Killing Civilians in Civil War: The Rationale of Indiscriminate Violence 9781626379060

Citation preview

Killing Civilians in Civil War

Published in association with the Handa Centre for the Study of Terrorism and Political Violence, University of St Andrews

Killing Civilians in Civil War The Rationale of Indiscriminate Violence Jürgen Brandsch

FIRSTFORUM PRESS A D I V I S I O N O F L Y N N E R I E N N E R P U B L I S H E R S , I N C. • B O U L D E R & L O N D O N

Published in the United States of America in 2020 by FirstForumPress A division of Lynne Rienner Publishers, Inc. 1800 30th Street, Boulder, Colorado 80301 www.firstforumpress.com

and in the United Kingdom by FirstForumPress A division of Lynne Rienner Publishers, Inc. Gray’s Inn House, 127 Clerkenwell Road, London EC1 5DB

© 2020 by Lynne Rienner Publishers, Inc. All rights reserved

Library of Congress Cataloging-in-Publication Data A Cataloging-in-Publication record for this book is available from the Library of Congress.

ISBN 978-1-62637-863-6 (hc)

British Cataloguing in Publication Data A Cataloguing in Publication record for this book is available from the British Library.

Printed and bound in the United States of America

The paper used in this publication meets the requirements of the American National Standard for Permanence of Paper for Printed Library Materials Z39.48-1992.

5  4  3  2  1

Contents

List of Tables and Figures Acknowledgments

1 Targeting Civilians in Civil Wars

Part 1 What Do We Know About Indiscriminate Violence?

2 State Use of Indiscriminate Violence

vii ix

1

21

3 Nonstate Use of Indiscriminate Violence

45

5 A Theory of Group-Selective Violence

83

4 Conceptual Building Blocks

Part 2 How Does Indiscriminate Violence Work in Practice?

63

6 Assessing the Effects of Indiscriminate Violence

125

8 Violence Against Civilians in Ethnic Wars

173

7 Group-Selective Violence Across History

v

149

vi

Contents

Part 3 Conclusion

9 Does Group-Selective Violence Work?

Appendixes Glossary Bibliography Index About the Book

199 209 219 221 237 239

Tables and Figures

Tables 4.1 5.1 5.2 8.1 8.2

Criteria for the Effects of Violence Against Civilians for the Perpetrator The Effects of Selective and Arbitrary Violence for the Perpetrator The Effects of Both Types of Group-Selective Violence Against Civilians for the Perpetrator Logit Regression Results for the Determinants of Violence Logit Regression Results for the Consequences of Violence

79 90

101 186 192

Figures 1.1 4.1 4.2 4.3 5.1 5.2 5.3 5.4

Three Audiences of Indiscriminate Violence Against Civilians Continua of Arbitrariness for Selective and Group-Selective Violence Coercion by Violence The Effects of Violence Against Civilians from the Perspective of the Perpetrator Types of Targeting The Coercive Effects of Specific Group-Selective Violence The Provocative Effects of Specific Group-Selective Violence The Net Effects of Specific Group-Selective Violence

vii

9

67 71

73 86 93 94 95

viii

Tables and Figures

5.5a

The Provocative and Coercive Effects of General Group-Selective Violence, Group 1 The Provocative and Coercive Effects of General Group-Selective Violence, Group 2 The Provocative and Coercive Effects of General Group-Selective Violence, Group 3 The Provocative and Coercive Effects of General Group-Selective Violence, Group 4 The Net Effect of General Group-Selective Violence The Effects of Specific and General Group-Selective Violence Ethnic Groups and the Usage of Violence Against Civilians Predicted Probabilities for the Usage of Violence Against Civilians Predicted Probabilities of an Additional Conflict Onset for an Average Ethnic Group

5.5b 5.5c

5.5d 5.6 5.7 8.1 8.2 8.3

98 99 99

100 101

120 185 187 194

Acknowledgments

Over the years, this book has benefited from the advice and support of numerous people. My first thanks have to go to Timothy K. Wilson. Tim supported this project enthusiastically from the outset, and I could not have asked for a better mentor than him. He has continuously encouraged me to move beyond established lines of thinking and given me the space to develop my own ideas. Without him, the project certainly would not have become what it is today. I also want to thank Javier Argomaniz for his support during various stages of the project. I also benefited greatly from the support and advice of my friends and colleagues throughout the years. Particularly, I would like to mention André Python, Adam Scharpf, Kristen A. Harkness, Marc R. DeVore, Sabine Carey, Anna-Lena Hönig, Bélén Gonzáles, Christian Gläßel, Mascha Rauschenbach, Katrin Paula, Nicholas Barker, Nino Abzianidze, Philipp Lichtenthaler, and Jonas Vestby. Aside from the academic community at St Andrews, this book was generously supported by the KonradAdenauer-Stiftung. Over the years the KAS has not only supported me financially, but also has become an intellectual home. I continue my affiliation with the Konrad-Adenauer-Stiftung, and the yearly meetings with the working group on terrorism and security are a constant source of friendship and inspiration. However, most important, I thank my family and close friends. I thank my parents for always supporting my decisions. Although these often may not have been the ones they would have taken, they never faltered in their support. For this I will always be grateful. Margot Wieland supported me during times when it looked like I would not be able to write the book. I am deeply thankful for her support throughout the years.

ix

x

Acknowledgments

Not least, I want to thank my wife, Christine, and our two sons, Elias and Theo. Since we met more than seventeen years ago, my wife has always supported and encouraged me. Without her help I could not have finished this project and I am deeply thankful for her being at my side. Elias and Theo are simply the best sons a father could wish for, and I am very happy to have them in my life. This book is dedicated to Christine, Elias, and Theo.

1 Targeting Civilians in Civil Wars

On November 9th in 1941, a German Wehrmacht company arrived in Baranivka, a Ukrainian village. The unit was investigating a partisan attack on a German patrol which had taken place a couple of days earlier in the same village. In this attack one German soldier and two officers had been killed, which made the German command of the rear area increasingly worried about a rising partisan threat. As a consequence, the command ordered an immediate investigation of the incident and the punishment of those who had collaborated with the enemy. Thus, when the company reached Baranivka, it had orders to retaliate in the harshest way possible. The soldiers began by encircling the village, and then the villagers were gathered near the house of the local physician where the German soldiers had been killed. The commanding officer interrogated the villagers, threatening to execute random people if they did not divulge who had been responsible for the attack. However, the villagers were seemingly unable to provide the demanded information. As a consequence, the officer chose ten male inhabitants randomly, and took them to the outskirts of the village where they were shot by a firing-squad. Afterwards most of the houses were burned to the ground and only a few buildings were left for the remaining villagers.1 This book is about incidents such as the one at Baranivka in 1941. It is about incidents of indiscriminate violence against civilians and the uncomfortable questions that they inevitably raise. What did the Wehrmacht forces at Baranivka actually assume would happen after they left? Did their commanding officer really think the area could be effectively pacified in this way? And, if so, did he have any good reasons for this belief? Even more disturbingly: did repression work? Against the backdrop of “total war” and

1

2

Killing Civilians in Civil War

the “war of annihilation” in Eastern Europe, this local incident might seem relatively trivial. And yet as we move our analytical focus out from this single atrocity, to consider wider perspectives throughout this book, it will become apparent just how important it is to understand such incidents more deeply. If we are ever to improve our general understanding of why civil wars so often take the deeply atrocious forms that they do and if we are ever to hope for finding remedies against this scourge for humankind, then we need to understand the apparently gratuitous slaughter of civilians better. When we begin to do so it soon becomes clear that this subject is potentially a huge one: because, sadly, there are Baranivkas (and similar such horrors) all over the globe. Groups of civilians have been and still are attacked in many conflicts. Often these are villages, but cities, ethnic groups, tribes or families also become the target of violence. To name just a few examples, Turkish security forces seem to have attacked and destroyed Kurdish villages after rebel attacks in the 1990s (Marcus, 2009, 221–222), Russian security forces attacked villages in Chechnya (Souleimanov & Siroky, 2016), and French forces initiated a system of collective responsibility in Algeria in the 1950s (Horne, 2006, 113–115). In all these instances the attacks were seemingly a deliberate part of the counterinsurgents’ strategy. However, this type of reprisal against the civilian population, which the literature often terms “indiscriminate violence,” poses a puzzle for researchers of armed conflicts: why do actors in civil war kill innocent civilians? After all, for all we know, these people had no inclination to directly support the enemy. Therefore, their death can only be either pointless or even detrimental to those responsible for killing them. In fact, the majority of practitioners and researchers are convinced that indiscriminate violence is either ineffective (Kalyvas, 2006; Pape, 1996) or counterproductive (Goodwin, 2001; Kalyvas & Kocher, 2007b; E. J. Wood, 2003). Their claim is supported by an abundance of case studies pointing out that indiscriminate violence can massively increase resistance or support for an opponent. Among many other examples researchers noted for the conflicts in El Salvador (Peceny & Stanley, 2010, 69), Northern Ireland (English, 2005, 151), and Vietnam (Race, 1972, 223–226) that indiscriminate violence leads to an upsurge in recruits for the opponent. Yet, this skeptical assessment of indiscriminate violence obviously has its limits: indiscriminate violence is used in so many conflicts that it is simply implausible to assume it is employed only due to an error of judgment. Consequentially, several authors have recently begun to investigate whether indiscriminate violence might, after all, be able to help a government in its counterinsurgency efforts (e.g. Johnston, 2009; Schubiger, 2014; Zhukov, 2015). These authors argue that indiscriminate violence may be used to coerce civilians into compliance or separate rebels and civilians from each other. But confusion still persists. No theoretical approach has emerged so

Targeting Civilians in Civil Wars

3

far that can explain how indiscriminate violence produces coercive effects that outweigh the potential counterproductive effects. The picture is complicated by a multiplicity of comparative and quantitative empirical findings both supporting and contradicting the notion that indiscriminate violence against civilians can have positive effects for the perpetrator. Many studies find indiscriminate violence to be detrimental to the perpetrator (ArreguinToft, 2003; Kocher, Pepinsky, & Kalyvas, 2011; Schubiger, 2014; Schutte, 2017b). This line of research has produced strong evidence of the increase in resistance following indiscriminate violence. And yet, a smaller number of studies provides evidence for positive effects; here, indiscriminate violence is found to repress rebels or increase the probability of winning for the perpetrator (e.g. Arreguin-Toft, 2005; Lyall, 2009). Therefore, unfortunately, the current state of the literature leaves us with many questions. In particular, it remains unexplained how the killing of innocent civilians can help the perpetrator. This is a deeply theoretical question as we do not know what benefits exactly a perpetrator can derive from indiscriminate violence. Going back to the example at the beginning, the central question is: how did the German army profit from killing ten random villagers, given that there is considerable evidence that such reprisals increase further resistance? I seek to close this knowledge gap by providing two simple propositions. First, I will start by acknowledging that indiscriminate violence can produce both types of effects, counterproductive and coercive, at the same time. But their respective strength depends on the circumstances. The theory I put forth will be able to integrate both effects into one framework in order to assess whether the perpetrator is able to achieve a positive overall effect. Second, and perhaps counterintuitively, I will point out that the purpose of violence may not be coercing the group targeted with violence. Currently, much of the literature on indiscriminate violence assumes that violence against one group seeks to coerce that same group. However, perpetrators may be interested in communicating to a wider audience that goes well beyond the targeted group. As I will show, widening our view of potential targets of coercion helps us to understand the gains that can be derived from violence. In short: we need to stand right back to see both a wider and a deeper landscape of possibility. With regard to the example of the village of Baranivka this means that the Wehrmacht may not have been interested in the coercive or counterproductive effects of the attack on Baranivka itself. Perhaps their horizons were wider, in fact, it may have willingly embraced the counterproductive effects within that village in order to demonstrate something to other villages. If the attack on Baranivka were able to coerce many villages in the vicinity, potential counterproductive effects within Baranivka might still be outweighed for the Wehrmacht.

4

Killing Civilians in Civil War

Deriving from these two propositions I hope to provide a theoretical framework that can be the nucleus for a new way of thinking about indiscriminate violence and its effects. The Focus of This Study and Definitions of Terms As I will outline in the next chapters, one particular problem within the literature on indiscriminate violence is an ambiguous usage of terms and concepts. Here, in particular, we look through a glass darkly. Although I will devote some space to the discussion of these problems later, it is necessary to clarify upfront the terms and concepts used in this book. The aim of this book put in one sentence is to study the effects of strategic indiscriminate violence against civilians in civil wars. Therefore, it is important to start with a brief clarification of the terms violence, strategic violence, civilians and indiscriminate violence. In general, I will follow existing definitions as much as possible in order to make my study compatible with previous research. Definitions of violence are notoriously problematic and the study of violence has been described as a “conceptual minefield” (Kalyvas, 2006, 19). In everyday language many actions and even speech can be termed violence (Nimni, 2017). However, this conceptual stretching entails the danger of misunderstandings. In order to reduce the risk of confusion, I will use a conservative concept of violence. I will follow Kalyvas in defining violence as inflicting physical harm and I will concentrate exclusively on lethal violence (Kalyvas, 2006, 20). While nonlethal forms of violence are important in their own right, the sheer amount of people killed in civil wars makes it a foremost interest of researchers and policy-makers to understand the murder of people. Strategic violence is the deliberate usage of violence by an organization for the purpose of benefiting the organization. It is particularly important to distinguish strategic from opportunistic violence (see e.g. E. J. Wood, 2010, 299). While strategic violence is decided by the leadership of an organization, opportunistic violence is violence carried out by individuals for their own individual gains. As a consequence, opportunistic violence may even be detrimental to the goals of the organization. I define a civilian as a person who is not fully employed as a soldier in either a rebel organization or the armed forces of a particular country. Therefore, the civilian category also includes collaborators and supporters of military organizations who are devoting only part of their efforts to helping a group (cf. Kalyvas, 2006, 19). Indiscriminate violence is a problematic term, as competing and ambiguous definitions exist. Currently, the central text that most of the lit-

Targeting Civilians in Civil Wars

5

erature refers to is Stathis Kalyvas’ book The Logic of Violence, in which he distinguishes between two categories of violence: selective and indiscriminate. Kalyvas defines selective violence as violence that targets opponents on an individual level: “Violence is selective when there is an intention to ascertain individual guilt” (Kalyvas, 2006, 142). Thus, actors use selective violence when they seek to punish individuals such as an informant working for the opponent. In contrast, Kalyvas defines indiscriminate violence very clearly as violence targeting groups (Kalyvas, 2006, 142). Yet challenges remain. Unfortunately, as I point out in Chapter 2, the application of the term indiscriminate violence often remains imprecise. Most importantly, this term may confound strategic violence (violence that follows a strategy to benefit an organization) with nonstrategic violence (e.g. violence that results from individual motives such as greed or violence stemming from errors in targeting). Although this does not pose a problem for Kalyvas’ original analysis (Kalyvas, 2006), which concentrates on the causes of violence, I show in Chapter 2 that this categorization leads to difficulties when seeking to apply it to the effects of violence. Arguably, we need to disaggregate further. As a consequence, I have opted for a categorization that separates several ideal types of violence: • Following Kalyvas, violence targeting individuals based on their behavior, which I will call selective violence. • Violence targeting groups such as villages, tribes or ethnic groups, based on their group-behavior, which I will call group-selective violence. • Violence without a targeting rule or a perceivable targeting pattern, which I will call arbitrary violence.

My categorization thus builds on previous efforts by Kalyvas (2006), Steele (2009), and E. J. Wood (2010) and is described in greater detail in Chapter 4. In this way, I can distinguish the strategic targeting patterns on the individual level and the group-level from the nonstrategic violence in arbitrary violence. Therefore, I will circumvent the term indiscriminate violence whenever possible and use the hopefully more precise categorization of these three ideal types. However, when discussing other papers and books, it is often important to use the original terminology. In those cases, the term indiscriminate violence must often be understood as an umbrella term encompassing both group-selective and arbitrary violence. Finally, since my main focus is on the group-selective part of indiscriminate violence, it is important to define what I mean by groups as well. Since I intend to provide a theory that is as encompassing as possible, I will work with only a minimum definition. I define groups as two or more people who are perceived by a potential perpetrator of violence as a distinct

6

Killing Civilians in Civil War

entity. This way, my theory will be open to incorporate violence against different kinds of groups as tribes, ethnic groups, villages or families. In sum, these terms set the cornerstones of the book’s theoretical and empirical ambition. While particularly the type of violence (strategic and lethal) being investigated is quite narrow, the targets of violence (types of groups and civilians) are more broadly defined, and so the focus is laid on the broader effects of a specific type of violence on groups. This way I want to make plausible the claim that effects of lethal indiscriminate violence and particularly group-selective violence are generalizable across groups and situations. Government and Nongovernment Violence Against Civilians Currently, most of the literature on indiscriminate violence against civilians concentrates on government violence, yet there is a growing amount of studies focusing on nongovernment actors. Unfortunately, the academic literature on government and nongovernment violence against civilians exists in relative isolation.2 In particular, the literature on terrorism often remains separated from other approaches to violence against civilians.3 One reason for this may be of course that terrorism studies are not exclusively focused on the study of civil wars. Instead, terrorism can be used as a strategy inside and outside armed conflict. However, many authors define terrorism in very similar terms to those outlined in the previous section as “indiscriminate violence.” Often the main difference is the fact that terrorism is attributed to nonstate actors while the literature on indiscriminate violence focuses mostly on government actors. As a consequence, it seems reasonable to explore how far government and nongovernment violence against civilians are comparable and if studying them together can be beneficial. While there are certainly differences in the causes of group-selective violence for government and nonstate actors, in this book it is argued that the effects are similar and comparable. Therefore, I will try to build a bridge between the two literatures which focuses on the effects of group-selective violence against civilians irrespective of the actor. The phenomena explored in terrorism and in government violence studies are sometimes very similar. While not all researchers of terrorism concentrate upon violence against civilians, a large amount of work focuses on this issue specifically (e.g. Goodwin, 2006b; Kydd & Walter, 2006; Pape, 2003; Richardson, 2007; Stepanova, 2008; Wilkinson, 2006).4 Several of these authors agree that terrorism using indiscriminate violence against civilians is different from other types of terrorism. For instance, the definition of Fortna (2015) points in that direction:

Targeting Civilians in Civil Wars

7

I define terrorist rebel groups as those who employ a systematic campaign of indiscriminate violence against public civilian targets to influence a wider audience. [...] However, civilian targeting is ubiquitous; almost all rebel groups (and almost all governments involved in civil wars) target individuals as a form of “control” to force cooperation and deter civilians from providing aid to the opponent. [...] By narrowing the definition to deliberately indiscriminate violence, I exclude this more common form of violence and focus on that which makes terrorism so terrifying: its randomness; and so outrageous: the intentional targeting of innocent civilians (as opposed to collaborators). (Fortna, 2015, 522)

Here, Fortna outlines the same distinction as Kalyvas in differentiating indiscriminate violence from selective violence and the awareness of this differentiation can be found quite often in terrorism studies. For instance, Kydd and Walter (2006) cite among others the two logics of attrition and intimidation, which broadly reflect selective and indiscriminate violence. Similarly, Hutchinson (1978) distinguishes between “endorsement terrorism” and “compliance terrorism” and Goodwin (2006b) explicitly focuses on indiscriminate violence in his conceptual development of categorical terrorism. Thus, it is possible to state that at least a part of the terrorism literature is conceptually comparable to the literature on indiscriminate violence by governments and focuses on the same phenomena. By focusing on indiscriminate violence alone, many of the definitional problems of terrorism vanish and most arguments for separating government violence from nongovernment violence become irrelevant. A central argument for the separation of the two phenomena is outlined by SanchezCuenca and De la Calle (2009) who argue that: “[t]he technology of socalled state terrorism (massive arrests, internment, mass executions, etc.) is very different from that of nonstate terrorism (car bombs, hostage taking, assassinations, selective shootings). We do not see the point of merging such disparate forms of violence.” (Sanchez-Cuenca & De la Calle, 2009, 34). However, when focusing on just one “technology” of violence, i.e., group-selective violence, there is no reason for not merging government and nongovernment actors anymore. In this case governments and nongovernment actors are restricted to the same technology. Others point out that governments and terrorist groups differ in their capabilities and in their institutional constraints (e.g. Hoffman, 2006; Pape, 2003). While these arguments may be true for many cases, they only point to potential differences in the ability and the usefulness of indiscriminate violence for a specific actor. Yet, I argue that once an organization decides to use this type of violence the effects should be similar. Nevertheless, whenever violence is used as coercion, the capabilities of the perpetrator are of course important. The more capable a perpetrator is to use violence, the higher will be the threat he can create for the population. This seems to be

8

Killing Civilians in Civil War

independent of the legal status of the perpetrator. Rebel organizations or governments are in principle both able to reach a high relative strength visa-vis their opponent (see e.g. D. Cunningham, Gleditsch, & Salehyan, 2013). In sum, it seems possible to study the effects of government and nongovernment indiscriminate violence against civilians in one framework. For this it is not necessary to argue terrorism and government violence are exactly of the same kind or that governments and terrorist groups face similar constraints. However, I will proceed upon the assumption that when governments and nongovernment groups engage in the same type of indiscriminate violence this violence will produce similar effects. Conversely, this also means that for my purposes I will speak of terrorism only in the sense of indiscriminate violence against civilians. Therefore, in Chapter 3, I will only address the literature on terrorism that focuses on this type of violence and will seek to integrate government and nongovernment violence in Chapter 5. The Effects of Violence Another aspect that needs clarification concerns the effects of violence. Violence in general and group-selective violence in particular can have effects on several levels. For instance, violence could coerce the targeted population, it could lead to a moral backlash, or it could trigger an international intervention. These effects come about as violence signals or communicates implicit messages to different audiences. In order to clarify which effects I am concentrating on, I distinguish between three main audiences.5 I will present these audiences as ideal types here. But in reality, the concepts may be fluid. The first audience is comprised by those who are directly impacted or could potentially be impacted by violence. When violence targets people in order to coerce, it targets some and threatens others. Violence and coercion thus concentrate on a domain of people who are actual and potential targets. For instance, in the case of Baranivka this would concern, among others, most villages in Ukraine. While the attack on Baranivka constituted one single event, it is clear that other villages in Ukraine could potentially suffer the same fate. For the sake of clarity, I will call this audience the target universe. Second, the perpetrator of violence usually has a constituency that under no circumstances will be targeted with indiscriminate violence. These are people who do not fall under the same targeting rule as the target audience and potentially support the actor who uses violence against civilians. For instance, in the Baranivka example this would be the German population within the boundaries of the core German Reich. Even if the Wehrmacht targeted more villages in Ukraine with group-selective violence, a German village would not.

Targeting Civilians in Civil Wars

9

However, violence can have other intentional and unintentional effects on this population as well. Since the constituency is closely connected to the perpetrator, the population can feel morally appalled that an actor that represents or claims to represent them is using violence against civilians. I will call this second audience the constituency. Third, beyond the constituency and the target universe other actors exist. These include among others all international actors and the sum of these actors can be seen as another audience. This audience is not included in the target universe and thus does not have to fear being targeted with violence. Neither is this audience directly connected to the perpetrator. For instance, the populations of other countries or international organizations which are not part of the conflict are often able to observe group-selective or indiscriminate violence to which they can then react. I will call this audience the nonparticipating observers. For an illustration see Figure 1.1. In addition, all these audiences consist of A) a population as well as B) organizations (military or political) that represent or claim to represent them. For instance, in the case of the Baranivka massacre the target universe consists of the Ukrainian population. However, the military organizations in the background are various partisan groups that claim to fight for the Ukrainian people. In addition, the constituency is the German population but the military organization representing them is the Wehrmacht.

Figure 1.1 Three Audiences of Indiscriminate Violence Against Civilians

10

Killing Civilians in Civil War

Such distinctions are important as the effects of indiscriminate violence on a population and military/political organizations may differ. It is also important since some of the arguments that I will review later make reference to the population while others refer to the organizations. However, as the exposure to violence and the preferences diverge between the population and the political representation, potential changes in behavior will be different, too. For instance, the partisans who hid in the wood near Baranivka may or may not have been concerned about the safety of civilians in that village. This, presumably, largely depended on their connection to the villagers and their own strategic considerations. However, the civilian population who were the target of the attack could not escape being concerned with attacks of this sort. As a future attack might target them, they will have been much more likely to adapt their behavior to avoid violence. A good general example of the dissociation between the preferences of an organization and the population is the widely used strategy of terrorists to provoke violence against their own constituency. In this case the population would prefer to not be attacked, but terrorist organizations benefit from these attacks. In fact, this strategy can be interpreted as a way to align the differing preferences of the population with those of the armed actor (see e.g. Lake, 2002). In this book I will concentrate on the intentional and unintentional effects of violence on the population of the target universe and I have several good reasons for doing so. For purely practical reasons, it is not possible to concentrate on all effects of indiscriminate violence in one book. The literature on most effects has increased considerably and it is necessary to set clear boundaries for the analysis. It is also generally accepted within the study of government and nongovernment violence that the main purpose of indiscriminate violence is to communicate with the target universe (e.g. Goodwin, 2006b, 2038, Kalyvas, 2006). The effects on other audiences are often merely unintentional side effects. In addition, the current controversy about the effectiveness or ineffectiveness of indiscriminate violence is mainly focused on the target universe. Scholars within the government and the nongovernment violence literature are at loggerheads when it comes to the effects on the target audience, but mostly do not disagree on the potential effects on the other audiences. In addition, I focus on the population instead of the military or political organizations. The reason for this is twofold. On the one hand, any type of effect of indiscriminate violence against civilians is more likely to be found among the civilian population itself. Since violence targets civilians and seeks to coerce them, the immediacy of violence can lead to direct behavioral changes while effects on military or political organizations need to be transmitted through the population. On the other hand, the current discus-

Targeting Civilians in Civil Wars

11

sion around the effectiveness of indiscriminate violence concerns the ability of this type of violence to have any positive effect for the perpetrator at all. As a consequence, in order to demonstrate that indiscriminate violence has an effect, it makes sense to concentrate on the population first. Later, additional research can then investigate how a behavioral change in the civilian population translates into behavioral change among organizations. However, it has to be borne in mind that indiscriminate violence can have effects on both other audiences as well. With regard to the constituency various authors have pointed out potential effects. For government violence, it is often argued that employing indiscriminate violence against the opponent’s population in a war might lead to increased resistance on the home front and ultimately to political defeat (e.g. Merom, 2003). Embarrassment can be a potent force. In fact, many case studies point to such a relationship. For instance, Townshend (2001) makes this connection for British violence during the Anglo-Irish War of 1919–1921 (Townshend, 2001, 351–355). In terrorism research, many authors make a similar observation. Some researchers argue that terrorism may alienate the constituency if violence is perceived as too brutal (e.g. Cronin, 2009, 348, Pape, 2003, 94), and De la Calle and Sanchez-Cuenca (2013) find empirical support for this connection for ETA (Euskadi Ta Askatasuna) attacks on the Basque community in Spain. However, others point out that the usage of terrorism can have a positive effect on the perpetrator’s constituency as well (e.g. increase support). Violence may even be used for this reason. For instance, Karber (1971) differentiates between the instrumental and the affective effect of terrorism: instrumental effects produce the immediate coercive effect on the target universe while affective effects produce a change in the long-term support for the terrorists among the constituency (Karber, 1971, 528). Similarly, in an early book on the Algerian War of Independence, Hutchinson (1978) coined the term “endorsement terrorism” which describes a strategy of increasing support within the FLN’s (Front de libération nationale) own constituency, the Algerian population in this case (Hutchinson, 1978, 145). Recently, this line of thought was taken up by Krause (2013), who argues that much of terrorist violence seeks to communicate to the constituency of the terrorists in order to increase support for the terrorist organization (Krause, 2013, 273). With regard to the nonparticipating observers—those watching from afar—it is also reasonable to assume there may be both positive and negative effects. In studying government violence, multiple authors have pointed to the potential negative effects of indiscriminate violence for the perpetrator. As violence is banned under international laws, using indiscriminate violence can lead to increased international attention with the potential consequence of international isolation, sanctions, reduction of

12

Killing Civilians in Civil War

financial aid or even outright military intervention (e.g. Arreguin-Toft, 2003; DeMeritt, 2012; Krain, 2012; Melander, 2009; Ron, 2003; Valentino, Huth, & Croco, 2006). However, particularly concerning nongovernment violence, many authors have pointed out that violence can generate the type of attention that is needed to produce international support. In fact, it seems to be one of the most highlighted facts in the terrorism literature that indiscriminate violence is able to increase the attention of the international community with potentially positive effects for the perpetrator (see e.g. Crenshaw, 1981, 386, Thornton, 2006, 50). To recapitulate, I will exclude the effects of indiscriminate violence on the constituency and nonparticipating observers from this study and concentrate only on the potential effects on the population of the target universe. In order to study these particular effects I will follow most of the literature and take the viewpoint of the perpetrator. From this perspective, violence can have positive or negative effects for the perpetrator on the civilian population. Conversely, I will assume that the civilian population in principle has three options of behavior towards the perpetrator: support, resistance, and neutrality. Support and resistance can come in many forms. For instance, civilians could join the perpetrators’ armed forces or the rebels’ armed forces, they could provide information to either side or engage in low-scale violence against either side. Neutrality would mean that civilians try to have as few contacts as possible with both sides of the conflict. In addition to these three forms of behavior, civilians can have the option of fleeing the theater of war. Some of the approaches discussed later refer to this particular option. However, in this book I will exclude this type of behavior and assume that people cannot flee. This enables me to focus on behavioral changes that seem particularly paradoxical, such as previously resistant or neutral civilians shifting their behavior to support the perpetrator of indiscriminate violence. Building on the previous literature, I identify three types of effects (which will be outlined in detail in Chapter 4) that potentially change the behavior of the civilian population: elimination, provocation and coercion.

1. Elimination is the most basic aspect of lethal violence. Lethal violence kills by definition. However, different types of targeting strategies kill different types of people. As demonstrated in the Baranivka example, group-selective violence seems to target the innocent. It therefore has a different elimination effect than say the assassination of a known collaborator which would kill an opponent and which would constitute selective violence. 2. The provocation effect makes civilians turn against the perpetrator and increases resistance among the civilian population. The resulting civilian behavior can take many forms, such as joining rebel forces,

Targeting Civilians in Civil Wars

13

using violence against the perpetrator, etc. Here, I identify three potential mechanisms that link indiscriminate violence with increased resistance which I will call: legitimacy, revenge and security. As I will discuss in further detail, revenge concerns emotional reactions to personal experiences of loss. In this case indiscriminate violence leads to increased resistance since people seek revenge for their loss. The legitimacy mechanism highlights the possibility that the usage of indiscriminate violence is morally so appalling that people will resist out of moral outrage. In addition, the security mechanism concentrates on the possibility that indiscriminate violence increases the perceived cost of staying neutral for the population. If indiscriminate violence is seen as random, people may perceive joining the rebellion as safer than staying out of it. 3. The coercion effect refers to the potential of indiscriminate violence to change the behavior of the population of the target universe. By threatening to punish a certain behavior, a perpetrator may be able to force the civilians into supporting him. A good amount of my discussion will concern this effect since currently only a few authors have devoted attention to the mechanisms that link indiscriminate violence with civilian support from the target universe. Structure of the Book From a methodological perspective this book is intended to construct and test a theory of the effects of group-selective violence against civilians. In this dual process, theory building seems particularly relevant. In 1995, Fearon wrote about the study of war: “The main theoretical task facing students of war is not to add to the already long list of arguments and conjectures but instead to take apart and reassemble these diverse arguments into a coherent theory fit for guiding empirical research” (Fearon, 1995, 382). I think the same is true for the study of indiscriminate violence against civilians. While empirical research in indiscriminate violence has increased steadily over the last years, the theoretical arguments have not been given a form that allows for an understanding of the countervailing positive and negative effects of this type of violence. Previous studies provide many important insights which I will highlight over the course of the book (e.g. Kalyvas, 2006; Schubiger, 2014; Douglass, 2012). However, I will reassemble and amend these arguments in order to construct a theory which can be tested. I start by critically reviewing existing theories on government indiscriminate violence in Chapter 2 and on nongovernmental indiscriminate violence in Chapter 3. The separation of these two chapters is not entirely stringent, as some arguments are not restricted to just one side. However,

14

Killing Civilians in Civil War

the separation is a major fault-line within the literature and, therefore, it makes sense to discuss both sides individually. These chapters are not designed as a mere narration of existing approaches. Instead, I want to engage in a process of consolidation and creative reconstruction of the literature. By pointing to some of the apparent weaknesses and potential strengths of existing approaches, the chapters will highlight the ways forward that are worth pursuing. Going systematically through the theoretical arguments for the positive view (i.e., violence can be beneficial for the perpetrator) and the negative view (i.e., violence is detrimental) on indiscriminate violence, I distill several key insights. First, the negative view on indiscriminate violence can be split up into arguments on the ineffectiveness of violence (violence cannot reach its intended goal) and the counterproductiveness of violence (violence produces unintentional side-effects). On the one side, I will argue that the arguments on the ineffectiveness of indiscriminate violence are only valid for violence that is perceived as random. Here the often wrongly made equation between violence against groups and violence that is perceived as random is a particular problem. On the other side, arguments on counterproductive effects mostly seem to be highly plausible. I will identify especially three relevant mechanisms within the literature that connect indiscriminate violence with counterproductive effects: legitimacy, revenge and security. Secondly, the positive view has recently gained strength in the literature on governmental violence. Some authors argue that indiscriminate violence can have a coercive effect. However, these authors do not explain how violence can lead to coercion or how the negative effects and the positive effects interact. As it is plausible that indiscriminate violence produces counterproductive effects, it remains unclear under what circumstances the coercive effect might outweigh the negative effects. This leaves us with three major obstacles for progress in understanding the effects of violence: 1) the conflation between different types of violence necessitates an improved categorization of types of violence that focuses on the effects and the perception of violence; 2) the way indiscriminate violence is able to coerce needs additional theorization; and 3) the countervailing effects of violence need to be brought into a single framework in order to study an overall effect. In Chapter 4, I address the three obstacles identified in Chapters 2 and 3, as overcoming them is the necessary precondition for building a theory on group-selective violence. I introduce a new categorization of violence. In particular, I distinguish between two targeting strategies, selective and groupselective violence, and violence that has no targeting pattern, arbitrary violence. I also review the concept of coercion and outline how violence against civilians can be coercive. I distinguish between two types of coercive strate-

Targeting Civilians in Civil Wars

15

gies: in the first strategy an actor seeks to coerce one target with violence (specific) and in the second strategy an actor seeks to coerce an audience by using exemplary violence against a target (general). Finally, I introduce a conceptualization of the effects of violence, which highlights that violence has simultaneously counterproductive (provocative) as well as positive effects and points out criteria under which the respective effect should be strong. In Chapter 5, I introduce my theoretical argument on the effects of group-selective violence. The theory is founded on rational choice assumptions and, therefore, explicitly outlines the preferences and incentives for the perpetrators of violence and the individuals within the target universe. This allows me to show that group-selective violence has a distinct set of effects on the groups and individuals within the target universe. Importantly, building on the insights of the previous chapter, I distinguish between two types of group-selective targeting strategies. On the one hand, an actor may focus narrowly upon the group itself. Simply, it targets a specific group in order to coerce it. I call this type of targeting specific group-selective violence. On the other hand, an actor might take a wider field of vision. It can target one group in order to coerce other groups more generally. I term this type of targeting general group-selective violence. While the first type of strategy is the model that most researchers seem implicitly to have in mind, I show that it is the second strategy that has the most potential to produce a positive net-effect for the perpetrator. In addition, I discuss the effects of specific group-selective violence and general group-selective violence and demonstrate that the potential positive effect of the former is much lower than the latter. Finally, although violence is targeted against groups I do not remain at the group-level for my explanations. The standard view among rational choice theorists is that individual rationality often does not translate into group behavior (Geddes, 2003, 194–198). As a consequence, it is important to base the theory in methodological individualism and show that individual rationality can translate into the expected group-behavior. In sum, the theory reconciles the previously theorized positive and negative effect of indiscriminate violence. By introducing two different types of strategies, specific and general group-selective violence, the theory is also able to point out why researchers may observe both unsuccessful and successful group-selective violence. The second part of the book is devoted to the empirical investigation of general group-selective violence (Chapters 6–8). In these chapters I concentrate on general group-selective violence as it is the strategy that most likely produces recognizable coercive effects. Since the current debate about indiscriminate violence mainly focuses on whether indiscriminate violence can be effective in coercing at all, it makes sense to focus on likely cases, to demonstrate the potential effectiveness of group-selective violence.

16

Killing Civilians in Civil War

In Chapter 6, I review the empirical findings on indiscriminate violence. This demonstrates that the problems identified in the theoretical literature have translated into problems with empirical strategies. However, it also shows that some authors have investigated group-selective violence, more or less without realizing that the phenomenon is different from selective violence or arbitrary violence. Interestingly, some of these findings already support the theory of group-selective violence. In Chapter 7, I discuss several historical cases of general group-selective violence. The purpose of this chapter is twofold. First, it is intended as a plausibility probe of the theory of general group-selective violence (Eckstein, 2000). Since the coercive effect of general group-selective violence has not been described as a generalizable effect, it makes sense to demonstrate how violence is used and what effects it seems to have in a set of diverse cases. Second, the chapter is intended to demonstrate the broad range of applications of general group-selective violence. Here it is particularly helpful to use historical cases as perpetrators tended to be much more outspoken about their strategies in earlier time periods. With the advent of international law, this willingness seems more and more reduced although the usage of group-selective violence appears to continue. I outline that cities, villages, tribes, and families have all been the target of general group-selective violence. In all contexts, perpetrators expect the coercive effects of violence will outweigh the negative effects, they seek to communicate their strategy to the target universe and they are willing to accept some negative consequences. In addition, the review of these cases provides some limited, anecdotal evidence of realized coercive effects. In most cases some indication can be found that violence increases support for the perpetrator. As a consequence, the cases can give additional plausibility to the possibility that general group-selective violence is used regularly for its coercive effects. In Chapter 8, the theory of general group-selective violence is tested with macro-data on recent ethnic conflicts and ethnic groups. Transferring the theory of general group-selective violence to multi-ethnic countries and building on the work of Toft (2003) and Walter (2009), I argue that violence is used in civil wars to deter other ethnic groups from rebelling. I use data on civil wars from the Uppsala conflict data program (UCDP) and the Ethnic Power Relations dataset (EPR) in a statistical analysis. I first test whether violence against civilians is used more often in cases where the incentives to do so are higher. Given the discussion of the theory, it should be expected that group-selective violence increases when more groups are present that could be coerced. I also directly investigate whether violence against civilians leads to fewer rebellions. The findings further support the idea that group-selective violence is used in many circumstances and a part of many recent conflicts.

Targeting Civilians in Civil Wars

17

In Chapter 9, I summarize my findings, discuss some limitations and draw several conclusions. All evidence taken together, I argue, this book provides support for the notion that group-selective violence can be effective in coercing. The findings of the book also pose an apparent moral dilemma. By finding that violence can be beneficial for a perpetrator it may seem that the book could—in the wrong hands—be used to improve repression by the ruthless. While this may be a natural first impression, I show over the course of the book that it is not the case. After all, perpetrators of groupselective violence exist already in abundance and they use violence against civilians in large amounts. As this book demonstrates, these perpetrators know about the potential positive effects of violence for them and pursue their ruthless strategies precisely because they work. However, this knowledge is not present among many of the researchers and practitioners who seek to counter violence and improve the lives of civilians in civil wars. Therefore, in a way this book is an effort to level the playing field. It is meant to produce equality of knowledge between perpetrators who use violence for their benefit and those who seek to counter violence. For the latter it is of the utmost importance to understand violence in order to fight it. Notes 1. For a detailed description of this incident see T. Anderson (1999). I will come back to the Baranivka example in Chapter 7. 2. Separating the literature on violence against civilians in these two categories (government and nongovernment violence) of course cannot do justice to all approaches. However, it can be seen as a pervasive trend that violence is studied separately. The reason for this seems to be that most authors focus on the causes of violence and less on the effects of violence. As a consequence, they emphasize the organizational differences between rebels and governments instead of looking for similar effects. 3. However, there are some exceptions to this general trend as, for instance, the literature on state terrorism, e.g. Stohl (1984), Stohl & Lopez (1986), Gurr (1986), Blakeley (2009), and Jackson, Murphy & Poynting (2009) and part of the literature on rebel group violence, e.g. Hultman (2007). 4. Other definitions that do not concentrate on civilians can be found for instance in English (2016), Neumann and Smith (2005), and Sandler (2003). A good overview of different kinds of definitions is provided by Schmid (2011). Schmid points out in his review of definitions of terrorism that slightly more than one third of academics agree that violence against civilians is a defining criterion for terrorism (Schmid, 2011, 76). 5. Others differentiate audiences even further. For instance, Schmid distinguishes ten target audiences (Schmid, 2005, 135). However, I would argue that the three audiences I am covering here are the most important ones.

2 State Use of Indiscriminate Violence

The study of violence against civilians is evolving fast. Since the end of the 1990s numerous articles and books have been published that explain why political and military actors would kill unarmed people. Several recent literature reviews attest to this development (Anderton, 2014; Hultman, 2014; Kalyvas, 2012; Valentino, 2014). In particular, the work of Kalyvas on selective violence and his so-called control-collaboration model has greatly advanced our understanding of violence against civilians (Kalyvas, 2006, 2012). However, the study of indiscriminate violence against civilians in particular remains obscure. Recent years have seen the publication of contradictory theories and empirical results (Arreguin-Toft, 2003, 2005; Johnston, 2009; Kalyvas, 2006; Kalyvas & Kocher, 2007b; Kocher et al., 2011; Lyall, 2009; Schubiger, 2014; Schutte, 2017b; Steele, 2009). The proposed arguments and findings seem to support both the possibility that indiscriminate violence works as a strategy and that it does not work. For the sake of convenience, I will call these two lines of thought the positive (indiscriminate violence is effective in achieving an intended goal) and the negative view (indiscriminate violence is not effective or is counterproductive). The reason for the divergent predictions of theories seems to be that many authors use concepts ambiguously, make implicit assumptions and do not delineate the domain or clarify the conditions under which their theoretical argument holds. Unfortunately, recent developments in the literature do not completely address this problem and it seems that some of the current empirical studies suffer from the unclear theoretical and conceptual work that characterizes the field at present (see Chapter 6). Therefore, it makes sense to provide a detailed criticism of the current theoretical literature on

21

22

What Do We Know About Indiscriminate Violence?

the effectiveness of indiscriminate violence before developing a more coherent theory. The Negative View: Indiscriminate Violence Does Not Work For a long time, it has been argued that indiscriminate violence is too blunt a tool to be able to coerce an opponent into compliance (e.g., Galula, 1964; Leites & Wolf, 1970; Nagl, 2002). But we need to explore this anatomy of (supposed failure) more closely. Two separate arguments are being raised here and it is important to distinguish the two. On the one hand, some authors maintain that indiscriminate violence has counterproductive effects. In other words, it exacerbates the problem it is meant to solve and there are unintended byproducts of violence as, for instance, an increase in the number of recruits for the opponent (e.g., Goodwin, 2001; Kalyvas & Kocher, 2007b; Souleimanov & Aliyev, 2015; E. J. Wood, 2003). On the other hand, others argue that indiscriminate violence is ineffective. It just does not do what it is supposed to do. Violence does not achieve its intended goal, as it is, for instance, unable to shape the behavior of its target (e.g., Kalyvas, 2006; Pape, 1996). Unfortunately, this distinction has not been made before which has led to some problematic conclusions. Counterproductiveness: Unintended Byproducts of Violence

Particularly with regard to guerrilla warfare it is generally argued that indiscriminate violence increases support for the enemy (e.g., Goodwin, 2001; Kalyvas & Kocher, 2007b; Souleimanov & Aliyev, 2015; E. J. Wood, 2003). Researchers often agree on the general direction of the effect. But they differ in the causal mechanisms that link indiscriminate violence and support. I was able to identify four theoretical linkages between indiscriminate violence and support in rebellion: opportunity costs, legitimacy, revenge, and insecurity. While I will discuss different mechanisms here, it is not always possible to pin down authors to just one of them. For instance, Goodwin (2001) seems to make an argument that fuses the legitimacy mechanism with the security mechanism. Therefore, the following is less an attempt to do complete justice to each author than a way to bring structure to the debate. Opportunity costs. The usage of indiscriminate violence against civilians

can decrease the opportunity costs for joining the rebellion (Azam, 2002, 2006; Azam & Hoeffler, 2002; Bueno de Mesquita & Dickson, 2007). Opportunity costs are the hypothetical costs that accrue if a person chooses

State Use of Indiscriminate Violence

23

the less rewarding of two options. Therefore, if not joining the rebellion becomes less economically attractive (for instance, if violence decreases the amount of jobs) it may increase mobilization for the rebels as people choose the now more rewarding or less costly option. Bueno de Mesquita and Dickson (2007) write: “When counterterror imposes hardship on an aggrieved population, support for continued violence is likely to increase both because the opportunity costs of violence decrease and because the people conclude that the government is not concerned with their welfare.” Azam extends this point and argues that violence against civilians and the decrease in economic opportunities stem from a common cause: looting. He points out that warlords have an incentive to loot either their own or their opponents’ constituency and he assumes that violence against civilians is a side effect of looting (Azam, 2002, 2006). Consequently, killing civilians can be connected to a reduction of economic opportunities which increases mobilization into rebel forces.

Legitimacy of the perpetrator. Indiscriminate violence can lead to a fall in legitimacy of the perpetrator. Massacres stain reputations perhaps indelibly. This, in turn, might lead to decreased support for the actor who uses indiscriminate violence and increased support for the opponent. For instance, Goodwin (2001) writes: Indiscriminate state violence against mobilized groups and oppositional figures is likely to reinforce the plausibility, justifiability, and (hence) diffusion of the idea that the state needs to be violently “smashed” and radically reorganized. [...] Unless state violence is simply overwhelming, then, indiscriminate coercion tends to backfire, producing an ever-growing popular mobilization by armed movements and an even larger body of sympathizers (Goodwin, 2001, 48).

In this view, indiscriminate violence changes the attitudes of the population towards the government which directly translates into support for the opposition. In a similar vein, E. J. Wood (2003) emphasizes the emotional benefits of participation. In her study on the civil war in El Salvador she observed that those involved in supporting the rebellion were doing so because of intrinsic benefits of participation. Since the government was generally despised as a result of its violence against the population, revolting against it provided emotional benefits, as, for instance, a perceived increase in dignity by the participants, which Wood calls “the pleasure of agency” (E. J. Wood, 2003, 17). Wood thus links the antipathy towards a government to processual selective benefits from participation; people can benefit individually from participating. This explains for Wood how the rebels were able to convince people to support them without receiving direct material benefits.

24

What Do We Know About Indiscriminate Violence?

In such circumstances, the argument goes, it might indeed be more appealing to die on one’s feet than live on one’s knees. Revenge. Indiscriminate violence can lead to a desire for revenge. For

instance, Kalyvas points out that in numerous case studies of guerrilla warfare rebels cite the desire for revenge as a core motivation for taking up arms (Kalyvas, 2006, 153–154).1 Therefore, the direct experience of indiscriminate violence can be the source of resistance. People who saw their friends and families being killed may react in an emotional way that supersedes a material cost-benefit calculation. With the feeling that they have nothing to lose they may disregard all security concerns and join the rebels to take revenge. In addition, it is possible that within cultures practicing blood revenge the revenge mechanism may be reinforced. Souleimanov and Aliyev (2015) argue that the custom of blood revenge seems to be a particularly powerful factor in mobilizing people, since the failure to act on the duty to exercise revenge leads to severe social sanctions: Our interviews with former Chechen insurgents confirmed this observation [that blood revenge is an important mobilizing factor], highlighting the need to restore their individual and clan honor through blood revenge as a key reason for their violent mobilization. Failure to do so would have been too heavy of a burden to bear. Indeed, many interviewees noted the social context of such failure and their fear of being considered cowardly and weak. As one veteran of the First Chechen War stated, “After what the [Russian] soldiers did to my household, it wasn’t possible for me to stay home and pretend nothing had happened. I couldn’t have looked people in the eye” (Souleimanov & Aliyev, 2015, 171).

However, while Souleimanov and Aliyev (2015) observe a strong link between blood revenge and mobilization they also note stark differences with regard to the type of crime being avenged and the region. For instance, cases of rape seem to mobilize avengers in large numbers. In addition, they find differences between cities and rural areas. While revenge was clan based in the mountainous south of Chechnya and thus mobilizing more people, in urban areas it was restricted to the core family (Souleimanov & Aliyev, 2015, 174). While blood revenge is certainly not a phenomenon that is common in all countries it is not unique to Chechnya, either. In fact, according to Souleimanov and Aliyev (2015) blood revenge is currently practiced in Afghanistan, Pakistan, Yemen, Somalia, Turkey, Albania, Kosovo, parts of North Africa, Chechnya and Ingushetia and may thus contribute to the mobilization effect of indiscriminate violence in many conflicts (Souleimanov & Aliyev, 2015, 162–163).

State Use of Indiscriminate Violence

25

Perception of insecurity. Indiscriminate violence can lead to a perception

of insecurity among civilians. This security-based view seems to receive wide support from the surveys of participants in rebellions and armed groups, where many respondents named improved security as a motivation for joining (Arjona & Kalyvas, 2012; Guichaoua, 2007; Humphreys & Weinstein, 2008). Although this argument is widespread it is perhaps most prominently developed by Kalyvas and Kocher (2007b) who link indiscriminate violence to the collective action problem faced by the organizers of a rebellion.2 Kalyvas and Kocher build their explanation around the idea that insurgents have to overcome a collective action problem in order to organize a rebellion. This idea was originally developed by Lichbach (1998). The collective action problem for rebels stems from the fact that participation in a rebellion is risky and costly, yet the benefits of the rebellion, e.g., a potential regime change, are nonexcludable. This means that even those who did not help in the rebellion will benefit from its effects, if it is successful (e.g., a more favorable regime). Consequentially, nonparticipation is the rational choice. A solution to this problem (alongside other measures) is that organizers of a rebellion can provide selective incentives (e.g., monetary benefits) to ensure participation (Lichbach, 1998). Building on this theoretical insight, Kalyvas and Kocher argue that indiscriminate violence may reverse the collective action problem for the rebels. They point out that when confronted with indiscriminate violence, nonparticipation might be either equally risky or even riskier for civilians than joining the rebellion. Therefore, the rebellion can provide security from indiscriminate violence as a selective incentive for joining and, thus, indiscriminate violence offers the rebels an additional way to solve the collective action problem; for civilians neutrality becomes unattractive and participation becomes the rational choice. On top of this, Kalyvas and Kocher point out that indiscriminate violence may have additional “perverse effects” which may make the reversal of the collective action problem even more pronounced. In fact, indiscriminate violence may disproportionately target neutral civilians or potential supporters of the perpetrator since they do not actively avoid being close to the perpetrator of violence. Assuming they will be safe, supporters of the incumbent may stay put but become the target of attacks anyway, while those collaborating with the rebels will leave and, in this way, avoid being harmed (Kalyvas & Kocher, 2007b, 191). However, the authors are very careful to point out that “indiscriminate state violence in and of itself does not automatically result in the growth of an insurgency: the rebels’ ability and willingness to capitalize on it are necessary factors intervening between indiscriminate violence and rebel recruitment” (Kalyvas & Kocher, 2007b, 190).

26

What Do We Know About Indiscriminate Violence?

Logic and Relevance of the Arguments

The above arguments all seek to explain that indiscriminate violence can have counterproductive effects. However, they are not all equally convincing when the definition of indiscriminate violence as lethal violence against groups (see Chapter 1) is applied. I will, therefore, briefly discuss all of them and point out the criteria under which the arguments should become relevant. The first mechanism concerns the opportunity costs of rebellion. Within this line of thought indiscriminate violence is supposed to lead to decreased economic opportunities which then increases the attractiveness of joining the rebellion. However, it seems as if the proposed mechanism assumes more than violence against civilians. None of the authors provides an argument of how lethal indiscriminate violence against civilians itself is related to the reduction of economic opportunities. Indeed, there seems to be no direct link between the two. The question to be answered would be: how does killing, say, five people in the street, reduce the economic opportunity costs of joining the rebels for other people? In principle, in an economic labor market model (that does not assume unlimited supply of workers) wages rise with a reduction of the workforce (e.g., Harris & Todaro, 1970). Interestingly, indiscriminate violence in itself and only from this economic point of view, could increase the opportunity costs of rebelling (i.e., make joining the rebellion comparatively costlier) since regular employment becomes more attractive with rising wages. This is obviously a gruesome way to look at things, but it is necessary in order to identify the channels through which indiscriminate violence has an independent influence. What seems to produce the logic that the aforementioned authors have in mind is not the fact that people get killed. Rather their arguments seem to implicitly assume that the killing of civilians is inherently connected to a destruction or looting of capital (e.g., appropriation of cattle, etc.) which makes working in the formal or informal economy less attractive. This connection is made explicitly in the work of Azam (2002, 2006). However, other authors have already pointed out that there is no necessary connection (MacGinty, 2004). Looting and economic destruction may indeed occur alongside indiscriminate violence. But indiscriminate violence itself does not necessarily reduce economic opportunities. A potential alternative to think about the opportunity cost argument would be to consider human or social capital instead of physical capital. In this case the killing of five people could lead to a reduction of capital and, as a consequence, to a reduction of jobs and wages. However, the important point here is to show that there is no clear-cut connection between violence against civilians and economic opportunities that works only through economic variables.

State Use of Indiscriminate Violence

27

The second mechanism concerns the legitimacy of the perpetrator. Within this line of thought it is argued that the killing of innocent civilians leads to a loss in legitimacy which then translates into support for the opponent. Therefore, it seems the argument makes a direct link between preferences and behavior. However, this link is far from straightforward. In fact arguments of this sort have frequently been criticized for not taking the potential costs of a change in behavior into account. As many examples show, civilians in civil wars will prefer short-term security over other goods in such a high risk environment. Leites and Wolf (1970) observe that “the efforts to limit damage may prevail over aspirations to better one’s condition or act according to one’s ideals” (Leites & Wolf, 1970, 126). Rather than fixate upon potential gains from resistance (e.g., a new political regime) they are mostly concerned with limiting the damage to what they possess. This view reflects long established findings from social psychology that attitudes and behavior are mostly unconnected (Wicker, 1969). In an early literature review Wicker points out that “[t]aken as a whole, these studies suggest that it is considerably more likely that attitudes will be unrelated or only slightly related to overt behaviors than that attitudes will be closely related to actions” (Wicker, 1969, 65). Put another way, the argument that attitudes lead to action rests on the assumption that there is no cost-benefit trade-off between action and nonaction. In addition, the existing studies in the civil war context seem either to focus on situations where participation in the rebellion is cost-free (or very low cost), as E. J. Wood (2003) seems to do3, or that civilians have not much to lose anyway. However, both assumptions are extreme as general conditions and will only hold in a few cases. This means that in all other cases, where civilians value their life and supporting the rebellion is risky, this explanation may be insufficient. In a way, the following two theoretical linkages between indiscriminate violence and rebel support (revenge and security) could be seen as reflecting specific cases under which either the value of life or the relative riskiness of rebellion is reduced. It seems then that this argument alone is not able to explain a shift in the behavior of people in general. Yet, it would be wrong to dismiss the argument entirely. Under certain circumstances, when supporting the rebellion is not costly or when the status quo is costly, a loss in legitimacy may lead to increased resistance. As the examples described by Wood suggest, providing food for rebels or maybe providing finance is not as dangerous as joining the rebels as a fighter. Therefore, demonstrating the illegitimacy of a government can lead to increased support from the population other than joining as a fighter, which is nevertheless very useful. The third mechanism concerns revenge. Individuals who are directly impacted by violence (for instance, those who have lost family members or friends) will experience feelings of revenge, which makes them more likely

28

What Do We Know About Indiscriminate Violence?

to join the opponent of the attacker. In a way Wood’s argument on the emotional benefits of resistance may also fit well in here. Evidence on a revenge mechanism is overwhelming (see e.g., Kalyvas, 2006; Lyall, 2009) and numerous case studies point to this mechanism, for instance, within the Mau Mau rebellion (D. Anderson, 2005, 46–47), the Kurdish rebellion in Turkey (Marcus, 2009, 133, Özeren, Sever, Yilmaz, & Sözer, 2014, 337), and the Algerian War of Independence (Horne, 2006, 118). In the case of the Vietnam War particularly rich anecdotal evidence seems to provide support for revenge motivating participation in rebellion. For instance, a study on the impact of bombing villages in Vietnam concludes that those who joined as a result of the US bombing campaign were some of the most motivated fighters amongst the Viet Cong (Donnell, 1967). Another study reports that revenge for dead relatives drove many of the interviewed prisoners to join the Viet Cong (Kellen, 1970, 52–53). In addition, some limited evidence from surveys indicates that revenge plays a role in the decision to participate in a rebel organization. For instance, Arjona and Kalyvas find that in Columbia about 13 percent of the respondents of a survey among ex-combatants claimed that revenge was the main driver in their decision to join an armed group (Arjona & Kalyvas, 2012, 158). It is therefore highly plausible that revenge plays a role in producing support for the opposition. Particularly in cases where blood revenge is a custom the emotional mechanism is combined with social sanctions that serve as selective incentives to join the rebellion (Souleimanov & Aliyev, 2015). However, studies on civil wars do not provide systematic evidence on the size of the effect and it seems that the mechanism is restricted in its reach. The findings of Arjona and Kalyvas (2012) suggest that at least in the Columbian case revenge did not motivate more than a minority of rebels. Yet, presumably, only those with close connections to the victims will experience a high desire for revenge. In the context of blood revenge this may extend to some people within a clan, but still the effect is confined to a narrow group of people for each individual act of government violence. Souleimanov and Aliyev (2015) point out that only few cases (and particularly cases of rape) led to higher amounts of mobilization (Souleimanov & Aliyev, 2015, 174). This means in turn that indiscriminate violence may lead only to a limited amount of resistance through the revenge mechanism. While each indiscriminate killing may increase recruits for the opponent, the increase is limited by the number of close relations of each person killed. The fourth mechanism concerns perceived security. This argument is at the same time the most compelling and the most problematic one. Kalyvas and Kocher argue that indiscriminate violence can increase the risk of staying neutral as violence also targets the innocent. This means that the rela-

State Use of Indiscriminate Violence

29

tive risk of joining the rebels may decrease to a level where joining is safer than not joining. However, Kalyvas and Kocher remain ambivalent about the patterns that indiscriminate violence follows. Although they acknowledge in line with Kalyvas (2006) that indiscriminate violence is defined as targeting on a group level (e.g., locality, ethnicity, sex, age, etc.), they do not elaborate on this targeting. Instead they treat indiscriminate violence and poorly targeted selective violence as roughly the same (Kalyvas & Kocher, 2007b, 187). In particular, Kalyvas and Kocher use an example for violence from the Phoenix Program. In this program the US sought to target those Vietnamese civilians collaborating with the Viet Cong during the Vietnam War. Within this program, they argue, poor intelligence led to the killing of a large number of innocent people which then led to a perception of insecurity. In fact, I agree that in such a case, civilians should not be able to perceive a pattern of violence and thus will perceive it as random. Since no one can predict the government’s failure to identify collaborators the risk of being killed should rise equally for everybody. However, this sort of violence is absolutely not equivalent to indiscriminate violence as defined by Kalyvas, which is the targeting of groups. When groups are intentionally targeted, perceptions of violence should be different, as civilians may be able to perceive this targeting pattern. If, e.g., only men are targeted by violence, for women the costs of staying neutral are unchanged. If violence targets one ethnic group, neutrality becomes costlier only for this one group. If violence targets only villages with known rebel presence, other villages can still feel safe. Therefore, indiscriminate violence is supposed only to increase the cost of neutrality for a person that falls under the targeting rule. Consequentially, this would lead to separate effects for those groups that are targeted and others that are not targeted. Within the targeted groups people will perceive violence as if it were arbitrary. Since they cannot do anything to avoid being targeted, violence cannot alter their behavior. To the contrary, outside the group that is targeted people may perceive violence as if it were selective. Since certain groups are punished for a specific reason people can in principle adjust their behavior. So, the counterproductive effect would only be felt within the targeted groups. The reach of the security mechanism is, therefore, limited. Violence only leads to unambiguously counterproductive effects through the security mechanism when no targeting pattern is perceivable. This means that Kalyvas and Kocher make a category mistake when it comes to the effects of violence. 4 By treating violence against groups as equivalent to poorly targeted selective violence, they conflate two types of violence with different effects. However, they are still right in their assessment that once the security mechanism is triggered, we can expect large amounts of

30

What Do We Know About Indiscriminate Violence?

mobilization. Once violence becomes so prevalent and unpredictable that people see neutrality as inferior to rebellion, we can expect that support for the opposition rises exponentially as more and more people will move from neutrality to opposition. This means that all four arguments on the counterproductiveness of indiscriminate violence need qualification. First, the opportunity cost argument seems problematic, at least when the focus is only on lethal indiscriminate violence. While indiscriminate violence may be accompanied by destruction and looting, which reduce economic opportunities, lethal indiscriminate violence itself does not reduce them, per se. Secondly, it seems implausible that the simple de-legitimization of the regime leads directly to overt action against the regime as long as taking action is risky and people have something to lose. Thirdly, emotional reactions and security concerns seem to have more explanatory power. However, these effects are confined in important ways. Revenge and blood revenge reactions should be triggered mostly by violence killing closer relations, which limits the reach of this effect. Here only a moderate increase of resistance can be expected. For the security considerations, it is important to conclude that the targeting patterns of indiscriminate violence matter. While violence that is perceived as arbitrary may generally increase insecurity, violence against groups should only increase insecurity for the civilians within those groups targeted. However, once the security mechanism is triggered support for the opposition will increase drastically. Ineffectiveness: Inability to Achieve the Intended Goal

So far the discussion has concerned potential unintended side effects of indiscriminate violence (counterproductive or provocative effects). However, in and of themselves, these effects do not preclude indiscriminate violence from being effective in achieving the intended goal at the same time. For instance, violence may produce feelings of revenge, but still be effective in coercing. Still, a couple of scholars assert that indiscriminate violence may also be ineffective in shaping the behavior of a target audience (e.g., Kalyvas, 2006; Koc-Menard, 2006). For the case of asymmetric warfare Kalyvas (2006) has summarized these arguments in his book The Logic of Violence which has become a major reference point in the discussion of indiscriminate violence. Here, Kalyvas identifies two mechanisms why indiscriminate violence can be ineffective (Kalyvas, 2006, 153–160).5 He recounts the argument that indiscriminate violence may not set incentives for collaboration (Kalyvas, 2006, 155). While selective violence (e.g., the killing of an enemy collaborator) signals to bystanders that unwanted behavior will be punished, this is not the case for indiscriminate

State Use of Indiscriminate Violence

31

violence (Kalyvas, 2004, 105–106, Kalyvas, 2006, 144–145). “Indiscriminate violence,” Kalyvas argues, “lacks almost every feature that is generally considered to be necessary for the effectiveness of sanctions: it is usually late, often arbitrary, inconsistent, erratic and totally disproportionate” (Kalyvas, 2006, 155). Since indiscriminate violence does not select on the individual level civilians can do little to avoid punishments. As a consequence, it cannot produce an incentive to collaborate with the perpetrator or stop collaboration with the enemy. Kalyvas also points out that indiscriminate violence may fail because of a missing link between civilians and rebels (Kalyvas, 2006, 151). This argument rests on the assumption that the perpetrator seeks to change the behavior of the rebels by attacking civilians. The underlying logic is that when civilians are attacked they will lobby the rebels to cease their activities. However, in order to do so, civilians need to have access to, and influence upon, the rebels’ decisionmaking. Should civilians lack this influence they may fail to change the rebels’ behavior and indiscriminate violence would, therefore, remain ineffective (Kalyvas, 2006; Lyall, 2009). Kalyvas carefully formulates these arguments as potential explanations for the ineffectiveness of indiscriminate violence and thus grants that they do not hold under all circumstances. However, it is nevertheless important to discuss the arguments thoroughly here as he and others draw from them the general conclusion that indiscriminate violence will mostly be ineffective. The argument on the lack of incentives for collaboration is not generally valid. As outlined above, Kalyvas seems to have an ambiguous concept of indiscriminate violence against civilians in mind when he discusses the counterproductiveness of violence. Unfortunately, this is also the case when he is considering its ineffectiveness. To begin with and as mentioned before, Kalyvas defines indiscriminate violence as violence that targets groups. However, when Kalyvas turns to explaining the ineffectiveness of indiscriminate violence, his account reads rather differently. Here, he even uses the terms random violence and indiscriminate violence interchangeably while he clearly focuses on the random aspects: Although random violence may work for a dictator [...], it is much less likely to achieve its aims in the midst of a civil war, where the presence of a rival makes defection possible. First, random violence defeats deterrence because it destroys the possibility of anticipation of a forthcoming evil and, hence, the ability to avoid it: it erases the relationship between crime and punishment, thus abolishing the concept of transgression. Its sheer unpredictability makes everyone fear lethal sanctions regardless of their behavior: innocence is irrelevant and compliance is utterly impossible. [...] Secondly, whereas compliance guarantees no security under conditions of indiscriminate violence, collaboration with the rival faction may both increase one’s chances of survival and allow for a sense of normative integrity. (Kalyvas, 2006, 143)

32

What Do We Know About Indiscriminate Violence?

As with the discussion of the counterproductive effects, Kalyvas’ argument seems to only speak to violence that is perceived as random (i.e., unpredictable violence). Only this type of violence is ineffective and again it does not necessarily apply to violence targeted at groups like families, villages or ethnic groups. For instance, violence against villages based on their support for the rebels is not completely unpredictable. To the contrary, measures of collective punishment in counterinsurgency operations are mostly applied for some sort of perceived or actual collaboration with the rebels which makes them sometimes very predictable. I will come back to this in Chapter 5 and explain exactly how violence against groups can shape incentives. In this regard, Souleimanov and Siroky (2016) provide an important distinction between retributive attacks (i.e., attacks based on previous behavior) and random attacks. Building on Kalyvas they argue that random violence is violence that targets large groups for no reason while retributive violence targets smaller groups (they restrict their analysis to villages) as a reaction to insurgent activities. They mainly argue that both types of violence have counterproductive effects, however, the effects differ in their location. While random violence leads to counterattacks in the same location, retributive attacks lead to insurgent violence in other locations, in order to avoid another retributive government attack. However, the important point here is that some forms of indiscriminate violence cannot be described as random or perceived as random. Violence that is linked to some behavior of certain groups can be perceived as a pattern.6 Kalyvas also argues that in cases where civilians have no influence on rebels, indiscriminate violence may be ineffective. However, this argument makes a tacit shift from civilian to rebel behavior as the target of coercion. The argument raised by Kalyvas does not dispute that violence has an impact on civilians, just that they may fail to influence the rebels. Yet, given the importance that most authors attribute to civilian support in irregular wars this cannot count as a failure of coercion. If civilians are willing to lobby for the government side, but remain unable to change the rebels’ behavior, this does not imply that the situation is the same as without violence. In fact, violence has already made civilians change their behavior (from fence sitting to opposition of rebel activity). Put differently, this is not an argument against a coercive impact of group-selective violence on the civilian population of the target audience. In fact, the argument assumes such a coercive effect on civilians. In sum, the arguments on the ineffectiveness seem to be unconvincing as general arguments. While random violence may be unable to coerce and violence may sometimes have limited impact on rebel behavior these are no arguments against a coercive effect of group-selective violence on the civilian population. It seems possible that violence can be used in forms that are

State Use of Indiscriminate Violence

33

not perceived as arbitrary and thus able to produce a coercive effect. Yet, this is not to deny that these arguments provide important points. It remains true that in cases where indiscriminate violence is perceived as random it is likely to remain ineffective. In addition, when civilians have no influence on rebels, indiscriminate violence might not change rebel behavior. It is important not to jettison these arguments wholesale, but to define more precisely the domain within which they remain valid. The Positive View: Indiscriminate Violence Works Despite a large majority of practitioners and researchers believing that indiscriminate violence is ineffective or even counterproductive, this sort of violence is frequently observed in civil wars. Kalyvas himself points out that it is unlikely that all violence that is perceived as being indiscriminate is simply selective violence that has been mislabeled (Kalyvas, 2006, 148). The resulting puzzle that indiscriminate violence is used even though it is commonly believed to be counterproductive has led to a recent rise in research on its potentially effective nature. Here, two approaches can be distinguished. First, some scholars argue that indiscriminate violence can be a tool to physically remove civilians from a certain area either by killing or forced displacement. This strategy is designed to deny rebels contact with their civilian support basis. Second, others maintain that indiscriminate violence can be used to coerce civilians into supporting the government. In this case civilians stay put, but are forced through indiscriminate violence to cooperate with the government against the rebellion.7 These approaches roughly resemble Schelling’s classic distinction between brute force and coercion, and in fact, authors frequently make reference to Schelling (1966).8 Physical Separation: Interdicting Rebel-Civilian Interaction

Arguments on the physical separation between rebels and civilians come in two forms. First, it is possible that actors seek to physically annihilate the supporting population with massive violence (Arreguin-Toft, 2001, 2005; Fjelde & Hultman, 2014; Hultman, 2009; Valentino, Huth, & Balch-Lindsay, 2004). Second, it is argued that indiscriminate violence can be used to displace civilians (Steele, 2009, 2011). Brute force. Most approaches emphasizing brute force concentrate on guer-

rilla warfare. The arguments rest on the assumption that civilian support is a sine qua non for military success in irregular warfare and authors consequentially focus on how indiscriminate violence can either deny this resource to the enemy or make the resource available to the perpetrator of violence.

34

What Do We Know About Indiscriminate Violence?

The annihilation argument seems to be the basis for several authors. Perhaps most prominently, Valentino et al. (2004) argue that indiscriminate violence against civilians is meant to kill the support basis of rebel forces. They point out that since civilian support is a necessity for rebels, governments can use mass-killings as a viable strategy to success. When governments are unable to target the rebels directly, killing civilians will cut rebels’ supply lines, which ultimately denies them the resources to fight. They further explain governments will use such drastic measures only if they are frustrated with the ineffectiveness of other counterinsurgency measures, confronted with a strong opponent or if the popular support for rebels is high (Valentino et al., 2004, 386–387). Arreguin-Toft makes a similar argument in his book on strategic interaction between strong and weak actors in war (Arreguin-Toft, 2001, 2005). He maintains that strong actors facing a guerrilla force will be more likely to lose if they counter this with direct military force (Arreguin-Toft, 2005, 38–39). However, attacking the civilian support basis of the rebels, a strategy he calls barbarism (ArreguinToft, 2005, 30–31), enables strong actors to level the playing field or even get back their numeric advantage. As Arreguin-Toft summarizes: “Barbarism works as a COIN [counterinsurgency] strategy because by attacking either or both of the essential elements of a GWS [guerrilla warfare strategy] sanctuary and social support it destroys an adversary’s capacity to fight” (Arreguin-Toft, 2005, 41). In addition, Hultman (2009) and Fjelde and Hultman (2014) apply this logic to ethnic wars. Fjelde and Hultman argue that in ethnic wars rebel organizations draw heavily on the support of their own group, since shared ethnicity facilitates cooperation between rebel groups and civilians. Therefore, they point out attacking the civilian population of the opponent’s ethnic group is a way of reducing the opponent’s resources and an indirect way of targeting his military strength. In a very similar approach Balcells (2011) points out that indirect violence (i.e., violence not perpetrated on a face-toface level, like bombings) is intended to weaken the enemy and therefore targeting areas where enemy support is highest.

Displacement. The displacement argument is most notably expressed by Steele (2009, 2011). Steele maintains that governments may use indiscriminate violence in order to strategically displace civilians who are perceived to be disloyal. In contrast to Kalyvas (2006), Steele assumes that at least in some cases civilian support may not be enforced through incentives as civilians have unchangeable loyalties for one actor. In these cases an actor cannot hope to coerce civilians with selective violence and indiscriminate violence is used against groups associated with the opponent (either ideological or ethnic) in order to displace these groups. Thus, while Kalyvas (2006) interprets civilian behavior mostly as resulting from incentives and

State Use of Indiscriminate Violence

35

disregards prewar loyalties for his theory, Steele uses the opposite direction and asks what happens if combatants cannot change civilian loyalties. In order to separate random violence from violence against groups she introduces the term “collective targeting” (Steele, 2009, 422). However, for collective targeting to take place actors again have to gain information on which groups to target. In the case of Colombia Steele argues election results were used by militias to identify which neighborhoods or villages were disloyal and, therefore, were targeted. In her view, attacks on civilians then directly lead to the displacement of those disloyal neighborhoods. A similar argument about displacement was made previously by Azam and Hoeffler (2002). They provide a game theoretic model that includes the option of reducing the military capability of the opponent by displacing the population. The government weighs the costs of displacing civilians against the costs of fighting and opts for the former if it is a cheaper way of discouraging the rebels. They argue that the strategy of displacement is more likely in cases where the civilian population is small and the strategy is, thus, feasible. Finally, Johnston (2009) argues that indiscriminate violence can be successful and he summarizes effective mechanisms under a strategy he calls “coercion.” However, he does not give a detailed account of what aspect of this strategy actually makes indiscriminate violence effective and although he calls the strategy “coercion” his main argument seems to be more about the physical separation of rebels and civilians. In Johnston’s view the coercion strategy that seeks to separate civilians from insurgents can be successful by using three different tactics: First, clear and hold operations, which Johnston sees as intrinsically violent against civilians; secondly, population relocation; and thirdly, bombardments of the civilian population in rebel heartlands, which induce flight (Johnston, 2009, 62). He argues that all three aspects of the strategy lead to a separation between civilians and insurgents and ultimately contribute to a higher probability of the government winning. However, he claims violence has to be used in a targeted manner and in moderation. A strategy of annihilation which is intended to eradicate the civilian population will generally fail as a strategy, which is in stark contrast to the above-mentioned approaches focusing on brute force. As governments using this strategy intend to target all civilians without discrimination, civilians have an incentive to seek the protection of the insurgents, which is close to the argument of Kalyvas and Kocher (2007b). Limitations of the brute force and displacement arguments. The argu-

ments outlined above explain a particular rationale of indiscriminate violence. Violence may be used to separate civilians from rebels either through deadly force or through displacement. However, there are limitations to most of these studies.

36

What Do We Know About Indiscriminate Violence?

Most approaches on brute force seem to have a narrow understanding of the effects of violence. Most importantly, virtually all of the studies seem to concentrate only on the benefits of indiscriminate violence and ignore its potential negative consequences. Only Johnston (2009) takes this into account when he introduces his second strategy of annihilation that he argues is counterproductive. As already discussed in the previous sections, it is very plausible that indiscriminate violence has counterproductive effects through the legitimacy, revenge and security mechanism. Yet, this potential drawback of indiscriminate violence against civilians is not discussed. The omission is even more striking since the authors on brute force for the most part focus on violence that does not discriminate between innocent and guilty individuals. This means that it can be expected that brute force violence has pronounced counterproductive effects. An implicit assumption (explicit only in Steele’s work) seems to be that the civilian population is already supporting the rebellion in such a wholehearted way that additional negative consequences are impossible. However, such an assumption is potentially only true for few cases. Kalyvas goes to great lengths to point out that civilian support is often situational to the degree of control that an actor has (Kalyvas, 2006, 92–100). Yet, if this assumption is not justified for most cases and counterproductive effects are possible, killing civilians as a strategy of physical separation would only work if its positive effects outweigh its negative effects. This would most likely be the case if the perpetrator killed so many civilians that there is simply no one left to join the rebels. Such a logic seems to be behind the argument of Valentino et al. (2004) who focus on massive amounts of violence. However, other arguments that assume even small amounts of killings can help to win fit the logic less well (e.g., Fjelde & Hultman, 2014). These arguments cannot demonstrate that the positive effects outweigh the counterproductive effects and are thus less convincing. The approach on displacement by Steele is interesting since it outlines a targeting strategy. Steele explicitly raises the awareness of the possibility that perpetrators are somewhat selective even when they use brute force. By targeting those areas that are perceived as disloyal, the perpetrator introduces a distinction between potential targets and nontargets that induces flight only from opponents. However, her approach does not provide a clear picture of the mechanisms that induce flight (which people are attacked and which are coerced) or the interplay of both positive and negative effects of indiscriminate violence. The approach of Johnston (2009) explores different ways to separate civilians from rebels. Still, both categories of violence, “coercion” and “annihilation” remain very broad. Coercion comprises violence ranging from hold-and-clear operations to bombings while annihilation seems to cover all violence on a massive scale. Johnston does not provide informa-

State Use of Indiscriminate Violence

37

tion on the boundaries of these two strategies but he mentions that the coercion strategy is associated with some sort of targeting (e.g., rebel strongholds). Bombing strongholds may then induce flight and thus separate rebels from civilians. Thus, in essence the argument of Johnston is the same as that of Steele. Johnston adds that untargeted massive violence (annihilation strategies) will lead to more resistance. However, he does not discuss why targeted violence leads to displacement while untargeted violence leads to resistance. It seems that approaches on the physical separation between civilians and rebels are potentially important in explaining indiscriminate violence against civilians. However, at least one important qualification is in order. Most approaches unjustifiably seem to disregard the potential counterproductive effects of indiscriminate violence. Yet, it seems that brute force can only avoid these provocative effects when it kills substantial amounts of the population. In these cases limited negative effects could be expected; in all other cases it remains unclear how far the positive effects for the perpetrator would surpass the negative effects. Coercion: Forcing Civilians into Support

The previous accounts focused on the physical separation of civilians from rebels to decrease rebel support. A second line of thought emphasizes coercion. Here, authors argue in general that indiscriminate violence, very much like selective violence, is used in order to change civilians’ behavior, not their physical presence. Indiscriminate violence is intended to force civilians into supporting the perpetrator or at least into withholding their support to the opponent (e.g., a rebel group). However, the authors concerned with indiscriminate violence differ in the way they model the effectiveness of violence. To begin with, Kalyvas himself proposes an argument. Although he makes a strong case against the usefulness of indiscriminate violence he nevertheless points out that it may coerce civilians if the rebels cannot provide security for them. The counterproductive effects of the security mechanism discussed earlier rest on the assumption that joining the rebellion may be safer than staying neutral. However, if the opponent cannot protect civilians from violence the government might use indiscriminate violence without negative consequences. Kalyvas writes: This analysis yields the following prediction: incumbents can afford to be indifferent about the type of violence they use when insurgents are unable to offer any protection to civilians. Put otherwise, costly discrimination can be dispensed with when insurgents are weak. When this is the case, indiscriminate violence does succeed in paralyzing an unprotected population. (Kalyvas, 2006, 167)

38

What Do We Know About Indiscriminate Violence?

Interestingly, this argumentation is the opposite of what Johnston (2009) concluded. In Kalyvas’ description overwhelming violence and the inability to find a way out for civilians may lead to support for the perpetrator. A very similar approach is presented by Zhukov (2014) who also focuses on irregular warfare. He builds a formal model in which he argues that both sides try to kill enemies but do so only with a certain degree of accuracy. The lower the level of accuracy (or selectivity) the higher the number of innocent civilians killed. If both parties use the same amount of violence (i.e., kill the same number of people) the side with the higher selectivity will kill more actual opponents. However, Zhukov argues that the less selective actor can offset the disadvantage by escalating violence to kill more people and with it also more opponents. The result of this strategy is an outbidding logic, where both actors will use more and more violence to decrease civilian security and increase support for their own side. The outbidding ends when one actor reaches his physical or moral limit of violent acts he can perpetrate. From this level on the civilian population will bandwagon with the stronger actor. In essence, the model says that whenever it is very risky to remain neutral, civilians will side with that actor who inflicts more harm on the opponent. As a result, in Zhukov’s model indiscriminate violence can be effective, but only when used in very high amounts (Zhukov, 2014, 43). Schubiger (2014, 133–190) argues that indiscriminate violence by the government can lead to civilian support for the government and counterinsurgent mobilization (i.e., setting up progovernment militias). She points to several general reasons why this may be the case. Schubiger outlines that in cases where violence is predominantly perpetrated in face-to-face interactions (as opposed to bombings or artillery shelling) civilians can try to signal to the government that they will support it. Setting up progovernment militias can be a way to provide such a signal in the hope of preventing attacks from the government. Secondly, she argues that under the influence of indiscriminate violence the militarization of local governance structures will become more widespread because of disruptions in traditional authority structures or a shift in local preferences towards security provision (Schubiger, 2014, 142). Another approach is put forth by Douglass (2012) who links selective violence with “random targeting” in a formal model. He explicitly defines random targeting as nonlethal targeting like the use of police checkpoints, cordon and search operations, or random arrests (Douglass, 2012, 131). This way his approach is outside the scope of this book. However, Douglass’ insight is important enough to be included here. He argues, that whenever insurgents hide among villagers, the government can target them selectively, if information is available, or use measures like random arrests in the hope of capturing some of them by chance. On their part, villagers can decide to pass on information on insurgents to the government. However, this decision influences the probability of the government using selec-

State Use of Indiscriminate Violence

39

tive or random targeting as well as rebel retaliation. The more information they give on existing rebels, the more rebels will be captured or killed selectively. This decreases the potential success of random arrests since fewer rebels hide in the village, which reduces the government’s incentive to use random targeting. Therefore, by giving information on rebels to the government villagers can prevent indiscriminate searches, yet they must balance this against potential rebel retaliation. Limitations of the coercision arguments. These approaches on indiscrimi-

nate violence as coercion make several very important contributions, but again it seems necessary to point out the gaps in order to make progress. Kalyvas (2006) emphasizes that indiscriminate violence can lose its counterproductive effects if the rebels cannot protect civilians. When the government enjoys an overwhelming power advantage, indiscriminate violence will not produce additional rebels. However, this does not explain why civilians should support the government. As pointed out above, the ineffectiveness and the counterproductiveness of indiscriminate violence are two separate stories. As a consequence, a negligible counterproductive effect (i.e., only marginal increases in rebel recruits) is not tantamount to becoming effective (coercing the civilians into supporting the government). Just because indiscriminate violence does not help the rebels does not mean it will help the government. The problem with this argument is, therefore, that it does not explain why the perpetrators would use indiscriminate violence in the first place. Even with a lack of protection by the rebels, there is no incentive for the population to support the government when we have no theory that explains the coercive effects of indiscriminate violence. In addition, the loss of counterproductiveness that Kalyvas proposes only concerns the security mechanism that Kalyvas and Kocher (2007) point to. Yet, the revenge mechanism might further produce counterproductive effects, even when the security mechanism is inactive. Thus, it is highly unlikely that the simple lack of rebel protection is sufficient in the loss of all counterproductive effects of indiscriminate violence. Zhukov (2015) seeks to extend the argument of Kalyvas. His model of escalating indiscriminate violence is interesting as it introduces an interaction between actors that is usually ignored. However, implicit in Zhukov’s model is the assumption that indiscriminate violence does not produce counterproductive effects; it is just less effective in killing opponents than selective violence. Thus increasing indiscriminate killing only increases the number of rebels killed. If indiscriminate violence had counterproductive effects the perpetrator could not unambiguously profit from increasing violence. Schubiger (2014) argues that progovernment mobilization is a way to signal loyalty in order to avoid being targeted by indiscriminate violence. This is an important point as it considers the possibility of avoiding violence.

40

What Do We Know About Indiscriminate Violence?

However, Schubiger does not go into the details of this argument and it remains unexplained under which circumstances civilians engage in such a signaling strategy. Put another way, it is unclear what makes civilians signal loyalty instead of joining the rebellion, as Kalyvas would hypothesize? Douglass (2012) provides perhaps the most interesting approach explaining the interplay between selective and indiscriminate violence. He outlines how villagers provide information on rebels to the government in order to make indiscriminate violence less likely. Indiscriminate violence can thus be seen as a threat in the background and civilians want to prevent it from happening. Information provision may thus be another option for civilians to signal loyalty to the government in order not to be targeted. However, Douglass’ approach also assumes that indiscriminate violence does not produce negative effects. Random targeting and selective targeting are assumed to have the same cost. This means the model excludes the detrimental effects of random targeting on the population. The model also explicitly focuses on nonlethal forms of random targeting which limits its applicability. As a consequence, the model cannot explain how lethal violence can become a credible threat. In sum, the theoretical consideration provides several important points for the effectiveness of indiscriminate violence. However, they also suffer from several overarching problems. A) They generally do not reconcile the counterproductive effects of indiscriminate violence with its potential for coercion. Yet it would be necessary to explain how the benefits of indiscriminate violence can surpass the counterproductive effects in order to make a strong argument for a positive net effect of indiscriminate violence. B) Most accounts on coercion seem not to engage with the details of coercion. It remains unexplained how coercion is supposed to work, which groups are to be attacked, which groups are supposed to be coerced, and what incentives individuals have to comply. As pointed out in the previous section on the counterproductiveness and ineffectiveness of indiscriminate violence, the perception of violence as arbitrary is what makes this violence ineffective and counterproductive. Therefore, it would be important to have a theory of group targeting that provides a logic for the effectiveness of violence as coercion. Group targeting is hinted at by several authors, but so far this concept has not been elaborated on (Johnston, 2009; Schubiger, 2014; Steele, 2019). In order to improve the understanding of indiscriminate violence, it seems necessary to improve our understanding of coercion and targeting. Conclusion The term indiscriminate violence has been used ambiguously. On the one side, indiscriminate violence is defined as violence against groups and

State Use of Indiscriminate Violence

41

treated as a strategic tool, yet on the other side, in its impact it is often described as if it was only perceivable as random. It seems that part of the confusion stems from the fact that Kalyvas’ categorization was conceived as a tool to describe the intentions behind violence (Kalyvas, 2006, 142). As far as intentions go, actors can target individuals or groups. Yet it is difficult to imagine that an actor selects individuals randomly. Therefore, in studying intentions of violence the dichotomous distinction between individuals and groups as targets makes sense. However, when we turn to the effects of violence, this is different. Selection can be perceived at the individual and at the group-level. In addition, civilians may fail to perceive a selection pattern and see violence as arbitrary. As a consequence, Kalyvas’ categorization cannot adequately capture the different possibilities in perceptions and necessarily conflates them. Although Kalyvas mentions the importance of perceptions he does not use perceptions and differences in perception as an explanation of the effectiveness of violence (Kalyvas 2006, 145). For that reason, it seems very important to further disaggregate the categorization proposed by Kalyvas in order to study the impact of violence. Violence can be perceived as selective, group-selective or arbitrary (I will come back to this point in Chapter 4). Arguments on the ineffectiveness of indiscriminate violence seem to be restricted to specific cases. A) The inability of indiscriminate violence to coerce or shape behavior is attributed to the fact that this violence makes no sense to those on the receiving end. As I have pointed out, this is only the case when violence is perceived as arbitrary. If violence, as Kalyvas himself acknowledges, follows a targeting pattern (even if that is on the grouplevel) it can be understandable. Targeting at the group-level could potentially lead to differing perceptions by those belonging to the targeted group and those belonging to other groups. The former may perceive violence as arbitrary while the latter perceive it as selective. B) The potential inability to shape rebel behavior either because of a missing link between civilians and rebels or because of diverging interests between rebels and population cannot be taken as a general argument against the effectiveness of indiscriminate violence. While these are important restrictions limiting the potential reach of the effect of indiscriminate violence, they preclude neither effects on the population nor those on rebel organizations. Arguments on the counterproductive effects of indiscriminate violence make important contributions but are all in need of qualification. A) The economic opportunity cost argument that is sometimes raised does not seem to explain counterproductive effects of lethal indiscriminate violence itself. B) The legitimacy argument may have an impact, but previous research indicates that this will only be the case when the costs (no risk) or the opportunity costs (the benefits of neutrality) for joining a rebel group are low. C) The revenge mechanism seems to be strongly supported by the previous literature.

42

What Do We Know About Indiscriminate Violence?

Yet, the reach of the effect may be restricted mostly to direct relations. Although the impact of revenge based mobilization can increase in the context of blood revenge, it will still be confined to a few individuals mobilized per person killed. D) The security mechanism seems to explain massive mobilization best. Still, importantly, the effect is not triggered by all forms of indiscriminate violence. For the security mechanism to become active it is necessary that a person perceives himself or herself as a potential victim of violence. However, it is not necessarily the case for all people when violence is perceived as group-selective. For those individuals whose groups are not being targeted, violence does not necessarily increase insecurity. Several authors emphasize that indiscriminate violence can be effective by separating the civilian population from the rebels. While this seems in principle plausible, most studies do not account for the fact that counterproductive and positive effects will both be present and do not explain under which circumstances violence produces higher positive than negative effects for the perpetrator. Given the possibility of high counterproductive effects it seems that a positive net effect can only be achieved in cases where the perpetrator exterminates the targeted population. Finally, studies arguing that indiscriminate violence can coerce introduce several important ideas that I will take up over the course of this book. However, they are not able to explain the puzzle of indiscriminate violence as they do not address critical aspects. As outlined above, some selection and targeting process is necessary to make indiscriminate violence effective. Otherwise indiscriminate violence will be ineffective and highly counterproductive. However, the current accounts do not explain how this targeting process works and what its effects are. Interesting points are raised by Douglass and Schubiger, who point to the fact that potential indiscriminate violence can be seen as a threat and that civilians need a strategy to signal loyalty, respectively. Unfortunately, they too do not yet provide a way to reconcile the counterproductive effects of indiscriminate violence with its potential effectiveness. Overall, the current literature is not yet able to solve the puzzle of why an actor should use indiscriminate violence if it is in fact counterproductive. This leaves us with several tasks. First, it is necessary to provide a better categorization of the perceptions of violence. Since the current categorization of selective vs. indiscriminate violence cannot adequately capture perceptions we will have to further disaggregate categories. Secondly, it is important to establish a framework for violence that encompasses all effects (positive and negative for the perpetrator) in order to understand the tradeoff between them. If positive and negative effects are not both captured it is impossible to know whether the net effect of violence works in favor of the perpetrator or not. Thirdly, in order to explain the coercive effect of indiscriminate violence it is necessary to use a more sophisticated understanding

State Use of Indiscriminate Violence

43

of coercion than is currently employed. Since coercion is based on the communication of threats we will have to establish exactly who is the target of violence and who is the addressee of the threat. Notes 1. A similar argument can also be found in Balcells (2010) and Balcells (2017). 2. A similar argument can be found in Mason (1989) and in Leites and Wolf (1970, 109). 3. Kalyvas and Kocher point out that the evidence in Wood’s book is not sufficient to establish that participation of the kind that Wood investigates is riskier than nonparticipation (Kalyvas and Kocher 2007, 185). 4. A category mistake is the semantic or ontological error to represent a phenomenon that belongs to category A as if it belonged to category B (Ryle, 2009). 5. It is important to note that I am focusing on a peripheral part of Kalyvas’ work here. For an overview of work on the so-called control-collaboration model see Kalyvas (2012). For authors further testing and extending the idea of selective violence see e.g., Kalyvas and Kocher (2009), Vargas (2009), Bhavnani, Miodownik, and Hyun Jin (2010), and Bhavnani, Miodownik, and Choi (2011). 6. Souleimanov and Siroky (2016) also argue that indiscriminate attacks can be effective in coercing a community (Souleimanov & Siroky, 2016, 699–706), but they describe this effect as ephemeral considering the provocative effects that would later appear. This is a point I will come back to in the next section. 7. A similar distinction of approaches can be found in Lyall (2009) and Zhukov (2014, 13). 8. See e.g., Zhukov (2015), Hultman (2009), or Johnston (2009). Schelling argues that brute force and coercion can be characterized by their intentions. While brute force intends to unilaterally destroy, coercion threatens to inflict pain if a victim does not behave in a certain way (this is what he calls the power to hurt).

3 Nonstate Use of Indiscriminate Violence

I defined terrorism as nongovernment indiscriminate violence against civilians in Chapter 1, where I also argued that the literature on nongovernment indiscriminate violence largely remains separated from the literature on government violence and often uses the label of terrorism to describe the same phenomenon. However, the term terrorism is also used in a broader sense encompassing violence targeting security services or selective targeting. For instance the selective attacks of Euskadi Ta Askatasuna (ETA), a separatist group operating in Spain and France, on politicians, the police or the military are mostly considered terrorism (De la Calle & Sanchez-Cuenca, 2013). This broad definition of “terrorism” will not be used here. Since I am only interested in the effects of indiscriminate violence and, particularly, group-selective violence, I will focus exclusively on arguments that outline how the indiscriminate targeting of civilians produces certain effects. Other arguments, e.g., on selective targeting, or violence against military targets, are excluded. Nevertheless, caution is still needed. Unfortunately, some authors who research terrorism are inconsistent in the application of their definitions and the line between attacks on military targets (mostly called guerrilla warfare) and violence against civilians is blurred. For instance, Moghadam (2006) and Goodwin (2006c) criticize the work of Pape, because attacks on military personnel and attacks on civilians are often lumped together (Pape, 2003, 2005). Pape uses the Hezbollah attack on the US and French troops in Beirut in 1983 as an example of terrorism, although this attack does not fit his own definition of terrorism. In addition, Louise Richardson’s book What Terrorists Want offers similar instances of mislabeling. Richardson’s first example of a terrorist attack in Chapter two is the

45

46

What Do We Know About Indiscriminate Violence?

Clerkenwell explosion in 1867, where the Irish Republican Brotherhood tried to free one of their members from prison. (For a detailed criticism of Richardson’s book see also Goodwin, 2006a) Again, this case does not fit her own definition of terrorism, as it only accidentally killed civilians and did not target them. However, as I will elaborate in detail in the following chapters, the difference between targeting civilians intentionally and killing them accidentally is huge when we consider the effects. If we thus focus on this subset of the literature, it becomes apparent that research on the military or political effectiveness of nongovernment indiscriminate violence against civilians has only recently got under way. Pape (2003) and Abrahms (2006) were among the first to systematically address this issue, with recent studies from among others Fortna (2015) and Aksoy, Carter, and Wright (2015) providing important improvements. Yet, interestingly, the study of nongovernment violence seems to have undergone the opposite development from the study of government violence. While research on government indiscriminate violence until recently mostly assumed that violence is overwhelmingly counterproductive, in terrorism studies the opposite was the case. Here a widely held assumption was until lately that terrorism works (see e.g., Kydd & Walter, 2006; Pape, 2003). 1 This relative consensus is challenged now in both fields with research on government indiscriminate violence emphasizing the possibility of positive effects (e.g., Lyall, 2009), and the literature on terrorism pointing out the ineffectiveness of violence (e.g., Abrahms, 2006). This puzzling and contradictory development is in itself an indication that current accounts on indiscriminate violence and terrorism struggle to adequately capture the effects of violence and underscores the need for better theoretical developments. In the following I will review arguments on the positive and the negative view on nongovernment violence and terrorism. Given that much of the negative view on nongovernment violence emerged as a reaction to the positive view, I will reverse the order and start with the positive view. Finally, I need to address an additional terminological issue. In this chapter, I will mostly use the term terrorism in order to stay true to the majority of the literature. Nevertheless, I will use the term only in the sense of nongovernment indiscriminate violence against civilians. If terrorism is used in another sense, I will mention this explicitly. The Positive View: Terrorism Works As already mentioned, until recently the generally held opinion was that terrorism can be an effective strategy. Exemplary for the writings of many, Bruce Hoffman writes: “Although governments throughout history and all

Nonstate Use of Indiscriminate Violence

47

over the world always claim that terrorism is ineffective as an instrument of political change, the examples of Israel, Cyprus, and Algeria [...] provide convincing evidence to the contrary” (Hoffman, 2006, 61). Similarly, Andrew Kydd and Barbara Walter conclude: “Hijacking planes, blowing up buses, and kidnapping individuals may seem irrational and incoherent to outside observers, but these tactics can be surprisingly effective in achieving a terrorist group’s political aims” (Kydd & Walter, 2006, 49). Nevertheless, explicit and detailed specifications of the mechanisms that link indiscriminate violence against civilians to changes in civilian or political behavior are at the same time in short supply. Compared with the vast amount of literature on terrorism in general there is surprisingly little theorizing on the effects of this sort of violence. I identified three effects connected to indiscriminate violence which are deemed to be positive for the perpetrator: First, the effect of coercion. A coercive effect is realized when a terrorist attack on civilians convinces the government or the population to change their behavior in order to avoid additional attacks. Second, the effect of provocation. A provocative effect may be realized when terrorism provokes a violent backlash, usually from the government that then alienates the population and helps the terrorists in recruiting. Third, the effect of spoiling. The spoiling effect is similar to the provocation effect, but here an attack very specifically seeks to undermine negotiations between the government and a moderate group. Since the literature on the theoretical mechanisms of spoiling is small I will discuss it in conjunction with provocation. The effects I identified are commonly cited to be the result of the most important strategies of indiscriminate violence against civilians in terrorism studies (e.g., Hoffman & McCormick, 2004, 244, Jones & Libicki, 2008, 4). Other authors have discussed additional strategies and effects as well (e.g., Crenshaw, 1981, 386–387, Kydd & Walter, 2006). Kydd and Walter (2006) point to five “strategic logics” of terrorism: attrition, intimidation, provocation, spoiling or outbidding. Martha Crenshaw distinguishes five “proximate causes” of terrorism: attention, disrupting the government, affecting public attitudes, provocation, or increasing organizational functions of the terrorist organization (Crenshaw, 1981, 386–387). However, the logic of these strategies only partially overlaps with my focus on indiscriminate violence against civilians and, therefore, it is necessary to exclude some of them for my purposes here. The Effect of Coercion

The notion that terrorism can have a coercive effect is frequently found in the literature. Either explicitly or implicitly, many researchers on terrorism follow this argument and it is widespread among proponents and opponents

48

What Do We Know About Indiscriminate Violence?

of the positive view alike (e.g., Abrahms, 2006, 2013; Enders & Sandler, 2011; Goodwin, 2006b; Pape, 2003, 2005; Thomas, 2014; Wilkinson, 2006). Early research already pointed out that terrorism seeks a demonstration effect. Thornton (2006) writes: “The relative high efficiency of terrorism derives from its symbolic nature. If the terrorist comprehends that he seeks a demonstration effect, he attacks targets with a maximum symbolic value” (Thornton, 2006, 77). However, few authors provide an explicit theory of coercion. One of the rare in depth discussions of the coercion mechanisms can be found in the work of Pape (2003). In his study on suicide terrorism Pape points out that the main aim of terrorism is to coerce the opposing government into changing its policy or to change the population’s behavior: At its core, suicide terrorism is a strategy of coercion, a means to compel a target government to change policy. The central logic of this strategy is simple: Suicide terrorism attempts to inflict enough pain on the opposing society to overwhelm their interest in resisting the terrorist’s demands and, so, to cause either the government to concede or the population to revolt against the government (Pape, 2003, 346).

Comparing terrorism to his work on military coercion in international wars (Pape, 1996), Pape notes that terrorism works under reversed conditions. While military coercion is mostly perpetrated by the stronger side, terrorism is a weapon of the weaker side. Pape argues that in military coercion actors can choose among two options of denial and punishment, but terrorists cannot.2 As denial presupposes a power advantage in order to deny the opponent a certain action (e.g., destroy some tanks to make it impossible for him to conquer or hold onto a certain territory), terrorists are unable to use this strategy. The punishment strategy, however, is the same as for states. The aim is “to cause mounting civilian costs to overwhelm the target state’s interest in the issue in dispute and so to cause it to concede the terrorists’ political demands” (Pape, 2003, 346). In essence, Pape’s explanation of terrorism is the same as for strategic bombing proposed in his earlier work (Pape, 1992, 1996). However, notably, he seems to be less pessimistic about the possibility of success, even though the perpetrator is now consistently considered to be the weaker side. The reason for this, he argues, lies in the fact that terrorism is mostly directed against foreign occupations and democracies that have only a limited interest in the issues that the terrorists demanded. 3 By contrast, if the interests at stake are high for the targeted government, terrorism would be unlikely to succeed (Pape, 2003, 344). A similar view on terrorism is presented by Goodwin (2006b), who focuses on indiscriminate violence. Goodwin observes that terrorist indiscriminate violence against civilians is rarely random, but more often directed

Nonstate Use of Indiscriminate Violence

49

against an opposing social or ethnic group. This targeting principle, which he calls “categorical terrorism,” is more likely when the terrorists perceive a population as complicitous civilians: When they employ a strategy of categorical terrorism, revolutionaries generally threaten and attack what we might call “complicitous civilians.” Revolutionaries view these categories of civilians as complicitous insofar as they are believed to (1) routinely benefit from the actions of the government or state that the revolutionaries oppose, (2) support the government or state, and/or (3) have a substantial capacity to influence or to direct the government or state (Goodwin, 2006b, 2037).

However, the proposed logic of coercion is similar to that of Pape. ”The main strategic objective - the primary incentive - of categorical terrorism is to induce complicitous civilians to stop supporting, or to proactively demand changes in, certain government policies or the government itself” (Goodwin, 2006b, 2038). It is important to note, however, that Goodwin explicitly argues that the terrorists seek to coerce the civilian population. Kydd and Walter (2006) provide another rational choice argument on coercion. Importantly, they argue that terrorists seek to signal resolve and their ability to inflict pain in the future to their opponents.4 Since the government is unsure about the resolve and the power of the terrorists, terrorists will have to use costly signals in order to convince the government that the threat of future attacks is real. Kydd and Walter point out that three factors in particular influence the effectiveness of the threat. First, the interests of the target of coercion are important. Targets will only yield if they have low interests in the issue at stake. Second, the possibility of retaliation makes a difference for terrorists. Because terrorists are almost by definition the weaker actor, they must fear the superior power of their opponent’s reaction. If a target can retaliate, for instance, in a genocidal manner, the potential cost of a coercion strategy may become too high. Third, the ability of a target to absorb the pain inflicted is important. If a target is able to withstand a larger number of attacks, it will be a less favorable target since more resources need to be invested to create a substantial threat (Kydd & Walter, 2006, 60–64). Furthermore, the coercion argument of terrorism studies has found its way into the literature on the behavior of rebel groups. For instance, Lisa Hultman, building on the insights of terrorism studies, argues that rebels, very much like terrorists in general, seek to raise the expected costs of fighting for the governments when they attack civilians. Hultman writes: “When a rebel group targets civilians, there are two costs that the government may incur: the political cost of not being able to defend its population, and the military cost of losing control over areas where the rebels instill fear among the population and consequently the fight against the rebels will be under harder conditions” (Hultman, 2007, 207–208). Both costs should

50

What Do We Know About Indiscriminate Violence?

lead to decreased government support. She points out that this strategy should be especially relevant in the case of military weakness of the rebels. In this case rebels cannot force the government by military means into making concessions and, thus, may need to coerce the government via violence against civilians (Hultman, 2007, 208). The above accounts on the coercive logic of terrorism as indiscriminate violence against civilians provide important insights. They give a clear account on coercion and most of these studies point out that terrorism must be conceptualized as a threat of future violence and it is this threat that may lead to policy change. Viewed in this way it becomes clear that civilians and governments targeted by violence face a trade-off between changing policy according to the demands of the terrorists and potentially being targeted in the future. They will do the former whenever the expected cost of future violence outweighs the cost of policy change. As a consequence, terrorists are often unlikely to achieve policy changes that touch on vital interests. Since they usually cannot pose a lethal threat to the government, the concessions governments are willing to make are limited. The focus on coercion also raises the attention for another feature of studies on terrorism. All of the above approaches highlight the fact that the target of violence and the audience of the threat differ in terrorism. For instance, referring to terrorist violence Schmid writes: “Terrorists use it more often indiscriminately and at random against a target group. Hence the enhanced fear they manage to generate among a much wider audience” (Schmid, 2005, 137). Taking this idea even a step further Schmid also compares terrorism to propaganda: “Violence and propaganda have much in common. Violence aims at behavior modification by coercion. Propaganda aims at the same by persuasion. Terrorism can be seen as a combination of the two. [...] Terrorism, by using violence against one victim, seeks to coerce and persuade others. The immediate victim is merely instrumental and serves to achieve a calculated impact on a wider audience” (Schmid, 2005, 142). Others have made the same point. Enders and Sandler (2011) write: “Terrorists broaden the audience beyond their immediate victim by making their actions appear to be random, so that everyone feels anxiety” (Enders & Sandler, 2011, 4). Similar remarks can be found in the working definition of terrorism by Lutz and Lutz (2013, 8–9) and the consensus definition of terrorism developed by Schmid (2011), to name just two of many. As I will outline later, this feature of indiscriminate violence has been under-appreciated in previous approaches of government violence. Yet, even in terrorism studies it often remains an aspect that is mentioned on the side without further discussion. However, as I will argue below, it will be necessary to build on this basic insight and integrate it into a wider framework. The literature remains heterogeneous regarding the supposed audience of the coercive message (the civilian population, the government, or both). Pape

Nonstate Use of Indiscriminate Violence

51

argues that terrorists may try to coerce both, but in his discussion he clearly focuses on the government; Kydd and Walter concentrate only on the government as the target of coercion; Goodwin focuses on the civilian population. As outlined in the introduction and in the previous section I think it is analytically important to prioritize the impact on civilians. While it is true that ultimately terrorists aim for a change in policy, the immediate impact of terrorism as indiscriminate violence against civilians is felt among civilians. However, putting the focus on civilians as the main audience also has important consequences for some of the arguments. For instance, Pape heavily concentrates on the interests of states in explaining why terrorists may not reach their goal. Yet, when the state is exchanged for civilians as the target audience, interests may be structured quite differently. While the government may be willing to trade some loss of life for a policy goal, civilians may prioritize security. In such cases terrorism has an impact on the behavior of civilians but not necessarily on the government. Government behavior on the other side will depend to a good degree on the actual, perceived or anticipated behavior of civilians, at least in open and democratic political systems. Similar to the positive view on government indiscriminate violence, the negative consequences of nongovernment violence are not considered systematically. Some authors mention in passing the possibility of a rallyaround-the-flag-effect, i.e., increased support for the government by the population (Fortna, 2015; Park & Bali, 2017; Williams, Koch, & Smith, 2013). However, the possibility that indiscriminate violence leads to negative consequences for the perpetrator among the target audience via feelings of revenge, insecurity or a loss of legitimacy is largely ignored. Therefore, the literature on the positive view of terrorism remains silent on potential counterproductive effects and so far no effort has been made to reconcile counterproductiveness with coercion. I will come back to this point in the next section, since some authors investigate effects like the rally-aroundthe-flag-effect in isolation from coercion. In sum, the literature on terrorism provides a useful starting point for thinking about the coercive impact of indiscriminate violence. Terrorism studies acknowledge far more the fact that the target of violence and the audience of the coercive message may differ. However, the literature on nongovernment coercion also suffers from some problems. There is considerable disagreement on the audience of coercion and disregard for the counterproductive effects of violence which makes establishing a theory on the effects of violence difficult. The Effect of Provocation

A second strategy of terrorism (i.e., indiscriminate nongovernment violence against civilians) concerns provocation. Provocation is in a way the opposite

52

What Do We Know About Indiscriminate Violence?

of coercion as it seeks to generate a violent reaction from the target audience and not compliance. Here, the terrorist’s intention is to make use of this reaction and turn it against the government. Fromkin famously termed this type of strategy “a form of jujitsu” because terrorists can use the government’s own strength against it (Fromkin, 1974, 688). Terrorists are too weak by themselves, but they can seek to make the government do the job for them. Within the terrorism literature many authors make reference to the existence of this strategy. Martha Crenshaw argues: “Terrorism may also be intended to provoke a counterreaction from the government, to increase publicity for the terrorists’ cause and to demonstrate to the people that their charges against the regime are well founded” (Crenshaw, 1981, 387). However, the provocation argument is complex, comprising at least three steps: First, the terrorists decide to provoke instead of trying to coerce; secondly, the government decides to react in a way that is somehow detrimental to its goals; and thirdly, the population then decides to support the terrorists. Unfortunately, only few authors go into the details of these mechanisms. One exception is Lake (2002) who elaborates on the provocation logic. Lake explains that by attacking, extremists may intend to provoke the target into a disproportionately violent response against the population of which the terrorists are part. The ultimate aim of the extremists is to change the preferences of their own constituency. If the target to an extremist attack lashes out against this part of the population, the population may either experience an increased feeling of revenge or may become convinced that the target itself is extremist and needs to be fought. In both cases extremists could profit from additional support (Lake, 2002, 19–20). Lake thus provides an explicit argument for the usefulness of provocation over a recruitment and support mechanism. He indicates two channels that lead to increased support for the terrorist. On the one hand, the retaliation by the government will necessarily produce harm that will lead to feelings of revenge. On the other hand, a harsh government reaction will make people distrust the government and decrease their belief that a negotiated settlement is possible. These two channels roughly correspond with the legitimacy and the revenge mechanisms discussed in the previous chapter on government indiscriminate violence. Lake explains further that the provocation strategy is a way to change the power relations between the target of provocation and the terrorists. The terrorists are not interested in negotiations now, they want to level the playing field in order to negotiate a better deal later (Lake, 2002, 19). A similar argument is provided by Byman (1998) who focuses on ethnic conflicts. He points out that ethnic terrorists may not only seek to gain support but possibly want to increase government attacks on their ethnic group in order to create a distinct identity among their ethnic constituency. Attacking the government and provoking a violent backlash will widen the

Nonstate Use of Indiscriminate Violence

53

trenches between ethnic groups and may heighten their identification as distinct entities (Byman, 1998, 154–156). Therefore, terrorists may follow this strategy especially on occasions when they want to activate their own group. The consequence of this approach is ultimately the same, the government reaction leads to support for the terrorists. The discussions by Lake and Byman are highly insightful and clarify the provocation strategy in many ways. However, they do not explain all aspects of it. In particular they do not provide a reason why a government could be expected to overreact. After all, the government could realize that an overreaction is not in its interest and, therefore, desist from such an action. One more specific explanation to this problem is given by Blankenship (2016), who does not see the government’s reaction as a strategic action but as an error. Blankenship argues that the main reason why states fall for the trap laid out by the terrorists is their bureaucratic capacity, or their lack thereof. He claims that states with a high bureaucratic capacity have the option to use violence selectively, while those with low capacities do not have that option. As a result, governments use indiscriminate violence only for the lack of alternatives. Finally, closely related to the subject of provocation is the literature on spoilers in peace processes. Spoilers are stakeholders in a peace process who believe that the peace negotiations are not furthering their interest and use violent means to sabotage the process (Stedman, 1997, 5). While spoiling can take many forms, organizations that use terrorism are mostly outsiders to the peace negotiations. For such a scenario, where a moderate party negotiates with the government and an outsider uses terrorism to sabotage peace, Kydd and Walter (2002) provide a game-theoretic argument. They outline a rational choice framework that explains why extremist groups increase the level of terror when the government is negotiating the terms of settlement with a moderate faction. In their view terrorism by extremists is able to increase mistrust between the two negotiating parties. Since the government is unable to know for sure the preferences of the moderate faction, terrorism can lead the government to believe that the moderates are secretly supporting the terrorists. Even in cases where the moderate faction is in fact willing to restrain the extremists, it is possible that extremists win, whenever the government is unable to observe the policing efforts of the moderates. In such cases, the government loses trust in the ability or the willingness of the moderate faction to uphold their bargain and negotiations break down. Thus, the proposed consequences are very similar to those of the provocation strategy. If successful, extremists can sabotage trust between moderates and the government and thus increase their own support. The research summarized above provides again some important insights into the working of the provocation strategy, although some gaps remain.

54

What Do We Know About Indiscriminate Violence?

It seems that most arguments on provocation cannot explain the actual provocative effect. As outlined above, a full argument on provocation comprises at least three steps: a provocative attack by terrorists, an overreaction by the government and increased support for the terrorists by the population. In general, most researchers do not explain all the steps and frequently the focus is on the third step which takes the government overreaction as a given. Once a government is provoked and reacts with indiscriminate violence, this overreaction will lead to increased support by triggering one of the mechanisms that were discussed in detail in Chapter 2: legitimacy, revenge and security. Yet, steps one and two do not receive the same attention. Regarding step one, it is odd that the literature does not provide an argument for the provocative effect of a terrorist attack. Given that a number of scholars argue that terrorist attacks have coercive effects (as outlined in the previous section), it is strange that no author can point out why some attacks should lead to provocation and others lead to coercion. In Chapter 5 I will argue that the effects of indiscriminate violence can consist of both provocative and coercive effects at the same time. It then depends on the net effect of both, whether the terrorist attack is provocative or coercive in the aggregate. However, until now, no author provides an explanation for the parallel existence of coercive and provocative effects. Regarding step two, there is no consensus as to why the government should overreact. If the government is a rational actor, it should weigh costs and benefits of a reaction and only carry out a reaction if it provides a net benefit. However, authors disagree as to why the government is unable to make this calculation. On the one side, some authors argue that it is difficult to steer the fine line between a reaction that is too restrained and a reaction that is too intense (Bueno de Mesquita & Dickson, 2007; Lake, 2002; Lapan & Sandler, 1993). While this argument has its merits, it does not yet provide an answer as to why this is so difficult. What is it that makes the government unable to choose the right dose of reaction? On the other side, some authors argue that the government is either irrational or incapable (Blankenship, 2016; Fromkin, 1974). Again, these arguments have their merit. Especially Blankenship’s point about the administrative capacity of a government is important. However, it may be neither the only nor the most central explanation. As I will note in Chapter 6, government indiscriminate violence often follows very precise patterns and it seems unlikely that governments blindly lash out against the terrorist. The literature on provocation seems to focus exclusively on government reactions. However, as discussed in Chapter 1, at least when focusing on provocation, it is important to take civilian reactions as analytically prior to a potential government reaction since they form the potential targets of future violence. However, none of the existing arguments on provocation

Nonstate Use of Indiscriminate Violence

55

provides an explanation of the reaction of civilians to an indiscriminate terrorist attack. To turn a blind eye on civilian reactions risks underestimating the potential provocation effects. As the terrorists also use indiscriminate violence against civilians it can be expected that terrorist attacks provoke the same effects among the targeted population as government violence: revenge, loss of legitimacy and decreased security. Counterviolence from the population that is independent from a government reaction is therefore a distinct possibility that has received no consideration so far. I will come back to this issue in Chapter 9 and outline that case studies often point to this kind of civilian reaction. Therefore, the provocation of civilians should receive much more attention in future research. In sum, this section highlights several important points. First, the literature on terrorism’s provocation effects mostly focuses on the effects of the government’s overreaction. However, it remains unclear how and why terrorism should produce an overreaction instead of a coercive effect. As with the literature on coercion, research on provocation remains isolated from other approaches on seemingly contradictory effects. Therefore, it will be important to study both effects of indiscriminate violence in a single framework. Secondly, approaches on terrorist provocation exclusively focus on a government reaction. However, it will be beneficial to investigate the effects on the population in order to understand the different provocative effects. The Negative View: Terrorism Does Not Work More and more authors have argued recently that terrorism may not work. Similar to the study of indiscriminate government violence, two types of arguments can be distinguished: On the one side, terrorism may be ineffective in achieving its goals and, on the other side, it may have counterproductive effects. Ineffectiveness of Terrorism

To begin with, some authors argue that terrorism is ineffective. Several researchers highlight the limits of a coercive effect and argue that coercive violence can only be effective if the demands of the terrorists are not too high (e.g., Pape, 2003; Berrebi & Klor, 2006). These arguments are based on a rational logic of coercion where the coercive threat presents the target with an altered set of choices: either face additional violence or give in to demands. Thus, Berrebi and Klor (2006) point out that terrorism with maximalist aims, as for instance destroying Israel, is unlikely to coerce the electorate into making concessions. Since only concessions tantamount to giving up the country would appease the terrorists, the costs may be too high.

56

What Do We Know About Indiscriminate Violence?

As a consequence, in situations where the demand is seen as excessive or where there is uncertainty over the goals of the terrorists, the effectiveness of the terrorist threat to coerce may be reduced. Max Abrahms has provided several arguments that concentrate on an intrinsic inability of indiscriminate violence to coerce. Abrahms (2006) argues that terrorism may be ineffective because the targets of coercion may be unable to understand the threat. Building on correspondent inference theory, Abrahms argues that an observer of terrorism will infer the objective of terrorists from the outcome of terrorism. Therefore, he points out, governments will infer from a terrorist act that terrorists want to destroy them rather than wanting a policy change: “The basic contention is that civilian-centric terrorist groups fail to coerce because they miscommunicate their policy objectives. Even when a terrorist group has limited, ambiguous, or idiosyncratic objectives, target countries infer from attacks on their civilians that the group wants to destroy these countries’ values, society, or both” (Abrahms, 2006, 56). Abrahms refines this argument later in another article (Abrahms, 2013). Here he argues similarly that terrorism does not work, since the targets misperceive extreme violence. However, he links his argument now to the rational choice literature on bargaining. He points out that high levels of indiscriminate violence will not only increase the coercive threat but also make the attacker look less trustworthy. Revisiting the bargaining theory literature, Abrahms (2013) argues that effective bargaining rests on credible threats (about which there is some consensus) and the promise to withhold the threatened action if the target complies (this, Abrahms argues, is generally not taken into account). As in the previous article Abrahms points out that targets of terrorism will infer from extremist violence that the terrorists are unappeasable. The usage of indiscriminate violence thus leads directly to a reduced credibility to withhold violence in the future. As a consequence, the attacker will be less likely to receive concession from the target since the target must assume that the attacker will not stop attacking anyway. Counterproductive Effects

Terrorism may also lead to counterproductive effects, which are mostly summarized under the heading of the “rally-around-the-flag-effect” in the literature (e.g., Chowanietz, 2011; Fortna, 2015; Park & Bali, 2017; Williams et al., 2013). The rally effect is well-known in the International Relations literature describing a phenomenon where public and elite support for the government is increased in response to a diplomatic crisis. Within this literature multiple authors have described potential mechanisms that produce the rally effect.5 A first explanation of the effect was introduced by J. E.

Nonstate Use of Indiscriminate Violence

57

Mueller (1973). Mueller argues that in case of a critical event the people unite behind their government in order to increase the nation’s prospects to master the crisis. Hetherington and Nelson (2003) call this the “patriotism school.” Another explanation comes from Brody (1991), who argues that the rally effect only materializes when opposition opinion leaders refrain from challenging the government. He points out that since the government can monopolize information in diplomatic affairs, at least in the immediate aftermath of a crisis, the opposition cannot provide a credible counternarrative. This line of thinking introduces an important differentiation between the behavior of elites and the population and emphasizes the opinion leadership of elites. However, the literature on the rally effect in international crises is far from having reached a consensus. For instance, some have pointed out that there is no consistent rally effect for military action by the US president (Lian & Oneal, 1993), while others point to more complex mechanisms that produce the rally effect (Chatagnier, 2012). Entman provides an additional contextualization when he argues that since the end of the Cold War international events have become more ambiguous, which makes it much more difficult for a government to frame the crises in a way favorable to them (Entman, 2004). Building on this literature, a number of scholars have argued that terrorism might produce the same rally effects as other foreign policy crises. For instance, Chowanietz writes: “The implications of terrorism for the national interest suggest a priori that this form of political violence offers fertile ground for rallies. Terrorism ticks the same boxes as military and diplomatic crises” (Chowanietz, 2011, 678). He thus argues that terrorist attacks should have the same effect on the political system as other crises. As the nation comes under attack, political opposition parties will rally around the flag in order to defend the country (Chowanietz, 2011, 680). Other authors have followed along the same lines. Among others, Williams et al. (2013), Randahl (2016), and Park and Bali (2017) take the rally effect into account when generating their hypothesis. If the arguments presented above bring some clarity to the debate on the effectiveness of terrorism, they do not all seem plausible. Beginning with the arguments on the ineffectiveness of terrorism, the point raised by Berrebi and Klor (2006) is clearly important. As described by the literature on the coercive effects of terrorism, acts of violence are supposed to threaten the opponent. But the threat is only able to be effective if the potential costs of future violence are higher than the costs of making concessions. This means that terrorism will only be effective within these boundaries. Abrahms’s arguments are different. In a very elaborate discussion Abrahms argues that the usage of indiscriminate violence against civilians undercuts the coercive effects as the strategy itself renders impossible the credibility of withholding violence later. The problem Abrahms thus describes

58

What Do We Know About Indiscriminate Violence?

is akin to the problem a victim of blackmailing has. For instance, if the victim decides to pay a sum for a certain information not to become public, it is difficult to guarantee that the blackmailer will not blackmail again over the same information. Abrahms’s argument is now that because of the method of using indiscriminate violence it will be impossible to convince the target of coercion that violence will be withheld later. Abrahms’s argument is very interesting as it links the method to the commitment problem. However, he remains unclear about the extent to which terrorism leads to a less credible commitment. For Abrahms’s argument to indicate a complete loss of effectiveness of terrorism, it seems necessary that the employment of indiscriminate violence would not only introduce some uncertainty about the trustworthiness (as in the general case of blackmailing, this is always present) but that a perpetrator becomes so untrustworthy that coercion is not possible anymore. Yet, it seems quite unlikely that just the use of violence against civilians itself always has this effect, as it would imply that killing civilians instead of, say, policemen completely blinds all observers and makes them believe that the attacker will never stop until everyone is killed. In this inordinate form, the argument seems to be untenable. Possibly, Abrahms’s argument suffers from a similar problem as Kalyvas’ argument on the ineffectiveness of government indiscriminate violence. Abrahms seems to have an extreme form of random violence in mind when he writes: Rather, citizens of target countries draw a direct correspondence between the extremeness of the challenger’s tactics and preferences notwithstanding the nature of his actual demands. If defenders inferred the extremeness of the challenger’s preferences directly from his tactics, escalation would create a credible commitment problem by rendering his vow to remove the pain unbelievable, regardless of whether the defender were to comply. [...] We should therefore find that defenders are indeed apt to conclude that a challenger harbors extreme preferences by dint of his employing extreme tactics (Abrahms, 2013, 665).

The logic of Abrahms’s argument seems to rest on his assumption that terrorist tactics are the most extreme tactic possible, which is why people will infer the most extreme preferences of the terrorists. However, it is not necessarily the case that terrorism is the most extreme tactic possible. As Pape (2003) notes, terrorists often attack in very clearly defined campaigns and stop once they reach concessions. This points to a clearly defined temporal selectivity. In addition, Berrebi and Klor (2006) point out that periods of relative calm can be seen as an important signal by terrorists that they are able to reduce attacks when concessions are granted (Berrebi & Klor, 2006, 902). This means that at least some terrorists are able to communicate

Nonstate Use of Indiscriminate Violence

59

that they are trustworthy and that they will show restraint when concessions are made. As a consequence, the categorical way in which Abrahms states his argument, that terrorism never works, seems implausible and may rest on a similar misconception of indiscriminate violence as arguments on government violence. Possibly violence against civilians decreases trustworthiness but the extent remains to be clarified. So far researchers have only provided arguments on the ineffectiveness of a coercive strategy and not on the provocative strategy. Seemingly there is no reason to believe that terrorism should be ineffective in reaching a provocative goal. This is especially interesting as it provides an additional reason to refute strong statements about the ineffectiveness of terrorism in general, as made by Abrahms. Since terrorism may often have its first goal in provocation, terrorism might be effective even if the arguments of Abrahms would be correct. Although it may often fail to coerce, this is not tantamount to its being ineffective. It may simply be trying to achieve different things. It is interesting to note that counterproductive effects of indiscriminate violence play only a minor role in the study of terrorism. Very few researchers argue that terrorism, instead of coercing, pushes the population into the arms of the opponent. For instance, Pape, Abrahms, and many others do not even consider the possibility of a rally effect. Compared with the literature on government indiscriminate violence this is puzzling as these effects are well established there. An exception here is the very small number of articles which are directly concerned with the rally around the flag effect. These authors emphasize that it is possible that terrorism leads to increased support for the opposed government. However, it is questionable if the rally effect found in international crises is directly transferable to terrorism. For instance, the study of Chatagnier (2012) includes among others nationalist terrorism in Spain and France, Islamic terrorism in France as well as right-wing terrorism in Germany. It seems at least questionable that all these terrorist attacks have the same consequences. The most obvious reason for this is that the terrorists target different parts of society. Nationalist terrorism frequently targets the majority population as opposed to the minority that the terrorists represent, Islamic terrorism often targets the whole of the democratic society, and right-wing terrorism targets mostly immigrant minorities. As a consequence, nationalist and Islamic terrorism may have the potential to be seen as a crisis for the nation, but right-wing terrorism surely cannot be seen in that way. Here the threat targets a minority and usually the majority society does not react with the same sense of crisis. In addition, it seems important that the mechanisms behind the rally effect need to be better specified. What is it that makes people react to terrorism in a certain way? Until now it has been unclear which property, or properties, of terrorism make it comparable to international crises.

60

What Do We Know About Indiscriminate Violence?

However, part of the problem with the rally effect stems from its similarity with the provocation effect. It may very well be the case that both effects are the same, only interpreted differently. While some may see it as a strengthening of the government that hinders the terrorists to reach a coercive effect, others may perceive it as a provocation of the government that helps the terrorists. It will therefore be necessary for terrorism studies to clearly specify what effects terrorism has and under which circumstances certain effects prevail. It will be one of the aims of Chapter 5 to provide a theory of group-selective violence of this sort. Finally, in the study of the negative view on terrorism again researchers seem to focus unduly on the effects on the government (an exception is Abrahms 2013). However, as discussed in the section on the coercive effect, the effects of terrorism differ for the government and civilians. While terrorists may threaten the government only with painful punishment, they can threaten civilians with death. As outlined in the civil war literature, civilians are very concerned about their physical security and rate this generally higher than their political or economic interests. Therefore, it is quite likely that the perceived size of the punishment for civilians is higher than the perceived punishment for the government. Taken together with the problem that civilians consistently tend to overestimate the risk of becoming a victim of terrorism (see e.g., Goodin, 2006, 123–131), a natural consequence is that civilians will be more likely to be coerced than the government. Thus by focusing on government behavior researchers may have missed effects on a lower level. Conclusion Regarding the positive view, I discussed especially two types of effects: coercion and provocation. On one hand, the literature on the coercive effects of terrorism seems to provide some notable points. Violence is seen as a coercive threat of future costs that can lead to concessions, as long as the concessions are not costlier than the threatened violence. In addition, the literature on terrorism emphasizes that the target of violence and the target of the coercive message may differ. While the literature does not provide a clear picture of who is being targeted (specific groups of civilians or the whole population) and who is being coerced (e.g., civilian population or government), I will use some of these insights to build a theory on groupselective violence. On the other hand, the literature on provocation effects points out that terrorism may lead to an overreaction from the government. However, these arguments mostly focus on the impact of government counterviolence. They provide only limited explanations for a government’s overreaction or the fact that terrorism sometimes leads to coercion and

Nonstate Use of Indiscriminate Violence

61

sometimes to provocation. As coercive and provocative effects are studied mainly in isolation, it remains to be explained why and under what conditions either of them prevails. Regarding the negative view, I discussed two types of arguments: ineffectiveness and counterproductiveness. In many ways, these arguments are similar to the literature on government indiscriminate violence. On the one hand, concerning arguments on the ineffectiveness, there are good reasons to believe that the coercive effect is restricted to cases where the costs of the demand do not exceed the expected cost of the threat. However, the arguments raised by Abrahms, at least in a strong form, seem less convincing. Abrahms argues that terrorism either produces the misperception that the terrorists want to destroy the target or nullifies the trustworthiness of the perpetrator. However, in this strong form, Abrahms’s statements seem implausible as terrorism is not necessarily the most extreme action possible. Even indiscriminate violence can be used with restraint in order to communicate a political message. Apart from the coercion literature, there are no arguments on the ineffectiveness of the provocation strategy. This is interesting in its own right as it indicates that terrorism could at least provoke even if it was unable to coerce. On the other hand, we find one argument on potential counterproductive effects. Some authors point out that terrorism can lead to a rally effect, i.e., an increase in support for the government of the attacked country. However, this argument seems to be closely related to the provocative effects. For researchers theorizing these effects it would be necessary to point out through which mechanisms they come about and under which circumstances they become strong. Unfortunately within the terrorism literature discussions of these mechanisms are virtually nonexistent. In sum, similar issues arise from this overview on nongovernment violence as from the overview before on government violence, although their relative importance is different. First, some theories of terrorist violence indicate that nongovernment indiscriminate violence is selective on the group-level (e.g., Byman, 1998; Goodwin, 2006b), while others emphasize its randomness (e.g., Schmid, 2005). While the controversy is far less developed in terrorism studies than in the study of government violence, a similar divide seems to exist. This divide also seems to translate into some arguments on the ineffectiveness of terrorism. As outlined, Abrahms (2013) seems to implicitly assume some extreme form of violence that is not understandable for the civilian population. However, the targeting and the effects of targeting are seldom discussed. As a consequence, for terrorism studies, too, it will be important to provide a categorization of violence that differentiates between arbitrary and group-selective targeting. Second, the discussion on coercion is more elaborated in terrorism studies than in research on government violence. Overall there seems to be

62

What Do We Know About Indiscriminate Violence?

more awareness of the fact that violence in itself is not coercion. Rather the implicit threat of future violence may coerce civilians or the government. Dread is the lever. In addition, the fact that terrorism studies often distinguish between targets of violence and targets of the coercive message will be very helpful in developing a theory of group-selective violence. Third, as with the study of government violence, nongovernment violence seems to produce both positive and negative effects. However, and especially in terrorism studies, the difference is not so clear cut as is the case for violence committed by governments. Seemingly, the same effects that are deemed counterproductive in government violence sometimes benefit terrorists, at least when they produce a violent overreaction that radicalizes the terrorists’ own constituency. Still, it remains a core problem of the literature that these effects are studied in isolation and not in conjunction. In terrorism research not a single study seems to address the potential tradeoffs explicitly. Yet, it seems highly plausible that violent attacks produce provocative and coercive effects at the same time. It is therefore necessary to provide a framework that studies these effects holistically. Notes 1. To be sure, there were also dissenting views. See, for instance, Laqueur (1987, 137–141) who argues that terrorism had either no effect or even a negative effect for the terrorists. 2. Denial aims at reducing the military capabilities of an opponent to a point where he is unable to attack. Punishment aims at inflicting pain, e.g., by targeting the civilian population; see Pape (1992, 1996). 3. On the relationship between democracy and terrorism see among others Wilkinson (1977), Wilkinson (2006), and more recently Chenoweth (2013). 4. In doing so they follow earlier economic approaches of modeling terrorism that conceptualize terrorism as a signaling game. For instance, Lapan and Sandler (1993) model terrorism as a game between a government and a terrorist group. 5. For a brief overview over the different schools of thought on the rally effect see e.g., Hetherington and Nelson (2003) and Chowanietz (2011). In the following I will build on these accounts.

4 Conceptual Building Blocks

In this chapter I will provide solutions to overcome the most important obstacles identified in the previous chapters. These will then serve as building blocks for the theory in the following chapter. Specifically, I will discuss the conceptualization of indiscriminate violence and provide an improved categorization that differentiates between intentions and the implementation of violence to overcome the problem of confusing different types of violence. I will take a closer look at the concept of coercion and will elaborate on the circumstances that make coercion effective. This step will help our understanding of the positive effects of violence for the perpetrator. Finally, I will provide a framework for the study of indiscriminate violence that incorporates both positive and negative effects. An Improved Categorization of Violence Against Civilians The categorization of violence against civilians that is used most often is selective violence vs indiscriminate violence. As outlined in Chapter 1, selective violence concerns violence targeting individuals based on their individual behavior. Indiscriminate violence is often defined as violence against groups based on the behavior of some within the group. Unfortunately this categorization seems to be insufficient when it comes to assessing the effects of violence. The main problem with these categories is twofold: First, as outlined in Chapter 2 they have been used ambiguously and researchers often equate violence against groups with violence that is perceived as random. Secondly, if we follow the definition of Kalyvas, the

63

64

What Do We Know About Indiscriminate Violence?

categories reflect only the two types of possible targeting intentions, but obscure differences in the possible implementation and the perceptions of violence. They are thus unable to capture the full spectrum of effects. Here, the intentions of violence concern the idea behind a targeting pattern. For instance, an actor may want to punish collaborators selectively. In addition, the implementation of violence concerns the actual usage of violence. The main question is whether actual violence is implemented according to the intentions. Finally, the perception of violence concerns the way that the target audience interprets the implementation of violence. In order to not further complicate things at this point, I will treat the implementation and the perception of violence as one single aspect. For my purposes, this difference is not important as I will focus mainly on the dividing line between intentions vs implementation/perceptions, which separates the idea of using violence from the actual usage of violence. However, it needs to be borne in mind that implementation and perception are separate in theory and practice. While the actual implementation will have an influence on perceptions, the relationship is not deterministic. Perceptions may err in both directions and violence may be seen as more or less selective than its actual implementation. Humans have the common tendency, called apophenia, to seek and perceive patterns even when no such patterns are present. In addition, certain aspects of violence may have an influence on perceptions independently of the implementation of the pattern. For instance, while the actual implementation of a targeting rule is independent of the amount of violence, perceptions may need repeated violence to recognize the targeting pattern and to separate it from informational noise that does not constitute a pattern. It is true in general that a pattern cannot be inferred from one instance or even from a few instances and this problem increases with informational noise. Since people faced with violence may intuitively confront the same problem of inference, they need enough data, so to speak, to establish the pattern. However, once the pattern is established for observers, this may be robust to errors made in the implementation. Thus, perceptions and the implementation of violence have separate implications that the further study of the impact of violence should take into account. As a result of some of these problems, some researchers have already worked on improving this dichotomous categorization. For instance, Abbey Steele introduced the category of “collective targeting.” Collective targeting she defines as violence against groups and indiscriminate violence she defines as violence not following any targeting rule (Steele, 2009). In addition, Elisabeth Wood distinguished between targeting intentions at the individual level, the group level as well as random violence (E. J. Wood, 2010, 299). Wood’s categorization was taken up for instance by Schubiger (2014) who emphasized that the three categories may be seen as one continuum. Similar efforts can be found in the study of terrorism. For instance, Good-

Conceptual Building Blocks

65

win’s notion of categorical terrorism also makes the point that terrorism targets groups instead of being arbitrary (Goodwin, 2006b). While these efforts go in the right direction, it remains problematic for the study of the effects of violence that these categorizations still focus on intentions alone. As a consequence they necessarily ignore the difference between group-selective and arbitrary violence and cannot adequately capture the effects of violence. To begin with, the category of random or arbitrary violence1 should not exist as an intentional strategy. As many authors and indeed Kalyvas himself pointed out, the strategic design of indiscriminate violence against civilians is practically never arbitrary (see e.g., Goodwin, 2006b, 140, Hutchinson, 1978, Kalyvas, 2006, 150, Wilson, 2010, 220). Violence targets either groups or individuals but it is never targeted at the world population as a whole. As a consequence, none of the authors uses the difference between arbitrary violence and violence against groups as an explanation for the effects of violence.2 However, arbitrariness can be introduced in the actual implementation or the perception of violence. Arbitrary violence may result from a process of inadequate implementation or the perception of violence. For instance, if a perpetrator is unable to implement selective violence with precision this will lead to a targeting pattern that is objectively not only selective at the individual level but incorporates a degree of arbitrariness. As a consequence, the categories of selective violence and group-selective violence need to be separated from arbitrary violence. While the first two represent ideal types of targeting intentions or strategies, arbitrary violence is mostly unintentional and results from inadequate implementation or a perception of violence. As a consequence, I will slightly modify the differentiation of Wood and refer to three separate types of violence: selective violence (violence against individuals), group-selective violence (violence against groups) and arbitrary violence (no targeting rule implemented or non perceived). These are, of course, ideal types. In Weber’s sense, they are exaggerated representations for the sake of clarifications and a tool to overcome the messy social reality. In practice we will see mostly mixtures of these categories (Weber, 1978, 1–22). It is important to distinguish between two processes that produce arbitrariness. First, as discussed, violence can become arbitrary due to errors in the implementation and due to perceptions. On the one hand, it is possible that the perpetrator intends to target civilians according to a strict pattern but makes errors in the implementation of this policy. For instance, in the example used by Kalyvas and Kocher (2007b), violence is intended to be selective but has a high margin of error and in practice is not selective anymore. In this case, the actual violence incorporates a high level of arbitrariness. On the other hand, it is possible that the perpetrator is implementing his targeting strategy exactly according to a targeting rule, but that civilians

66

What Do We Know About Indiscriminate Violence?

are unable to perceive the pattern, either because of the strategy itself or because of the surrounding circumstances. For instance, if an actor wants to punish every collaborator but is capable of doing so only in a few instances, people may perceive the targeting as disconnected from collaboration because targeting may also correlate with other factors as ethnicity, location, etc. In addition, the surrounding circumstances may complicate perceptions. For instance, the presence of multiple perpetrators that apply different logics of targeting may make it difficult to perceive a definitive pattern. As others have pointed out, this is what happened in German counterinsurgency during World War II where among others German commanders had extreme leeway in their implementation of reprisals (see e.g., T. Anderson, 1999). So, in the case of a single commander the actual targeting may have followed a distinct pattern, but the targeting pattern of the commanders combined in aggregate did not. As a consequence, observers may be unable to perceive the pattern. Second, and in contrast to the above, group-selective violence always includes an additional process that introduces arbitrariness. Since the targeting rule only specifies the group to be targeted and not the individuals on the intragroup-level it does not matter who is being killed from among that group. Many observers have noted this. For instance, Wright (1988) coins the term “representative violence” and Tim Wilson describes it as the “restricted randomness” of violence (Wilson, 2010, 220).3 As any member of the target group but not the members of other groups can be victimized, violence is restricted and arbitrary at different levels. It is restricted at the group-level and arbitrary at the intra-group-level. This second type of arbitrariness seems to be responsible for much of the confusion in the categorization of violence against civilians, as group-selective violence is intrinsically associated with some arbitrariness. However, it is important to realize that this intrinsic connection only exists on the intra group level. On the group-level, groupselective violence can be perfectly selective (i.e., targeting only those groups that fall under the targeting rule) or be perceived as arbitrary due to inadequate implementation or another perception issue. Consequentially, it is necessary to rethink the conceptualization of the three types of violence and incorporate the implementation and the perceptions of violence. While it is possible to use the categorization of Wood as a basis and to separate selective, group-selective and arbitrary violence, it is important to recognize that arbitrary violence is different from the other two types. While selective and group-selective violence constitute targeting patterns, arbitrary violence is the absence of a targeting pattern. I propose, therefore, to understand these categories as the end points of continua rather than discrete states of being. In this my thinking is close to Schubiger (2014), who also emphasizes that types of violence represent different points on a continuum. However,

Conceptual Building Blocks

67

it is difficult to see it as one continuum where group-selective violence represents the middle category. Perpetrators either select groups or individuals as targets. Therefore, selecting groups does not get closer to arbitrariness due to any erroneous implementation or perception, but is just introducing a different targeting rule. This means that we should think of the categories as representing two different continua. Governments can intend to be selective or group selective. However, violence can be implemented and perceived in a way that includes various levels of arbitrariness (see Figure 4.1). When group-selective violence becomes more arbitrary due to error

Figure 4.1 Continua of Arbitrariness for Selective and Group-Selective Violence

68

What Do We Know About Indiscriminate Violence?

(or for any of the other reasons) it becomes less selective on the grouplevel; yet, it always remains nonselective on the intra-group-level. In addition, as Figure 4.1 shows, we can think of these continua in different layers. The first layer is the intentions of violence. It is difficult to think of a perpetrator intentionally introducing arbitrariness in the targeting pattern, therefore this layer differentiates only between the two targeting patterns. Nevertheless, arbitrariness may be introduced either due to the implementation (layer 2) or the perceptions (layer 3). Although these two layers are not independent from each other, neither are they connected in a deterministic way. However, as already mentioned, for the purpose of this study I will not differentiate between layers 2 and 3 in order to reduce complexity. Differentiating between types of violence in this way, thus frees groupselective violence from the baggage of being generally perceived as random. This differentiation, thus, opens up the possibility to build a theory of groupselective violence in the next chapter. Finally, a justification is in order since this categorization introduces two terms that are not frequently used (group-selective and arbitrary violence). The main reason for the introduction of these terms is that throughout many of my discussions with colleagues several misunderstandings kept recurring. For the category of arbitrary violence all alternative terms (i.e., random violence and indiscriminate violence) tend to be misleading. As outlined in the previous chapter, indiscriminate violence was used for so long (and still is) in an ambiguous way that I am convinced it is not possible to simply impose a new meaning on the term without the risk of serious confusion. The completely opposing definitions of Steele and Kalyvas who define indiscriminate violence as random violence and the targeting of groups, respectively, are a case in point. In addition, the term random violence seems to suggest to many readers the intentionally administered randomness in a statistical sense which makes it difficult to separate intentions from perceptions. As a consequence, I opted for the less specific term arbitrary violence. For the category of group-selective violence at least the term “collective targeting” stands out as an alternative. My intention was to choose a term that focuses on the similarities between violence selecting on individuals and violence selecting on groups, which are both targeting strategies. Collective targeting seems to suggest to many readers a meaning closer to arbitrary violence and thus risks producing a similar misunderstanding as the term indiscriminate violence. Coercion and Violence Against Civilians As outlined in the previous two chapters, several authors are convinced that government and nongovernment indiscriminate violence can be tools for

Conceptual Building Blocks

69

coercion, but the literature does not often discuss the necessary conditions for coercion to work. This problem is more accentuated in the study of government violence than in the study of nongovernment violence. Particularly, research on terrorism provides key insights that I will be able to build on (see e.g., Abrahms, 2013; Pape, 2003). In both literatures researchers often do not go into the details of the mechanisms of coercion. Therefore, before proceeding with a theory of group-selective violence, it seems necessary to take a closer look at the concept of coercion. Coercion as a Concept

A very useful definition of the concept of coercion can be found among others in an article from Byman and Waxman: “[c]oercion is the use of threatened force, including the limited use of actual force to back up the threat, to induce an adversary to behave differently than it otherwise would” (Byman & Waxman, 2000, 9).4 This definition clarifies that coercion is the sending of a message that consists of a threat to use force and a demand for a certain behavior. For the sake of clarity, I will call this the coercive message. Therefore, coercion must be distinguished from what Schelling calls brute force (Schelling, 1966, 2). Brute force is the use of physical power to reach an aim. For instance, if a state wants a piece of land from another state it could use pure military strength to invade and occupy that land. In this case, there is no necessity for bargaining or compliance by the targeted state. Coercion, however, is the threat of violence that seeks to circumvent the actual use of violence if the target complies with the demand. It is a form of communication in that one actor, the sender, is sending a message to another actor, the receiver. For instance, if a state wants a piece of land from another state it could issue a threat to use force if the other state does not hand over the land without a fight. The message itself has two main components, a demand and a threat. It is demanding a certain action (or inaction) and threatening a punishment in the case of noncompliance. Should the threat be successful and the demand be effectively communicated, the expansionist state will obtain a territory without having used military force at all. Schelling calls this the use of latent violence (Schelling, 1966, 3). As the definition from Byman and Waxman makes clear, coercion can involve the actual use of violence. For the sake of clarity, I will call this type of violence coercive violence. Yet, the purpose of violence here is still very different from brute force. The aim is not destruction but again communication. This means that the employment of coercive violence then can have two aspects: communicate the coercive message (threat and demand) and establish credibility of the threat. On the one hand, violence can incorporate the message itself. In many instances and especially where it is not possible to communicate the

70

What Do We Know About Indiscriminate Violence?

coercive message directly, the usage of violence transports the threat and the demand (Schelling, 1966, 141–151). As Schelling writes on limited wars: “In fact, there is probably no characteristic of limited wars more striking than this, that one communicates by deeds rather than words, or by deeds in addition to words, and makes the actions form a pattern of communication in spite of the fact that each side is literate enough to understand what the other is saying” (Schelling, 1966, 146). In this case, the coercive message is transported by coercive violence. For instance, when Israel sought to prevent the Fedayeen attacks from across the Jordanian border in the 1950s, it retaliated with aerial attacks after each incident on Jordanian territory. In this way the bombing raids by Israel in themselves transported the message which behavior was to be stopped and what the consequences were for noncompliance (Byman, 2011). On the other hand, an actor can use violence to make the threat of even more violence credible. As Nieburg points out: Violence has two inextricable aspects: its actual use [...], or its potential use. [...] The “rational” goal of the threat of violence is an accommodation of interests, not the provocation of actual violence. Similarly, the “rational” goal of actual violence is demonstration of the will and capability of action, establishing a measure of the credibility of future threats, not the exhaustion of that capability in unlimited conflict (Nieburg, 1963, 44).

In this case, violence is meant to signal that the actor is capable and willing to use more violence. Much of the rational choice literature on terrorism recognizes this aspect as “the sending of a costly signal” (see e.g., Kydd & Walter, 2006; Lapan & Sandler, 1993; Overgaard, 1994; Pape, 2003). So coercive violence is used either to increase the credibility of the threat of future costs or to also send the coercive message altogether. As I will discuss later in this chapter, both of these aspects are important to understand the effectiveness of violence as a coercive threat. However, the fact that coercive violence is supposed to carry the coercive message is especially important in the study of indiscriminate violence. Most perpetrators of indiscriminate violence cannot acknowledge that such violence is intentional since the use of it is seen as morally unacceptable and even prohibited under international law. As a consequence, a perpetrator willing to use indiscriminate violence against civilians is unlikely to communicate this directly. Doing so would preclude the option to deny responsibility by e.g., attributing violence to collateral damage. Take for instance one recent example form Myanmar. Since the end of 2016 NGOs and international organizations have highlighted that the army of Myanmar has been conducting an operation of ethnic cleansing against the Muslim minority of the Rohingya (Holmes, 2016). While many details remain opaque, a largescale army operation against the Muslim population is evident. However,

Conceptual Building Blocks

71

contrary to all evidence, the government of Myanmar strongly denies that such a campaign is under way (Holmes, 2017). As a consequence, in studying the coercive effects of indiscriminate violence, we are interested in understanding the circumstances under which coercive violence can be effective in communicating the coercive message, and subsequently, in changing the behavior of the target (see Figure 4.2). Targets of Violence and Targets of Communication

So far, the discussion has identified neither the target of coercive violence nor the target of the coercive message. In most studies on indiscriminate violence against civilians, the implicit assumption seems to be that both are the same group. In the literature on government indiscriminate violence not a single author distinguishes between the target of violence and the target of coercion. However, as the literature on terrorism emphasizes, this does not have to be the case. Terrorist violence is often seen as symbolic or seeking a demonstration effect (e.g., Schmid, 2011, Thornton, 2006, 45–47).

Figure 4.2 Coercion by Violence

72

What Do We Know About Indiscriminate Violence?

For this reason, I introduce a distinction of two types of coercion that is prominent in criminology, which will help to build a theory on the effects of group-selective violence in the next chapter. In criminology, an important aspect concerns the effectiveness of measures of criminal deterrence, which are the measures that can deter criminals or potential criminals from committing a crime (e.g., a prison sentence). Here, two types of criminal deterrence are distinguished: specific and general deterrence (e.g., Gibbs, 1975, 32–39, Stafford & Warr, 1993).5 Specific deterrence signifies instances in which an actual offender is sanctioned and this sanction deters him from future offenses. For instance, a jail sentence for a burglary steers the person not to commit another similar crime after he leaves jail. By contrast, general deterrence means that the sanctioning of one offender deters other potential offenders who may not have yet committed any crime. General deterrence provides the potential offender with an indirect experience of punishment. For instance, if someone observes a burglar being sentenced to time in jail this may reduce his probability of becoming a burglar himself (Stafford & Warr, 1993, 123–124). The distinction between specific and general deterrence points to the possibility that the intended coercive threat may not exclusively aim at the target of punishment but sometimes focus on other potential offenders. The target of violence may just be one example of some misbehavior that merits exemplary punishment. However, the perpetrator may derive his main utility from providing the indirect experience of punishment to others and thereby demonstrating which actions will be sanctioned. This principle is of course nothing new: Voltaire once famously observed that from time to time the English would shoot an underperforming admiral “to encourage the others” (Voltaire, 1988, 111). However, in the study of indiscriminate or better group-selective violence it seems to have been ignored so far. A Conceptualization of the Effects of Violence Against Civilians Before I go on to theorize the effects of group-selective violence, it is necessary to provide a conceptualization of the effects of violence against civilians in general and to offer a tool for assessing its effectiveness. As both the literature on government and on nongovernment violence highlight the fact that violence can have positive and negative effects at the same time, this framework needs to incorporate all effects in order to assess the potential net effect of violence. As briefly mentioned before, it is possible to disaggregate the effects of violence into three different components: coercion, provocation and elimination 6 (for an illustration, see Figure 4.3). Violence may have a

Conceptual Building Blocks

73

Figure 4.3 The Effects of Violence Against Civilians from the Perspective of the Perpetrator

coercive effect. Several authors have pointed out that even indiscriminate violence against civilians can have a coercive effect in that it forces civilians to support the perpetrator. Violence against civilians often has provocative effects. Previous research argued that indiscriminate violence creates feelings for revenge, delegitimizes the perpetrator and increases the perceived insecurity among the population. Lastly, at its core, lethal violence is eliminatory. Although this may not be its main purpose, lethal violence, by definition, kills. In order to investigate the effects of violence against civilians in the next chapter, I will follow these three broad components. Importantly, it will be necessary to evaluate whether group-selective violence is able to produce a positive net effect (see Figure 4.3). Since only the elimination and the coercion effect work in favor of the perpetrator, but provocation mostly works in the opposite, negative direction, it is a priori unclear if the overall effect is positive or negative when all three effects are present. Only by discussing the likely intensity of all effects and off-setting the positive against the negative effects, one can judge whether violence is able to produce an overall positive outcome for the perpetrator. In addition, it is necessary to provide a tool that enables us to assess the occurrence of the particular effects of group-selective violence. As a consequence I will discuss in detail the conditions under which provocative and coercive effects should materialize.

74

What Do We Know About Indiscriminate Violence?

Coercive Effects

The previous sections clarified the general logic of coercion, but, so far, the criteria for the effectiveness of the coercive message to change the receiver’s behavior have not been explored. Unfortunately, the literature on coercion does not provide a clear-cut framework for its effectiveness. However, within the International Relations literature the following six aspects in particular received attention7: comprehensibility, credibility to use violence, credibility to withhold violence, feasibility, desirability and interaction.8 Comprehensibility. To be effective, the coercive message must be compre-

hensible.9 Signals must stand out from background “noise.” As already mentioned, when indiscriminate violence is used it may not be possible for the perpetrator to communicate a message verbally and directly. Since killing civilians is in most circumstances seen as morally reprehensible and difficult to justify, the perpetrator may not want to be explicit about his intentions. Not making the threat explicit in a statement has the distinct advantage that attacks on civilians can be labeled as unintended collateral damage later. Therefore, when targeting civilians and especially groups of civilians, the coercive message may need to be transported by coercive violence. Within this message the threat is always to use additional violence in the future and the demand is to prohibit a behavior that is punished. For instance, in the case of Baranivka the threat is to shoot more people in the future and the demand is to hand over the people who were responsible for the attack. However, in order to send such a message it is necessary that a clear actionreaction connection exists.10 I expect that two aspects in particular are important in making that connection: timing and location.11 As an example, take for instance the US cruise missile attack on a Syrian airfield after the Syrian government used chemical weapons in April 2017. Here the US military attacked the airfield from which the Syrian plane had taken off to carry out the chemical attack in order to send a “specific message” (Gordon, 2017). As this US attack demonstrates, the United States acted immediately after the chemical attack and targeted a specific location: the airfield where the plane that carries the chemical weapons started. In that way, the US was able to make a strong connection between the unwanted behavior and the punishment via timing and location. However, in some cases the targeting may have to be repeated in order to become understandable as a targeting rule. Because it may not possible to infer a definitive pattern from one incident and difficult to infer from a few, repetition may be needed to make the pattern perceivable. Credibility to use and to withhold violence. To be effective, the threat of

the coercive message must be credible. In other words, the receiver of the threat must believe that the sender is not bluffing when he threatens more

Conceptual Building Blocks

75

violence in the future. Since most of the time the threatened punishment is costly to the sender of the threat, it may seem questionable whether he will follow through with his promise. For example, had the US threatened to attack the Syrian airbases only after the next chemical attack and had not attacked directly, it would have been doubtful if the US would do so in the future. Potentially, the US would be perceived as unwilling to attack. However, with group-selective violence against civilians, the credibility to use violence is mostly directly visible. Since coercive violence is used to communicate, the credibility to use violence is already established by the exemplary usage of violence. It is, therefore, not the case that a perpetrator issues a verbal threat that can be perceived as a bluff. To the contrary, employing coercive violence establishes credibility. The credibility to withhold violence is more difficult to establish but remains closely connected to its intelligibility. Here the coercer must demonstrate that he will use violence only in response to an unwanted behavior and otherwise will refrain from using violence. Therefore, the level of restraint that the coercer demonstrates in his targeting is important. If the issuer of a threat gives the impression that he will punish the target anyway and irrespective of its compliance, the threat cannot coerce. For instance, had the US not stopped after one attack on the Syrian airfield, it would not have been a credible coercive punishment. In this case the Syrian government could have drawn the conclusion that the US is punishing them independently of their usage of chemical weapons. Therefore, it seems that the credibility to withhold violence can be demonstrated by a restrained usage of violence either temporally and or locally. The credibility to withhold violence is therefore closely connected to the criterion of basic comprehensibility. In terms of violence against civilians, a coercer may show restraint by targeting civilians exclusively along one targeting rule which punishes a certain behavior only once. This means that in the specific case where coercive violence is supposed to transport the coercive message, the credibility to withhold violence can be established with the targeting pattern.

Feasibility. To be effective, the coercive message needs to demand something that is feasible. Those who are threatened by the coercer will need the ability to comply with the demands. If the target of the threat has no control over the undesired behavior the threat will be in vain. For instance, in the case of the bombing of the Syrian airfield the implicit message was that the United States would punish the usage of chemical weapons. Since the Syrian regime controls the usage of these weapons it is capable of not using them. However, had the US asked the Syrian government, for instance, to disarm all progovernment militias, the feasibility would have been less straightforward.

76

What Do We Know About Indiscriminate Violence?

In the case of group-selective violence, Chapter 7 will show that many types of demands can be issued by the perpetrator of violence and not all are equally feasible. However, in most cases the perpetrator demands actions that the population is able to comply with. For instance, as I will outline later, the indiscriminate killing in Baranivka led multiple villages in the vicinity to hand over partisans and collaborators. For the theory building exercise I will, therefore, assume that the demands are feasible. Desirability. A coercive message must structure the incentives of the target

in such a way that compliance is the most desirable option for the target. This is the point of Berrebi and Klor (2006) and of Pape (2005), who argue that terrorism can only coerce if the potential future costs of the threat outweigh the cost of the demanded action. Of all the options available to the receiver of the message, the action desired by the sender must be the most appealing one. This implies that the potential punishment and the probability that the sanction will materialize must be sufficiently high to make the target give up something it values. In addition, the desirability depends on the target’s preferences. If the target values the opposite action high enough it may not give in even in the face of the most severe punishment. For example, in the example of the US missile attacks on Syria, for the Syrian government it is important to weigh the threat of additional missile attacks against the military benefit it derives from using chemical gas. Only if the first one outweighs the latter one, can the Syrian government be coerced. In the case of group-selective violence, this means that the threat of future violence must outweigh the costs of the demand. If the perpetrator of violence is strong enough and can threaten to use massive violence with a high probability, compliance will be more likely. In addition, if the population is not sternly opposed to the demands the possibility of compliance rises, too. For the purpose of theory building I will also assume that the preferences of the targets allow violence to become coercive. As the examples in Chapter 7 will show, this is often the case with government violence. In addition, I will discuss for all types of violence how the variation of the amount of violence impacts on the coercive effectiveness. Interaction. Finally, the effectiveness of a coercive threat depends on the pos-

sible interaction. If the receiver of a coercive message has the opportunity to countercoerce or to shield itself against future coercion, coercion will be less effective. On the one hand the receiver can react to the coercive threat in kind. If the receiver can take action to retaliate it may make coercion costlier for the sender than he is willing to accept. As a hypothetical example, if Syria were able to sink an American aircraft carrier for every missile attack on an airfield, the US would be less likely to attack and Syria would be less likely to comply. On the other hand, the receiver may also invest in technology that makes the

Conceptual Building Blocks

77

threat irrelevant. For instance, if Syria were able to quickly build a missile defense system, the US threat would be neutralized. As a consequence, the capabilities of the target to react play a role in the effectiveness of the threat. As a consequence, interaction between the sender and the receiver of a coercive message is an important aspect for the effectiveness of coercion that needs to be taken into account in practice. However, for now I will exclude this aspect for purely practical reasons. The main purpose of this book is to show that group-selective violence can be effective under given circumstances. Therefore, I will assume that the receiving side of the threat cannot retaliate or adapt. As I will show in Chapter 7 this is the case in many examples. It is important to keep in mind, though, that in other cases where the receiver has additional options, coercion may fail. All six criteria have an important impact on the effectiveness of an act of coercion and it is important to understand how they relate to the question: how can “indiscriminate violence” be effective? Here two steps have to be distinguished. First, as a necessary condition, coercive violence needs to effectively transport the coercive message. This means violence needs to be comprehensible. Second, if the message is understandable the coercive message must be able to shape the behavior of the target. This means that the threat of violence must be credible, the demanded action must be feasible and desirable and potential countermeasures must not change the calculation for the target. As a consequence, for building the theory in the next section I will focus only on the comprehensibility. Comprehensibility is the crucial aspect here. Although in practice all other aspects remain important, none of them will make a systematic difference to different types of targeting. It is the comprehensibility that authors have been concerned with (e.g., Kalyvas, 2006), and I will concentrate on this issue in order to show how different types of violence against civilians (selective, arbitrary, as well as specific and general groupselective violence) differ in their potential to make themselves understandable. Provocative Effects

As outlined in Chapter 2, three mechanisms can be identified that link violence to provocative effects: legitimacy, revenge and security. The legitimacy mechanism proposes that the usage of indiscriminate violence against civilians delegitimizes the perpetrator and thus leads to increased resistance. However, the argument does not specify what aspect of indiscriminate violence supposedly makes it so appalling for the population. Is it the massiveness of the violence or is it that innocents are being killed? With regard to the amount of violence, indiscriminate violence may not be very different from selective violence. Kalyvas points out that selective violence is often massive in its amount and difficult to distinguish from indiscriminate violence (Kalyvas, 2006).

78

What Do We Know About Indiscriminate Violence?

However, from surveys in war-torn countries it can be seen that at least the targeting of innocent civilians is widely perceived as morally wrong.12 For instance, the International Committee of the Red Cross finds in 1999 that 78% of people asked in war-torn countries and 59% in other countries were opposed to the view that the killing of civilians could be justified (ICRC, 2016, 7). In addition, the PEW research center found in July 2017 in a survey among US citizens that 59% of the population think that targeting civilians can never be justified, and 83% think that it can rarely (e.g., in cases of self-defense) be justified (Center, 2017, 101–103). Therefore, in the following, I will tentatively assume that the delegitimizing effect of indiscriminate violence comes from targeting people who are not individually responsible for a certain behavior. Since this effect may not apply under all circumstances, I will also express potential reservations with this view over the course of the next chapter. The revenge mechanism instead argues that violence against civilians will lead to feelings of revenge among those close to the victim. On the one hand, as outlined above, this mechanism seems to be confined to a small group of people like family members and friends. Unless norms like blood revenge are present it seems that only a limited amount of people will be touched by this. On the other hand, it can be expected that the revenge mechanism is always present whenever people are killed. While different targeting strategies may have an influence on the size of the effect, no type of targeting should be able to completely suppress this effect. Finally, the security mechanism hypothesizes that indiscriminate violence increases the costs and the insecurity of staying neutral and thus leads to support for the opponent. As I have pointed out, this should be the case only when violence is perceived as arbitrary. In cases where there is a targeting rule that for instance targets a group, only members of this group should feel an increase in insecurity. As a consequence, the security mechanism is closely connected to the comprehensibility criterion of coercion. When violence follows a clear pattern that is understandable the security mechanism will be triggered only for those who fall under the targeting rule. However, when violence becomes less understandable and is perceived as more arbitrary, more people will see neutrality as too risky. Yet, unlike the revenge mechanism, the security mechanism also depends on the amount of violence. Only when violence becomes widespread does the risk of neutrality surpass the risk of joining an armed group. Eliminative Effects

Finally, at its core, lethal violence is eliminatory. Although this may not be its main purpose, lethal violence by definition kills. As a consequence, all types of violence against civilians will have an elimination effect but, as I

Conceptual Building Blocks

79

will argue below, the target of violence differs between types of violence. While selective violence intends to kill the guilty, group-selective violence intends to kill the innocent. This means that those types of violence that target the innocent do not have an eliminatory effect that is positive for the perpetrator. As they only target the innocent, they do not reduce the threat posed by the opposing organization. Summary of the Conceptualization

This subsection sought to bring together the criteria that are necessary for evaluating the effects of violence against civilians in one conceptualization. Table 4.1 summarizes these criteria. All types of violence against civilians (selective violence, group-selective violence and arbitrary violence) will have positive and negative effects to some extent. In the following Chapter 1 important step will be, therefore, to theorize if these types of violence show systematic differences in the way they generate these effects. Building a theory of the effects of violence along these criteria will demonstrate that groupselective violence cannot produce the same effects as arbitrary or selective violence and that the respective effects of different types of group-selective violence differ starkly. This assessment of the effects will also allow me to derive hypotheses about the direction and the size of a possible net effect. Conclusion Over the course of the previous sections several points were elaborated, and I will summarize them here briefly. First, in this chapter a new categorization

Table 4.1 Criteria for the Effects of Violence Against Civilians for the Perpetrator

Coercion effect (positive) Comprehensibility Provocation effect (negative) Legitimacy Revenge Security Elimination (positive) Targets guilty Targets innocent Net effect for the perpetrator

Range of the Effect of a Specific Type of Violence

weak to strong weak to strong weak to strong weak to strong weak to strong weak to strong weak to strong yes or no yes or no negative to positive

80

What Do We Know About Indiscriminate Violence?

of violence against civilians was introduced that concentrates on the effects of violence. As I pointed out, previous categorizations seem to have focused on intentions. However, this approach obscures the possibility that violence may be perceived as arbitrary because of erroneous implementation or because of a wrong perception. In addition these categorizations bear the risk of confusing group-selective and arbitrary violence. By focusing on the implementation and perceptions of violence it is possible to conceptualize arbitrary violence not as an intentional approach but as a problem complicating strategies of selective and group-selective violence. The ideal type of completely arbitrary violence is then the end-point of two continua, when either of the two targeting strategies becomes, or is perceived as, increasingly arbitrary: see Figure 4.1. Secondly, this chapter also elaborated on the concept of coercion. Importantly, it is necessary to ascertain that coercion is exercised by the threat of violence and not by violence per se. If violence is used as coercive violence it has two aspects: first, it is supposed to make a threat credible and, secondly, violence can be used to convey the coercive message when verbal communication is not possible. Regarding the effects of coercive violence the main question thus is: how can coercive violence make itself comprehensible? In addition, I distinguished two coercive strategies relevant for group-selective violence: specific and general deterrence. Specific deterrence is punishing one person in order to make this person change his or her behavior, general deterrence targets one person in order to shape the behavior of others. Thirdly, since previous approaches did not focus on the effects of violence, it was necessary to introduce a new framework of these effects. This framework divided the effects of violence into three types, coercive, provocative and eliminatory, and elaborated the circumstances under which these effects should materialize. However, the effects of violence against civilians work in opposite directions. While the eliminatory and the coercive effects work in favor of the perpetrator, the provocative effect most often works against his intentions. The net effect of violence against civilians is determined by off-setting positive against negative effects. Notes 1. I will speak of arbitrary violence rather than random violence in order to avoid confusion with a statistical understanding of randomness. 2. Relatedly, Steele and Wood seem not to see collective targeting as a means of coercing civilians into a certain behavior but more as a means of brute force and displacement (Steele, 2009). 3. For instance, Kocher, Pepinsky, and Kalyvas also see this possibility, but do not elaborate on it (Kocher et al., 2011, 204).

Conceptual Building Blocks

81

4. This definition is indeed very close to what is used in research on terrorism. For instance, Sandler, Tschirhart, and Cauley (1983) define terrorism as “premeditated, threatened or actual use of force or violence to attain a political goal through fear, coercion, or intimidation” (Sandler et al., 1983, 37). 5. While similar concepts are sometimes used in International Relations they vary slightly in their conceptualization, which makes them less useful here (see e.g., Huth & Russett, 1993; Morgan, 2003). In International Relations the terminology is also slightly different, mostly distinguishing between “general” and “immediate” deterrence. Here, immediate deterrence assumes an ongoing crisis between two states and focuses on the deterrence of military action within that crisis. By contrast, general deterrence does not assume a crisis but focuses on the deterrence of challengers in general. Most of the literature on deterrence in IR focuses on immediate deterrence (Levy, 1988, 488–489). 6. I will use the term provocative effects rather than counterproductive effects here since effects may sometimes work in the perpetrator’s favor. 7. In other fields, similar criteria exist to describe the effectiveness of coercion. For instance, in criminology, commonly, three aspects are considered: the certainty, the celerity and the severity of punishment (Gibbs, 1975). To an extent all three aspects are considered in the six criteria listed in the IR literature. Empirically, the literature on criminal behavior seems to have found more support for certainty and celerity than for the severity of punishment (Nagin, 2013; Nagin, Cullen, & Jonson, 2009). Interestingly, some findings seem to be similar to those of the literature of indiscriminate violence. Seemingly, criminal punishment at the individual level can have counterproductive effects as well (Pogarsky & Piquero, 2003). 8. The first five criteria can be found in (Schelling, 1966). Other accounts of coercion emphasize also interaction. For instance, Byman and Waxman (2000) point out that the capability of the receiver to counteract the threat has an important influence on the potential effectiveness (Byman & Waxman, 2000, 10). Pape makes a similar argument for both terrorism and strategic bombing (Pape, 1992, 1996, 2003, 2005). Also, in the field of violence against civilians, Kalyvas and Kocher point to the importance of interaction (Kalyvas & Kocher, 2007b). 9. Leites and Wolf (1970) also point to understandability as an important criterion of selective violence (Leites & Wolf, 1970, 108). 10. This point is also closely related to the notion of retribution put forth by Souleimanov and Siroky (2016). 11. Schelling hints at the possibility that many aspects play a role in linking action and reaction. Among others he mentions target choice, appropriateness and timing in an example from the Vietnam War (Schelling 1966, 145). He then abstractly argues that the response has to remain in “the same idiom” (Schelling 1966, 148). So it may well be that there are further nuances possible that can make a punishment more understandable. 12. However, the notion of innocence and guilt is difficult to fathom. For instance, it remains an open question if guilt can only be ascribed through individual action. In some contexts, it may be the case that guilt by association is a socially accepted way to think. For instance, it is not uncommon that people accept responsibility for the behavior of other members of a group, a phenomenon known to social psychologists as collective guilt (see e.g., Doosje, Branscombe, Spears, & Manstead, 1998).

5 A Theory of Group-Selective Violence

In this chapter I will outline a theory of group-selective violence. The theory will highlight that group-selective violence is particularly focused on achieving coercive effects. This is not to deny that attacking groups can have other aims like displacing or even eliminating a group; however, as I will outline later, coercing groups is a strategy that seems to be frequently used but is often not understood. The Coercive Logic of Group-Selective Violence King, Keohane and Verba argue that theories should be designed to be as encompassing as possible (G. King, Keohane, & Verba, 1994, 113–114). Since I do not have a priori reasons to believe that the theory I am developing only applies to some groups, I will follow their advice and state my theory as applying to all kinds of groups and for the purpose of this study, give only a minimal definition. Here groups will be defined as two or more people that are perceived by outsiders and by members of the groups as belonging together. Therefore, groups in this sense could be families, hamlets, villages, clans, ethnic groups or whole countries. Group-selective violence then is violence against whole groups in order to punish some unwanted behavior. In the terms of Souleimanov and Siroky (2016) it is retributive. Typically, this unwanted behavior stems from a small number of people within the group; however, the perpetrator of violence is either unable or unwilling to identify those individuals. For instance, think again of the well-documented case of Baranivka that was briefly outlined in Chapter 1. Here, in 1941 a detachment of the Wehrmacht carried out a punitive

83

84

What Do We Know About Indiscriminate Violence?

attack on the village, killing ten people. The attack was a response to a partisan attack several days before during which three German soldiers were killed. Since the Wehrmacht officer in charge was unable to determine who among the villagers was responsible for the attack, random people from the village were executed (T. Anderson, 1999). The rationale behind such a punishment is to coerce entire groups into a different behavior. The coercive threat of future punishment is supposed to give groups an incentive to collaborate with the perpetrator. In the case of Baranivka, the officer wanted the inhabitants of the village to hand over those who were responsible for informing the partisans about the German patrol and punished the village when it did not comply. This way groupselective violence shifts the responsibility for policing the group away from the perpetrator of violence to the group itself. The implicit demand is that the group, as a whole, stops collaborating with the enemy and thus has to expel, or tame those who seek to do so. If some people within the groups do not comply, the perpetrator will punish those groups as a whole. Therefore, in group-selective violence, the punishment itself is not directed at the guilty individuals. As the perpetrator cannot distinguish between collaborators and innocents, he will arbitrarily pick some people from among the group. This means group-selective violence introduces the kind of “restricted randomness” into civilian targeting, which was already mentioned before (see Wilson, 2010, 220). Any member of the target group can serve as a target but the members of other groups will not be victimized. As a consequence, on the individual level, group-selective violence differs from selective violence in the type of targets focused on. In selective violence, the targets of coercive violence and the coercive message are those who collaborate or potentially collaborate with the enemy. Selective violence punishes, and threatens to punish, a certain behavior shown by individuals. As a consequence, people who might have undertaken this action will then be more likely to abstain, because the expected costs have risen. However, in group-selective violence the perpetrator of violence targets and coerces predominantly those who are uninvolved and had no intention of collaborating with the opponent. By killing those who are uninvolved, the coercer communicates that they will not make a difference between those who are guilty and those who are innocent as long as there are collaborators within the group. In fact, the perpetrator will assign guilt by association and perceive the whole group as guilty. The assumption of the perpetrator usually is that at least some of those within the group who are uninvolved have knowledge of those who collaborate. Not using this knowledge to support the perpetrator makes them guilty. This way the uninvolved individuals will be threatened with further violence unless the group as a whole changes its behavior. The implicit

A Theory of Group-Selective Violence

85

demand within the coercive message is that uninvolved or neutral people need to support the perpetrator in order to avoid violence; neutrality is not enough anymore. Therefore, in a way, perpetrators outsources his obligations to find and silence opponents to the uninvolved civilians. If they do not want to be targeted they must ensure that the group is acting in accordance with the demand of the coercer. By implication this also means that for group-selective violence to work it is neither necessary nor useful to target actual opponents. In fact, group-selective violence will work best when it targets those who are uninvolved. Thereby it increases the coercive threat for other uninvolved civilians and possibly changes their behavior from neutrality to support for the perpetrator. In the light of this discussion one key point of the theory of groupselective violence is to put the focus of research on the uninvolved or neutral part of the group. While many authors seem to assume that the targeting of opponents is the main aim of violence (see Chapter 2), this is not the case for group-selective violence. According to the logic outlined here, targeting the uninvolved may to some degree trigger provocative effects, but nevertheless, it is absolutely necessary for the intended coercive effect of group-selective violence. It is, therefore, important that we acknowledge the fact that group-selective violence intentionally targets the uninvolved in order to coerce others who remained uninvolved. In addition, an implicit assumption in most studies on indiscriminate violence is that violence against one group primarily has effects on this same group. For instance, many empirical studies are interested in the effects of violence on the target itself. Yet, the effects of violence may be more encompassing. After all, as the literature of terrorism emphasizes so strongly, violence can have demonstration effects beyond the targets. As a consequence, I will use the differentiation between general and specific deterrence from criminology to distinguish two types of group-selective targeting strategies. First, an actor can punish a group in order to produce effects within this specific group. For instance, a counterinsurgent could punish a village with the aim of changing the behavior of this village in the future. That means the coercer kills some villagers to show the rest of the village that continued cooperation with the rebels will lead to further violence against the village. In this case, even though different individuals are involved (those killed and those coerced are naturally not the same), the target of coercive violence and the target of the coercive message is the same group. I will refer to this type as specific group-selective violence. Second, an actor can punish a group in order to demonstrate something to additional, surrounding groups. In this case, violence against the first group is not intended to change the behavior of this specific group; the purpose of violence lies in the larger demonstration effect. For instance, a

86

What Do We Know About Indiscriminate Violence?

counterinsurgent could punish a village in order to communicate a message to the surrounding villages. Violence is then used to set an example for those who might potentially harbor rebels themselves. I will call this type general group-selective violence (for an illustration of the typology see Figure 5.1). However, it is important to note that the separation of specific and general group-selective violence is again ideal typical. In practice, both these types of group-selective violence are not necessarily separated. For instance, an actor who seeks to coerce the targeted group may seek to coerce other groups at the same time, or vice versa. In addition, actors may not be able to delineate groups in a clear-cut fashion as groups may overlap or groups or group boundaries may be unclear. However, in the following, these practical complications will not be discussed. Instead, I will focus on the ideal typical strategies and their respective effects, as this will enable us to understand the coercive potential of group-selective violence. In sum, the discussion of the logic and the effects of group selective violence has highlighted two aspects that have not received enough attention so far. On the one hand, group-selective violence targets the uninvolved, the innocent civilians. Perpetrators seek to change the behavior of those individuals that remained neutral, in order to outsource the reduction of an unwanted behavior by a group. On the other side, by using groupselective violence perpetrators may not only seek to coerce the targeted groups but also the surrounding ones. Since other groups may conceivably also engage in the unwanted behavior perpetrators may seek to deter those Figure 5.1 Types of Targeting

A Theory of Group-Selective Violence

87

who have not engaged in it yet. Thus, the theory highlights the need to study particularly those who remained uninvolved in order to understand the effects of group-selective violence, groups and individuals. Only by focusing on their behavior can we assess how group-selective violence may benefit the perpetrator. The Effects of Violence Moving forward, I will now discuss the effects of violence as I laid them out in the framework in Chapter 4. In order to calibrate our understanding, I will start with the effects of two types of violence that were already described by Kalyvas: selective and arbitrary violence.1 Subsequently, I will discuss both types of group-selective violence: specific and general group-selective violence. In each case, I will point out whether the criteria for coercive success are present, in how far the potential mechanisms for provocative effects are triggered, whether violence targets the innocent or the guilty (elimination effect) and whether this leads to a positive net effect. Selective Violence and Arbitrary Violence

I will begin with discussing the effects of selective and arbitrary violence. For both types of violence, Kalyvas already provides brief descriptions of their effects: Kalyvas (2006, 144–145) and Kalyvas (2006, 142–144). However, Kalyvas’ interest lies with the causes of violence and he is much less concerned with the effects. As a consequence, while he assumes the presence of a coercive effect and the absence of a provocative effect for selective violence, he does not elaborate on the criteria for their occurrence (see e.g., Kalyvas, 2006, 141–145). Therefore, it will make sense to discuss selective and arbitrary violence as well, in order to have a reference point for our understanding of the effects of group-selective violence. It is important to note again that I will discuss all these types of violence as ideal types (cf. Weber, 1978, 1–22). This means that selective violence is implemented and perceived as completely selective on the individual level. Only individuals who did engage in an unwanted behavior are punished. In addition, arbitrary violence means that violence is implemented and perceived as completely arbitrary. There is no perceivable targeting pattern whatsoever. Selective violence. Selective violence, as Kalyvas points out, is directed

against individuals that are from the perspective of the perpetrator guilty of a punishable crime, e.g., those who collaborate with the enemy. The perpetrator thus targets with punishment all those who are engaging in the

88

What Do We Know About Indiscriminate Violence?

unwanted behavior.2 As with all violence against civilians, selective violence will produce provocative, coercive and eliminatory effects (see Chapter 4). Selective violence has a coercive effect. As selective violence kills the guilty, it sends an implicit coercive message. This message includes the demand to refrain from collaborating with the enemy and the threat to kill those who do not comply. As a consequence, in the terms of criminology, selective violence is a form of general deterrence and not specific deterrence. Regarding the effectiveness of coercion, selective violence seems to be able to score high on the main criterion. As discussed, in order to be understandable, violence must form a pattern of action and reaction which can be signaled through two aspects: timing and location. The punishment must follow the unwanted behavior quickly and it must target a connected location, i.e., in this case the right person. In selective violence both aspects can be fulfilled and the signal to the surrounding population is that collaboration leads to punishment. One problem with all types of coercive violence is the question of whether the action-reaction connection is observable to the general population. While this is not always the case, many examples show that civilians often have good information on their fellow family members, townspeople, or tribesmen and can distinguish cases. In addition, information on violence, even in premobile communication times, seemed to have spread very quickly. For instance, in Algeria the term téléphone arabe gained prominence during the Algerian War of Independence (1956–1964) and described the quick informal travel of information even in rural areas (Leites and Wolf 1970, 107). This means that selective violence can be highly effective in coercing.3 Selective violence will also produce only a minimum of provocative effects. To begin with, selective violence does not target those who did not engage in the unwanted behavior, the uninvolved civilians. As outlined in the previous chapter I tentatively assume this is the critical factor in determining the legitimacy of violence. As a consequence, I expect selective violence to reduce legitimacy less than attacks on innocent people. Regarding the revenge mechanism, I noted that it is restricted to close relationships of those killed. However, it can be expected that even selective violence can have some provocative effects through the revenge mechanism. For every person killed, it is possible that family members or friends seek retribution. Therefore, even selective violence may produce some negative effect via revenge. Finally, selective violence does not increase insecurity for the general population. Since violence targets only those that are engaging in an unwanted behavior, others have nothing to fear. This means that in sum selective violence produces very few provocative effects. Selective violence has a strong eliminatory effect. Whenever an actor kills a guilty person this leads to the elimination of an actual opponent. In this sense, selective violence is in part brute force, as it kills those indi-

A Theory of Group-Selective Violence

89

viduals who are supporting the enemy and whose elimination helps the perpetrator. Thus, depending on the amount of selective violence used, it can be even effective in the sense of just eliminating those who are opposing the perpetrator. Taking all these effects together, it seems that selective violence should have a highly positive net effect for the perpetrator. On the one hand, it eliminates opponents and scores high on the criteria impacting on the effectiveness of coercion which leads to a relatively large positive effect. On the other hand, it has only minimal provocative effects with only the revenge mechanism being of relevance. As a consequence, the net effect of selective violence should work largely in favor of the perpetrator. This description of the effects of selective violence is in line with the brief discussion in Kalyvas (2006, 144–145).

Arbitrary violence. Arbitrary violence is in most aspects the opposite of selective violence. As outlined before, arbitrary violence is not a targeting pattern but is better described as the complete absence of a perceivable pattern. Thus, by definition, it kills people without regard to their previous behavior in a haphazard way and, as a consequence, arbitrary violence produces the opposite effects from selective violence. Arbitrary violence fares much worse than selective violence in terms of coercive effect. As arbitrary violence is not based on a connection between an unwanted behavior and punishment, it cannot send a coercive message. Thus, it completely lacks any comprehensibility. As arbitrary violence by definition does not follow any pattern, no regularity can be perceived by the civilian population. It makes no sense to them. In sum, this means that there is no coercive effect to be expected. Arbitrary violence is highly provocative. Through all three mechanisms it produces negative effects. The killing of predominantly innocent people will produce a great sense of outrage that delegitimizes the perpetrator much more than selective violence. In addition, arbitrary violence will also produce provocative effects through the revenge mechanism. However, these should be equally high as those of selective violence, assuming that the same number of people is killed. Regarding the security-based mechanism, arbitrary violence produces the highest possible provocative effect. Since there is no pattern at all, the population will have to assume that violence can hit them, too. Therefore, neutrality will become riskier and thus less desirable. Depending on the amount of violence used by the perpetrator this can lead to massive support for the opponent. Finally, arbitrary violence has a limited eliminatory effect. Arbitrary violence targets the guilty and the innocent alike and as Kalyvas and Kocher point out, most likely disproportionally targets the innocent (Kalyvas & Kocher, 2007b, 191). Therefore, although it kills, it does not often

90

What Do We Know About Indiscriminate Violence?

kill opponents and in extreme cases, it may kill only innocent people with no connection to an unwanted behavior. Therefore, the eliminatory effect of arbitrary violence is not of much help for the perpetrator. Only in cases where the perpetrator escalates violence to genocidal levels can he produce a positive net effect (compare this to the discussion on indiscriminate violence as brute force in 2). Arbitrary violence produces almost the opposite effects to selective violence (the effects of selective and arbitrary violence relative to each other are summed up in Table 5.1). On the one side, the provocative effects are high, while, on the other hand, the coercive and the eliminatory effects are low. As a consequence, the net effect of arbitrary violence should be highly negative for the perpetrator. Again, this discussion is in line with what Kalyvas briefly described as the effects of indiscriminate violence (Kalyvas, 2006, 142–144). Specific Group-Selective Violence

In specific group-selective violence, an actor is confronted with only one group that becomes associated with some unwanted behavior. The unwanted behavior could be, for instance, that people from among the group support a rebellion. Whenever the actor does not have enough information to identify those who are responsible for the behavior, he can seek to punish the group as a whole through the logic of synecdoche in which some are taken to represent all. As outlined above, the logic behind this kind of violence is to punish a group in order to make the same group behave in a different way. However, in the case of specific group-selective violence and in contrast to selective violence, the target of coercive violence and the target of the coercive message are the same group. For Table 5.1 The Effects of Selective and Arbitrary Violence for the Perpetrator Coercion effect (positive) Comprehensibility Provocation effect (negative) Legitimacy Revenge Security Elimination (positive) Targets guilty Targets innocent Net effect for the perpetrator

Selective Violence strong strong weak weak weak weak strong yes no positive

Arbitrary Violence weak weak strong strong strong strong weak yes yes negative

A Theory of Group-Selective Violence

91

instance, a village is attacked in order to make this village behave in a certain way. Therefore, the critical issue for the perpetrator is to communicate that the attack is meant as a deliberate punishment. In addition, when the actor uses lethal violence against civilians, he cannot punish the whole group equally, but has to target a small number of people arbitrarily in order to set an example. Recall the case of the reprisal carried out in Baranivka where random villagers were executed. The aim of this kind of violence is to increase the threat of being killed for each remaining individual of the group, in order to make everyone support the aims of the perpetrator in the future. However, violence has an element of arbitrariness on the group-level. As a consequence, the effects of specific group-selective violence are a mixture of those of selective violence and arbitrary violence with both negative and positive effects experienced within the same group. If a perpetrator is confronted with only one group, the coercive effect may be limited, as the comprehensibility of specific group-selective violence is difficult to generate. People may struggle to see the point or the pattern. In order to make coercive violence convey a coercive message, it is necessary to establish an action-reaction relationship. This relationship is supposed to signal that the perpetrator is singling out some of the group for punishment in response to an unwanted behavior. However, in the case of specific group-selective violence this connection can only be signaled through timing and not through location. Since there is only one group to target and since the perpetrator does not know which individuals from among the group is responsible for the unwanted behavior, he cannot target the right people or the right location. As a consequence, the comprehensibility is limited but not impossible. Targeting the group immediately after an incident can provide this connection. However, it also means that coercive violence has to stop quickly again. Since the only connection to the unwanted behavior is timing the punishing actor can only use coercive violence that is close in time to the punished behavior. Otherwise the punishment risks signaling arbitrariness and loses its coercive effect. Targeting a group with specific group-selective violence will produce provocative effects. To begin with, specific group-selective violence kills predominantly innocent people within the targeted group. Therefore, this type of violence decreases the legitimacy of the perpetrator. However, since the targeting still follows an action-reaction it seems reasonable to think that the decrease in legitimacy is lower than in completely arbitrary violence. In addition, violence will always lead to a certain amount of people seeking revenge. As for the two types of violence which were described in the previous sections, for each person killed others may seek revenge. Finally, specific group-selective violence is at a high risk of triggering the security mechanism as it is not selective within the group. The arbitrariness

92

What Do We Know About Indiscriminate Violence?

of the targeting within the group means that the threat of future attacks is directed against everyone. Since it may not be clear after one attack what exactly is the association that is punished and what kind of behavior will satisfy the perpetrator of violence, the perpetrator is confronted with a dilemma. If he uses too few attacks he may not be able to establish the pattern and the population may infer a high degree of arbitrariness. If the perpetrator uses too many attacks, people may infer that the perpetrator is not interested in a change of behavior but seeks to eliminate them. Yet, if the group starts to think that violence becomes eliminatory, people will join the opposition. As a result, the perpetrator that seeks to coerce in a situation with just one group will have to use violence in a very restrictive manner in order to not trigger a backlash. It seems to be this kind of situation that some authors have in mind when they speak of the thin line that an actor has to walk in order to be repressive and not provocative (see e.g., Bueno de Mesquita & Dickson, 2007, 22, Lake, 2002, 377). Finally, the eliminatory effect of specific group-selective violence is similar to arbitrary violence. Specific group-selective violence targets one group but is not selective at the intra-group-level. Therefore, we can assume that the elimination of opponents will be as limited as in arbitrary violence. In sum, it seems that the net effect of this type of violence remains inconclusive and it is difficult to say if the perpetrator is able to achieve his goal. Specific group-selective violence may be restricted in communicating a coercive intention since it can only use timing and not location as an indicator of the action-reaction relationship. In addition, it seems that in this case the perpetrator has to steer a thin line between coercive effectiveness and provocation. If the perpetrator increases violence, both the coercive effects and the provocative effects will rise. However, it can be expected that the provocative effects will rise faster than those of coercion. To begin with, coercive violence can be expected to have diminishing marginal returns. This means the increase of coercive effects for a one unit increase in violence (e.g., one person killed) is higher for small amounts of violence than for high amounts of violence. Put differently, the first person killed will make a big impression while the killing of another person after already 100,000 people have been killed does not increase the coercive effect as much. As a consequence, a the coercive effects as a function of increasing amounts of violence could look like Figure 5.2, which is a typical function of diminishing returns. Here the slope of the function (i.e., the first derivative) decreases continuously for increasing amounts of violence although the absolute value increases continuously. On the other hand, provocation as a function of increasing amounts of violence will look differently. The provocation effects initially rise at a

A Theory of Group-Selective Violence

93

Figure 5.2 The Coercive Effects of Specific Group-Selective Violence

small rate since mostly the revenge mechanism and the limited legitimacy mechanism come into play, but the effect of provocation will rise at a much higher rate once the security mechanism is triggered. This means for higher amounts of violence where violence becomes perceived as more and more arbitrary, provocation should increase dramatically. As a consequence, provocation as a function of varying amounts of violence could look like Figure 5.3. Here the slope of the provocation function is continuously increasing. Since the function of the coercive effects has a continuously decreasing and the function provocative effects has a continuously increasing slope, summing negative and positive effects must lead to an inverted u-shaped function. This means there is one specific amount of violence when an additional increase of violence starts to lead to a decreasing net effect. In economic terms, this could be described as the point where the marginal returns of coercion equals the marginal costs of provocation. If the perpetrator surpasses this point, additional violence will have a decreasing or

94

What Do We Know About Indiscriminate Violence?

Figure 5.3 The Provocative Effects of Specific Group-Selective Violence

even negative net effect for him. For an illustration see Figure 5.4. In this graph, the hypothetical negative provocative effects are summed up with the hypothetical coercive effects. As can be seen from Figure 5.4, the combined net effects have a maximum at certain amount of violence. After this point, additional increases in violence decrease the net effect and the net effect reaches zero in the end. This result of the effects of specific group-selective violence is also close to descriptions of other authors. Johnston (2009) in particular can be mentioned here who argues that indiscriminate violence can coerce initially and may be counterproductive at high levels. In sum, it seems that specific group-selective violence is capable of achieving a coercive effect. However, the dilemma that the perpetrator is confronted with makes it very difficult to reach a high positive net effect for the perpetrator. The perpetrator certainly cannot escalate violence to extremes as this would lead to a decrease in the net effect. This is, however, only true if the perpetrator cannot annihilate the group. If annihilation or the killing of very substantial parts of the specific group is a possibility, the

A Theory of Group-Selective Violence

95

Figure 5.4 The Net Effects of Specific Group-Selective Violence

negative effects of violence may vanish again. This would then of course be an effect of brute force and not of coercion. General Group-Selective Violence

To illustrate the logic of general group-selective violence, imagine a situation in which an actor, A, is confronted with several groups, B, C and D. Assume now that one of these groups, B, becomes associated with a behavior unwanted by A (e.g., people from among the group B support a rebellion against A). While actor A identifies the responsible group, he has no information on the responsible individuals. As with selective violence, actor A can now single out the specific group and punish its members collectively, which only necessitates information on the groups responsible for the unwanted behavior. Here, the intention of actor A is to use exemplary violence in order to communicate to the other groups, in this case C and D. If a group is punished it sends the message to other groups that a similar

96

What Do We Know About Indiscriminate Violence?

behavior would meet additional punishment in the future. Should the coercive message be effective, the surrounding groups C and D will have an incentive to hand over those supporting the rebellion or to make sure that no one from among them collaborates in the future. However, in contrast to specific group-selective violence, the target of the coercive message and the target of coercive violence are different groups. While group B is targeted with coercive violence, the coercive message is destined for groups C and D. In the following I will discuss the coercive, provocative and eliminatory effects of general group-selective violence. However, the most important difference of general group-selective violence compared with other types of violence is that perceptions will differ widely for members of the punished group and members of the groups receiving the coercive message. In general, members of the punished group B will perceive this sort of violence as similar to arbitrary violence. For them it is impossible to adapt their behavior in order to avoid violence. However, members of those groups not targeted, C and D, will perceive this violence as similar to selective violence. They are able to adapt, although adapting is only possible as a group and not individually. As a consequence, the effects of general group-selective violence are split up between the targeted and the nontargeted groups. The targeted group. First, the targeted group will experience no coercive

effect. As with specific group-selective violence, the targeting at the intragroup-level is nonselective. However, now the perpetrator does not seek to communicate to the targeted group and does therefore not need to show restraint in attacking the group. In extreme cases, the perpetrator can even seek to eliminate this group entirely as this would send the strongest signal to other groups and reduce the potential provocative effects. As a consequence, the coercive effect on the targeted group is nonexistent. Second, the targeted group will experience a high provocative effect. The legitimacy mechanism may lead to an increase in opposition. Violence against one group where only a few are responsible for an unwanted behavior predominantly targets innocent people. Therefore, observers should be more likely to condemn the perpetrator compared with cases where the guilty were targeted (e.g., as in selective violence). However, there are two reasons why this is not necessarily the case. A) General group-selective violence already provides a high degree of selectivity. While violence does not select on the individual level it does use information to select on the group level. Therefore, the probability that those attacked are at least deemed partly responsible for the unwanted behavior is much higher than in completely arbitrary violence. B) It is possible individuals accept the concept of guilt by association. Take for instance the Baranivka incident where Anderson reports that even some relatives of the people killed blamed the incident on the partisans rather than the Germans (T. Anderson, 1999).

A Theory of Group-Selective Violence

97

The revenge based mechanism should also come into play. Since people from among the group are killed, those close to them have a higher probability of experiencing feelings of revenge and thus of joining the opposition to the perpetrator. However, the effect should be confined in that only close relations of the victims will experience emotions of revenge. Most importantly, the security based mechanism may be strong. Since violence does not distinguish between the guilty and the innocent within the targeted group, no one within this group can feel safe. No change of behavior can save the members of the targeted group from being potential victims. As a consequence, and as Kalyvas and Kocher (2007) suggest, the risk of staying neutral increases and may even be higher than the risk of joining the opposition. If the latter is the case, members of the group should have a high incentive to fight against the perpetrator. Third, the elimination effect of general group-selective violence is more precise than in arbitrary violence, as it is able to target a group that is associated with an unwanted behavior. The information of the perpetrator enables him to be selective on a group level, although not to the extent that individual opponents can be identified with any confidence. Thus, the probability that some of those who are killed are responsible for the unwanted behavior is higher.

The groups not targeted. The surrounding groups experience the situation differently. For the groups that are not targeted, group-selective violence produces a strong coercive effect. Regarding the comprehensibility, the coercer can establish a clear action-reaction relationship that satisfies both aspects, timing and location. Since the perpetrator can target a specific group that is associated with an unwanted behavior he signals which types of behavior will be punished and that abstaining from that behavior is sufficient not to be targeted. As a consequence those groups that are not targeted have a strong incentive to comply with the implicit demand of the perpetrator. Regarding the provocative effects, the impact on the groups that are not targeted should be limited. Solely the legitimacy based mechanism could lead to an important increase in opposition. If targeting the one group is seen as deeply unjust, other groups may fight the perpetrator. However, here too the previously mentioned objections can be raised: the legitimacy of targeting groups may be quite high when people accept the concept of guilt by association. The revenge mechanism will only play a role if the surrounding groups are connected to the targeted group. However, it is safe to say that social relations should be closer within groups than between them. This means that the revenge mechanism may be lower for surrounding groups, although not necessarily entirely absent. Finally, the security based mechanism should be absent. Since people from the surrounding groups can avoid punishment, they should not see neutrality as less secure.

98

What Do We Know About Indiscriminate Violence?

For the groups that are not targeted no elimination effect exists, since the perpetrator does not kill people from among these groups.

The combined effects of general group-selective violence. This means that general group-selective violence can partly circumvent the trade-off between provocative effects and coercive effects that is so problematic for specific group-selective violence. With the negative and the positive effects split up between groups, the perpetrator can escalate violence against one group to increase the coercive effect without risking a provocation effect among all groups. Since violence triggers the security mechanism in the targeted group in any case, escalating violence will not lead to a drastic increase of detrimental consequences for the perpetrator. Figures 5.5a through 5.5d illustrate this for the case of four groups. Only the first group is targeted and experiences provocative effects. Groups 2–4 experience only coercive effects. As a consequence, the net effect for general groupselective violence is different. This is illustrated in Figure 5.6. Since the coercive effects have multiplied compared with specific group-selective violence, the net effect stays positive for all amounts of violence in this example (although it decreases at some point). In addition, if this scenario is extended to cases where many more observing groups are involved,

Figure 5.5a The Provocative and Coercive Effects of General Group-Selective Violence, Group 1

99 Figure 5.5b The Provocative and Coercive Effects of General Group-Selective Violence, Group 2

Figure 5.5c The Provocative and Coercive Effects of General Group-Selective Violence, Group 3

100

What Do We Know About Indiscriminate Violence?

Figure 5.5d The Provocative and Coercive Effects of General Group-Selective Violence, Group 4

potentially no amount of escalation of violence against the first group will decrease the net effect. However, the perpetrator will still face a trade-off. The negative effects of attacking one group need to be offset by the coercion effects against the other. As a consequence, general group-selective violence produces the largest net effect when there is a large number of small groups. In this case, the negative effects are confined to one small entity and the coercion effects on the remaining groups are huge. The fewer and larger the groups are, the less dominant will be the positive effects and in extreme cases the negative effects might even outweigh the positive ones. Table 5.2 provides a comparison between the effects of both, specific and general group-selective violence. Finally, group-selective violence will always produce higher provocative effects and lower net effects than selective violence, even in the case of general group-selective violence. This raises the question, why is selective violence is not always used? The reason seems to be that selective violence is costly and requires certain capabilities. The overarching challenge of selective violence is that it needs an enormous amount of information in order to punish the guilty individuals (Kalyvas, 2006, 174–176). Gathering this information is costly and requires an amount of expertise

101 Figure 5.6 The Net Effect of General Group-Selective Violence

Table 5.2 The Effects of Both Types of Group-Selective Violence Against Civilians for the Perpetrator

Coercion effect (positive) Comprehensibility Provocation effect (negative) Legitimacy Revenge Security Elimination (positive) Targets guilty Targets innocent Net effect for the perpetrator

Selective Group-Selective Violence weak weak strong strong strong strong weak yes yes indeterminate

General Group-Selective Violence strong strong weak weak strong weak weak yes yes positive

102

What Do We Know About Indiscriminate Violence?

(infiltration of networks, etc.) that is not easy to build up in the short term. Consequently, and as observed by Kalyvas, coercers often struggle in the beginning of a campaign to obtain this information (Kalyvas, 2006). The cost of information gathering, thus, points to an important advantage of group-selective violence and especially general group-selective violence. Group-selective violence seeks to shift the costs of information gathering onto the civilian population. By associating the innocent civilians with the unwanted behavior, the coercer sets an incentive for the civilians to find the guilty themselves and to tame them. While this makes the net effect of violence drop, it also reduces the costs of information gathering drastically. As a consequence, it is possible that group-selective violence is even more efficient for an actor when we consider the cost of information gathering. While the effects of selective violence may be stronger, the cost of applying it may even be prohibitively high. Indirect Influences on the Effects of Violence So far, I have only discussed aspects of violence that are directly related to the coercive, provocative, and eliminative effects. However, some aspects may also be indirectly related to the different types of targeting. For instance, it is possible the one type of violence loses its intelligibility under specific circumstances, while others do not. In this section I will focus on two aspects that are of high importance, although this is surely not an exhaustive list: 1) the type of behavior that is being punished and 2) the introduction of arbitrariness. Both these aspects seem to pose problems particularly for specific group-selective violence. Type of Behavior to Be Changed

The type of behavior that the perpetrator of group-selective violence is seeking to change may make a difference. For instance, in the case of coercive violence against civilians an important difference is whether the coercer seeks to halt a continuous behavior or whether he wants to deter a discrete behavior from happening again. This distinction is associated with Schelling (1966), who introduced the difference between two types of coercion: deterrence and compellence. Schelling points out that deterrence and compellence differ in the way coercion is framed and that this has implications for the perception of the demand. Deterrence, on the one hand, is the defense of the status quo. When deterring a coercer seeks to persuade an opponent not to do something. The opponent than has the choice between complying (resting inactive) and defying (taking action). Compellence, on the other hand, is different; here

A Theory of Group-Selective Violence

103

the coercer seeks to change the status quo. In issuing the threat the coercer seeks to compel the opponent to do something. The opponent then has the choice between complying (taking action) and defying (resting inactive). To provide an example, imagine again two states interested in the same territory. In the deterrence scenario, the territory belongs to state A and state B prepares to invade. In that case, A can threaten to use massive retaliation should B invade and thereby try to deter B. In the coercion scenario, which is the mirror image of the first one, the territory belongs to B and A issues a threat to hand over the territory or they will attack. So, deterrence seeks to preserve the status quo and compellence seeks to change it. Recently, the argument was raised that compellence is more difficult to achieve than deterrence (Schaub, 2004). Schaub (2004) builds on findings from behavioral economics arguing that it is more difficult to make someone give something up than deterring him from getting it. Carried over to the study of coercion this supports the claim of Schelling that deterrence is easier than compellence. To illustrate this, think again of the US attack on the Syrian airfield which was a reaction to one chemical attack by a Syrian aircraft. In this case the US reacted to a single event (one attack) in order to deter another such chemical attack. The situation would have been different if the Syrian government had decided to start a bombing campaign with chemical weapons, dropping hundreds of bombs every day. In this case the Syrian behavior would have been approximately continuous. In general, it is possible to coerce an adversary in both situations. However, to coerce an adversary into stopping a continuous behavior makes it necessary to punish him repeatedly until he stops, while punishing one singular event can be done with one singular punishment (Schelling, 1966). This means that coercion in the face of a continuous behavior needs in general more repetition and, thus, higher amounts of punishment to be effective. In addition, it is not possible to use the timing of punishment as part of the communicative message. Since the punished behavior is ongoing, it is not possible to communicate by timely proximity that a certain behavior is unwanted. As a consequence, continuous behavior is more of a problem for specific group-selective violence than for general group-selective violence. As outlined above, specific group-selective violence needs to be highly restrained to remain effective because it cannot use the location indicator to make itself understandable. It may also trigger the security mechanism and then become detrimental to the perpetrator. However, this becomes next to impossible when continuous behavior is being punished. If the actor has to continuously punish the opponent, in specific group-selective violence this will exacerbate the provocative effects and ultimately squander any positive effect for the perpetrator. In addition, specific group-selective violence is unable to use the location as a communication device in its targeting.

104

What Do We Know About Indiscriminate Violence?

Timing is often the only option. Yet, if it is impossible to use timing because the behavior that is punished is continuous, specific coercion loses its only option to make itself understandable, i.e., to convey the coercive message. This heightens the risk that people will perceive violence as so arbitrary that they switch from being coerced to being provoked. As a result, for a strategy of specific group-selective violence, punishing continuous behavior will very likely not produce a positive net effect. Arbitrariness

So far, the three intentional strategies of violence against civilians (selective violence, specific and general group-selective violence) have been discussed as ideal types. Therefore, the strengths and weaknesses of these strategies were discussed without any arbitrariness in the implementation or perceptions (except for the arbitrariness of group-selective violence at the individual level). However, in practice this precision is of course not achievable, as errors may be made, or people may not recognize the targeting pattern for circumstantial reasons. It is, therefore, necessary to ask what happens if some error is introduced into the targeting pattern. Are strategies robust in producing a coercive effect? In general, the introduction of error and thus arbitrariness blurs the understandability of the coercive message in that the targeting pattern of timing and location is not as clear-cut anymore. For selective violence and general group-selective violence this should be less of a problem. Since they rely on both aspects of comprehensibility, timing and location, it may be possible to communicate the coercive message even if some moderate arbitrariness is introduced. However, specific group-selective violence can only rely on timing (the location aspect is already fully arbitrary). As mentioned, this means that specific group selective violence does not leave much room for mistakes. Therefore, the introduction of arbitrariness may tip the balance and cause violence to be perceived as untargeted. In sum, both aspects indicate that specific group-selective violence may not be able to coerce in many situations. In the face of continuous behavior or when arbitrariness is introduced its ability to communicate is even further reduced, which highly restricts the range of its applicability as an effective strategy. Underlying Assumptions All theories are reductions of the world and, consequentially, have to rest on certain assumptions. While this is normal it is important to make these assumptions as explicit as possible in order to know under which circum-

A Theory of Group-Selective Violence

105

stances the theory is applicable. Here I want to raise the awareness particularly of three assumptions made in my rational choice framework. First, the theory assumes that the perpetrator of violence can observe the behavior of groups. This means that the perpetrator can see if groups are complying or not. However, although in many cases it seems sensible to make this assumption, it is not necessarily the case. Should the perpetrator in fact be unable to observe group-behavior, the individuals within the group only have a limited incentive to comply with the coercive demands. Incomplete information by the perpetrator would thus introduce a moral hazard problem4 where individuals have an increased incentive to only pretend to comply (Akerlof, 1970; Spence, 1973). In this regard, Schubiger’s observation is especially interesting, that progovernment militia mobilization may be a way of signaling support to the government. Schubiger highlights that indiscriminate violence may lead to mobilizing civilians precisely because they seek to prevent an attack. Progovernment militia mobilization is therefore additional information provided by civilians in order to inform the government about their intentions, which ultimately reduces the probability of an attack. However, while Schubiger emphasizes bottom-up mobilization, mobilization may also be introduced as an expected behavior from above. For instance, Garrard-Burnett (2010) describes that Guatemalan people were forced into self-defense militias under the threat of severe punishment (Garrard-Burnett, 2010, 100–101). This strategy seems to have worked in terms of coercing the population (Garrard-Burnett, 2010, 101). A similar strategy was seemingly used by Turkish military in its fight against the PKK (Marcus, 2009, 221–222). Second, currently the theory assumes that information about an incident of violence spreads evenly among the target universe. While there is evidence that such information spreads quickly, even in the absence of communication technology, any application of a theory has to take into account that this may not always be the case. There is a reason that most perpetrators engage in active communication of their violence whenever they think that they can stomach the negative consequences. For instance the US dropped millions of leaflets in Vietnam, in order to explain why they were bombing specific locations (Hosmer, 1996). Also the German army put signs on destroyed villages specifying the reason for the punishment (Mazower, 2001). Therefore, the ability of a perpetrator to use regular communication in addition to the communication by the deed will have an impact on the effects that can be achieved. This will especially be the case for instances where the communication by the deed leaves room for ambiguity, as in specific group-selective violence. Third, the theory leaves civilians only a limited amount of choices. Effectively, civilians can either comply or not comply. However, in practice

106

What Do We Know About Indiscriminate Violence?

civilians may have additional repertoires of action that go beyond mere compliance.5 They may be able to negotiate with the perpetrator, leave the area or appeal to some of the other audiences, e.g., the international community. Consequentially, the logic of violence that is outlined in this study may only apply to cases where the perpetrator does not accept anything less than full compliance and where civilians have no room for maneuver. Hypotheses on Targeting

From the previous discussion it is possible to derive a number of testable hypotheses and I will outline the most important ones here. They will serve as the basis for the next chapters during which I will check the theoretical considerations against the empirical evidence. However, since the theory is potentially far-reaching, I will only be able to test some of the hypotheses. The remaining ones will be left for further research. I will begin with four hypothesis on the usage and the targeting of groupselective violence, followed by three hypotheses on the effects of the two types of group-targeting. Firstly, as discussed, group-selective violence is expected to be less effective than selective violence, although not necessarily less efficient, given the potentially high costs of gathering information in remote areas. Yet, in any case it is safe to say that group-selective violence will only be used in cases where resorting to selective violence is not a viable option. This could be the case because an actor has no information on the actual guilty individuals or he could be unable to reach these individuals. Whenever the usage of selective violence is not feasible, an actor has an increased probability of using group-selective violence (Hypothesis t1, [Ht1])

Secondly, the discussion above also suggests that actors engaged in group-selective violence will be selective on the group-level but indiscriminate or random in choosing targets on the within-group level. The main effect of group-selective violence comes from targeting the innocent as the coercive effect then forces the innocent civilians into supporting the perpetrator of violence. As a consequence, the actor will deliberately target the innocents. An actor engaging in group-selective violence will deliberately target the innocent civilians within a group (Hypothesis t2 [Ht2])

Thirdly, another important point of the theoretical discussion is that specific group-selective violence is less effective than general group-selective violence. This means that actors able to engage in the latter should do so.

A Theory of Group-Selective Violence

107

When actors are engaged in an environment with multiple groups they should be more likely to use general group-selective violence instead of specific group-selective violence (Hypothesis t3 [Ht3])

Fourthly, an actor engaged in general group-selective violence can have an incentive for exterminatory violence against the targeted group. Since group-selective violence has provocative effects in the targeted group and since harsher punishment increases the coercive effects, an actor may be tempted to be exterminatory in the application of general groupselective violence. An actor using general group-selective violence has an increased incentive to exterminate attacked groups (Hypothesis t4 [Ht4])

Hypotheses on Effects

First, the theory outlined here hypothesizes that group-selective violence will always produce provocative and coercive effects simultaneously. Even when a high coercive effect can be achieved, group-selective violence will produce some provocative effects through the legitimacy and the revenge mechanism. Group selective violence will produce provocative and coercive effects (Hypothesis e1 [He1])

Secondly, the provocative effects will manifest themselves mainly in those groups that are being targeted with group-selective violence. Provocative effects will materialize in the targeted groups (Hypothesis e2 [He2])

Thirdly, however, the location of the coercive effects will be found in different groups for specific and general group-selective violence. While in specific group-selective violence the effects will emerge in the targeted group, in general group-selective violence they will emerge in the surrounding groups. In specific group-selective violence the coercive effects will materialize in the targeted groups (Hypothesis e3 [He3]) In general group-selective violence the coercive effects will materialize in the surrounding groups (Hypothesis e4 [He4])

Fourthly, the coercive effect of group-selective violence will depend heavily on the understandability of violence. As outlined above this will be communicated with a clear action-reaction pattern and will potentially be supported with additional communication by the perpetrator.

108

What Do We Know About Indiscriminate Violence?

The perpetrator will try to communicate the strategy of grouptargeting to the target audience (Hypothesis e5 [He5])

Fifthly, the coercive effects may be outweighed by the provocative effects. If an actor fails to effectively communicate the strategy of group targeting, either by a lack of targeting or a lack of supporting communication, resistance will increase. If the perpetrator fails to effectively communicate the strategy of group-targeting to the target audience the provocative effects will outweigh the coercive effects (Hypothesis e6 [He6])

In the next chapters of the book I will proceed with increasing the plausibility of the theory and testing some of these hypotheses. The Effects of Group-Selective Violence on the Micro Level

The previous sections outlined the logic and the effects of group-selective violence on the aggregate group level, which obscures the underlying individual rationality. It is a common criticism of theories on the group level to only abstract groups while not considering the incentives for individuals. After all, groups are not living beings and group behavior critically depends on the action of individuals. Therefore, in this section I will elaborate on the mechanisms at play at the individual level. In a first step, I will take a pure rational-choice approach that exclusively focuses on individual decisionmaking (see e.g., Geddes, 2003, 175–211). This approach has the advantage that it only focuses on individual incentives and does not necessarily need specific group properties to demonstrate that individual rationality translates into group behavior. As a consequence, the result of this discussion should apply to all groups irrespective of their group-level traits. In a second step, I will discuss two specific group properties that may facilitate individual incentives to cooperate and with it the effectiveness of group-selective violence. This will give some indication as to which groups might be more responsive to coercive threats than others. The Individual Level

In order to understand the specific aspects of group selective violence it is again instructive to contrast it with selective violence. Selective and groupselective violence are dissimilar since group-selective violence is used to punish even if only very few from among the group do not comply. While

A Theory of Group-Selective Violence

109

selective violence sets clear individualized incentives, group-selective violence demands collective action. In selective violence, as already discussed, the perpetrator kills a guilty individual and thereby sends a coercive message to other potential opponents. The message demands that individuals refrain from opposing the perpetrator and threatens to kill them otherwise. As a consequence of this coercive message, someone thinking about collaborating with the enemy will have to factor the increased individual risk of being punished into his cost-benefit calculation. It is entirely up to his own calculation whether he complies or not and whether he incurs the risk of being targeted by violence. However, with group-selective violence people cannot avoid punishment individually, but must ensure a certain group behavior in order to escape being targeted. This means that for the individual only successful collective action can save them from being punished in the future. In this case it is necessary to turn here to the wider literatures on collective behavior as it is controversial under which circumstances collective action is likely or even possible at all. At least since the publication of Mancur Olson’s influential book The Logic of Collective Action, social scientists are wary of assuming that just because an action is in the interest of the group, the group will behave accordingly (Olson, 1965). Olson’s work demonstrated that individual rationality can run counter to the group’s interests and he focuses especially on the provision of public goods. Public goods are generally characterized by two properties: nonrivalry and nonexcludability. Nonrivalry means that after a public good is provided the consumption of the good by one individual does not reduce the size of the good for consumption by other individuals. Nonexcludability implies that after the provision of the public good it is not possible to exclude an individual from its consumption. An oftenused example is the provision of clean air or the reduction of air pollution. In this case, the consumption of one person does not reduce the amount of clean air for others and no one can be excluded from the consumption of clean air once it is provided. Olson’s argument is mainly that the provision of public goods can be inefficient since individuals have an incentive to free ride on the efforts of others. This means that the properties of a public good give every individual an incentive to not contribute to the public good in the hope that others will do so. The good would then be provided nonetheless and they can benefit from it without having paid for it. For instance, consider the following example. Each household has a wood-fired oven for heating, but the ovens belch smoke which pollutes the air. Every household can buy a filter that reduces the pollution from the wood-fired oven to zero. So if every household buys a filter the general air pollution can be reduced to zero, too. However, due to the large number of people involved the individual contribution,

110

What Do We Know About Indiscriminate Violence?

the filter, is marginal for each household, which means that buying a filter for one household alone does not change the general situation. Thus, without coordination no one will buy a filter since the individual costs are higher than the individual effect on air pollution. Even if all existing households had a filter and an additional household had to decide on whether to buy one, too, this household would have an incentive to not do so, as the individual pollution is also marginal (not reducing the public good significantly) while the cost of buying a filter is substantial. Since all households face the same incentives the outcome of such a situation is the underprovision of the public good, in this case clean air. For the study of group-selective violence this means that a coercive threat against a group may not translate into a collective effort to comply. People may have individual incentives that make them try to free ride on the efforts of others. Consider again the example of the German attack on Baranivka. In this case the German army punished a village because it suspected ties to the partisans among the population. Such punishment sends the demand to villages not to collaborate with the partisans and threatens to kill people arbitrarily from among the villages which do not comply. Individuals are threatened with a punishment if the group as a whole does not comply with the demands of the perpetrator. In this case the public good is to ensure that not a single individual can be linked with the partisans (and thus the group is not punished). The individual contributions to achieve this are inaction with regard to collaboration and actions to prevent potential collaborators from supporting the opponent. In the extreme case, noncompliance, however, is already reached when a single individual from among the group or the village collaborates with the partisans and is apprehended. An important corollary in Olson’s work is the proposition that larger groups will be more likely to provide inefficient amounts of the public good since the incentives for free riding rise (Olson, 1965, 36).6 Olson wrote: “unless the number of individuals in a group is quite small, or unless there is coercion or some other special device to make individuals act in their common interest, rational, self-interested individuals will not act to achieve their common or group interest” (Olson, 1965, 2). As a consequence, we would expect for our case of group-selective violence that collective action among threatened groups would be difficult to achieve, and that larger groups like ethnic groups are less likely to provide a public good than smaller groups like families. However, Olson’s work and with it the standard notion of the collective action problem includes several implicit assumptions that are not necessarily fulfilled. Importantly, in the case of group-selective violence the standard model provided by Olson seems not to be applicable. Over the past half century numerous researchers have outlined conditions under which Olson’s general claims do not hold (see e.g., Esteban &

A Theory of Group-Selective Violence

111

Ray, 2001; Pecorino, 2015; Pecorino & Temimi, 2008; Sandler, 2015). In the domain of group-selective violence there are various reasonable pathways to overcome the collective action problem. In the following I will outline some of them to clarify that even with a very limited set of assumptions it is possible to achieve a collective behavior in the face of group-selective violence. I will discuss mainly the effect of that violence on the previously uninvolved civilians but I will also briefly refer to those who are collaborating with the opponent. To begin with, it is important to outline the relevant variables in the specific case of group-selective violence. If we focus on the individual choice of an uninvolved group member who is under threat of being targeted by group-selective violence, this person will have to take several aspects into account: a) the size and the probability of the potential attack (both impact the probability of becoming a victim of violence), b) the value the individual places on his own life, c) the cost of providing information to the potential attacker, d) the expected effect that information provision has on the probability and the size of an attack and, most critically, e) the distribution of relevant information among group members of his own group. Going back to the example of Baranivka this would mean that prior to the killing of the ten villagers, each villager would have to calculate the expected cost of an attack and compare that with the expected cost of providing information to the potential attacker. For the moment I will assume that each villager has some information that would identify some partisan supporters in the village and thus marginally reduce the probability of an attack. I will also assume that the value of life is high and that providing information to the potential attacker is costly. The expected cost of the attack is specifically the expected cost of being killed in such an attack. The cost of providing information is for instance made up by the probability of retaliation by the partisans or the loss of reputation within the group. If the benefits on an individual contribution are less than the cost of this contribution each individual faces the classical collective action problem. All group members can hope that other people will provide relevant information and that they themselves can save themselves the cost of doing so. However, since all group-members think this way, no information will be provided and the group will be attacked by the perpetrator. If denouncing were cost free (e.g., no fear of retaliation and/or intrinsic motivation to support the potential attacker) the collective good of preventing an attack would always be provided, as all people would contribute. However, it is important to remind ourselves that these considerations only concern the individual level and I will consider aspects like altruism or norms and institutions only in the next section. For the purpose of studying the impact of group-selective violence it is important to consider if the assumptions made in the model outlined by

112

What Do We Know About Indiscriminate Violence?

Olson are met. As outlined above the collective action problem arises because of the fact that individual contributions to the collective good are small and—from an individual perspective—not rational since they only translate into a marginal provision of the public good while the costs are substantial. However, in the case of group-selective violence this may not necessarily be the case and there are several potential aspects to this. Individual contributions may have a more than marginal impact on the public good. For instance, consider a situation in which each individual has full knowledge of all partisan supporters. In this case, each individual would have the option to provide the full public good (i.e., preventing the attack) by denouncing all partisan supporters. The collective action problem would be reduced as the individual contribution in itself would be beneficial to the individual (as long as the cost of denouncing is lower than the cost of an attack). However, this may only partially address the collective action problem as each person may still hope that others will denounce and thereby get around paying the potential cost of denouncing. The collective action problem can be overcome by exclusion mechanisms (Buchanan, 1965). This argument would assume that the public good can be turned into a club good where the benefits can be enjoyed only by those who have contributed. In the field of group-selective violence, this would be the case if the potential perpetrator of violence treated those individuals who denounced partisan supporters differently from those who did not. For the perpetrator the provision of information would lead to increased information about the loyalties of individuals and the perpetrator could then further discriminate within the group. This is a possibility but somewhat out of my scope here, as it assumes a more sophisticated strategy by the perpetrator of violence which then leads to a shift to more selective violence. There is another, less obvious but more compelling, solution to the problem, which rests on the assumption that all individual contributions are necessary to achieve the provision of the public good. This deserves a bit more attention as it provides a full solution to the collective action problem that does not need regulating institutions in place. Olson assumes a specific aggregation function, which is the function by which the individual contributions are transformed into the public good. In Olson’s case this is the sum of the individual provisions which means that the sum of the efforts constitutes the public good. However, in cases where the aggregation function is not characterized by the sum of the contributions, the efficient provision of the public good can be much more likely. When the aggregation is the sum of the individual contributions, in large groups each individual has a marginal impact on the public good. As in the example of clean air, every individual can contribute to making the

A Theory of Group-Selective Violence

113

air slightly cleaner by paying for instance for additional chimney filters, etc. In such a case, the individual contribution is marginal in large groups. Therefore, withholding this contribution does not change the amount of the public good and free riding is rational. After all, one chimney pollutes the air only to a very limited amount. However, it is possible that the particular collective action problem is not characterized in this way. In some instances it might be that the aggregation is characterized by a minimum function (also known as weakest-link function). Hirshleifer (1983) and Hirshleifer (1985) introduced this function with several other authors building on his work (see e.g., Cornes, 1993; Cornes & Sandler, 1984; Vicary, 1990). In such a case, the individual contribution of the smallest size constitutes the public good. An often-used example is the height of dikes when several people are each responsible for a certain part of the dike. Each individual decides for himself how high he builds his part of the dike, but it is the smallest part of the dike that determines how high the tide can get before all the land is inundated. Hirshleifer and others demonstrated that collective action situations which are characterized by a minimum function, increase the incentives for each individual to contribute. Since the minimum contribution determines the public good, free riding is unlikely. Individually reducing one’s contribution immediately reduces the size of the public good. If an individual decides to not build a dike, inundation is certain. As a consequence, when the preferences of individuals are homogenous the minimum function leads to an efficient provision of the public good (Hirshleifer, 1983). Subsequent experiments confirmed this theoretical insight (Harrison & Hirshleifer, 1989). In addition, the weakest-link function also sets incentives for collaborations between individuals of a group. Vicary (1990) demonstrates that in this case richer individuals have an incentive to transfer money to less well-endowed individuals in order to provide higher amounts of the public good. Interestingly, the group size proposition of Olson does not hold in the case of a weakest-link function. Even if the size of the group increases each individual has an incentive to provide the optimal contribution because otherwise the public good will not be provided at all (D. C. Mueller, 2003, 22–25). Going back to the study of violence against civilians, it seems that the weakest-link function may characterize well the situation of groups which are threatened with group-selective violence. Let us look at a previously uninvolved group member. As mentioned before a critical issue is the distribution of information among group members which could be reported to the perpetrator (e.g., denouncing collaborators). If this information is not available to everyone in the group—which given the secretive nature of asymmetric warfare is plausible—some people may have unique information to contribute. If we assume for instance that each partisan supporter is

114

What Do We Know About Indiscriminate Violence?

known to only one uninvolved group member, the weakest-link function would apply. However, even if we assume that each partisan supporter is known to not just one person but a small group of people, the result would still be similar. Olson argues that very small groups can overcome the collective action problem easier than others. This means that the clandestine nature of the partisan activities which compartmentalizes groups to small clandestine circles may provide the basis for successful collective action. This in turn will lead group members to all contribute to a common collective good. Since the incentive structure makes it impossible to free ride, group members will be more willing to further the group goals than in other situations. If we now look at collaborators within the group the situation would be similar. Since each instance of collaborating with the opponent can potentially trigger a violent attack by the perpetrator, the collaborators are confronted with a similar weakest-link function when it comes to achieving the collective good of saving the group from violence. If they do not contribute (in this case this means refraining from collaborating) the collective good will not be provided. As a consequence, free riding on the efforts of others is not possible. However, this rests on the assumption that collaborators are concerned with the welfare of the group. However, this is not necessarily the case. For instance in Vietnam, Viet Cong operatives from outside were operating in villages across the countryside. In these cases, uninvolved civilians may need to act in order to prevent an attack. Numerous studies seem to suggest that experience of violence or the threat of violence leads to an amplification of collective action or support in favor of the community. For instance, studies on the development of parochial altruism (i.e., altruistic behavior that favors a specific group) suggest this connection. For instance, Choi and Bowles (2007) argue that conflict and war are the primary drivers of parochial altruism. In the civil war context, we also have evidence that conflict leads to cooperative behavior. For instance, Bellows and Miguel (2009) find in a household survey in Sierra Leone that households which were exposed to higher intensities of conflict were more likely to engage in communal activities afterwards. Similarly, Voors et al. (2012) find in a field experiment in Burundi that individuals exposed to violence were among others more likely to behave altruistically. Gilligan, Pasquale, and Samii (2014) provide similar results from a study in Nepal. Here too, altruistic behavior is increased by the exposure to conflict. In addition several studies on the experience of terroristic violence found that it increases altruistic behavior (Vázquez, Pérez-Sales, & Hervás, 2008, 70–71). Finally, observations of a similar kind can be found often in the literature on ethnic violence. For instance, Byman (1998) argues that ethnic terrorism and the subsequent violence from the government is mainly meant to form identity

A Theory of Group-Selective Violence

115

rather than the other way around. In addition, Brubaker (2002) argues that violence is particularly efficacious in creating groups and form what he calls “groupness” (Brubaker, 2002, 171). The argument is also a reflection of Kalyvas’ notion that it is the microdynamics of conflicts that shape identities, rather than identities shaping the dynamics of conflicts (Kalyvas, 2008, 2012). This seems to indicate that the outside threat of being attacked as a collective, facilitates collective-action which may even lead to norms and institutions for collective action that outlast the actual threat (Ostrom, 2000). While it is not possible to discern from these studies if groupselective violence is responsible for the increase in collective action or identity formation, the above theoretical considerations would suggest that group-selective violence can create an incentive structure that increases individual contributions to a collective good. Those communities which were threatened by group-selective violence and which consequentially had to collaborate may also be more likely to develop norms and institutions for cooperation. This means that rather than a necessary precondition, norms for cooperation or group-cohesion may be a consequence of group-selective violence. Kalyvas makes a similar observation with regard to selective violence, when he finds that the avoidance of violence can lead to an institutionalized ethos of solidarity (Kalyvas, 2006, 307). To summarize, this discussion was intended to make plausible that group-selective violence can provide an incentive structure that makes collective action likely, even without including specific assumptions about the group’s properties. Particularly by confronting innocent or uninvolved individuals with the necessity of providing a public good that is defined by a weakest-link aggregation function, it is likely that individuals will contribute (i.e., comply with the demand of the perpetrator). Therefore, it is not necessary to assume that groups have norms or institutions that facilitate collective action. However, such norms and institutions may be able to increase the potential for collective behavior. The Group Level

So far, the model outlined here has only taken individual aspects into account and ignored the possibility that individuals are related to, and interact with, other members of the group. However, one of the most basic aspects of groups is the fact that they are grounded on the social connections between their members (Forsyth, 2009, 2–4), and it is well-established that the group-level can have an effect on various individual decisions: to take just one example among many possible, in research on terrorism and insurgency it is recognized that groups (e.g., groups of friends or families) can facilitate recruitment into armed groups (Sageman, 2004).

116

What Do We Know About Indiscriminate Violence?

As a consequence, it is likely that the way social groups are bound together impacts on the effectiveness of group-selective violence. In this section, I will briefly discuss which aspects of groups could be relevant for group-selective violence and how research from sociology and social psychology may inform differences in the effectiveness of the coercive message on different types of groups. As discussed before, the main goal of group-selective violence is to coerce the neutral and innocent population into supporting the perpetrator. In the previous section, the coercive threat entered the individual calculation of an innocent person exclusively through a reassessment of the individual’s own security. This way a person considers that any attack on his group may kill him, too. As a consequence, the individual is more inclined to ensure that the group as a whole complies with the coercive demand, which leads to efforts to prevent the group from being targeted. However, two group-aspects can have an influence in this process. First, the attitude of the individual towards the group can change the assessment by the individual. If the individual values other members of the group or the group itself, he will have an even higher incentive to comply with the coercive threat. Secondly, the ability of individuals to exercise social control will be important. The more some individuals within the group are able to influence others, the more likely it will be that efforts to prevent the group from being targeted are effective in the aggregate. This first aspect concerns the attitudes of the individual towards the group or the other individuals. A person may value the other members of the group or the group itself as important for their own wellbeing and this can have several dimensions. A) The individual can feel an emotional affection to members of the group or to the group itself. The most obvious cases of this kind concern affection for family members, spouses and friends. In such a case the individual would include the potential harm to those with whom he has affectionate bonds in his cost-benefit calculation. 7 However, it is also possible that individuals have an abstract affection towards the group. An example of this kind of emotion is patriotism, which is generally seen as an affection between individuals and their nation (Bar-Tal & Staub, 1997; Kosterman & Feshbach, 1989). Here the individual would incur a disutility from any attack on the group. Thus, while the implications slightly differ, positive emotions towards the group and positive emotions for some of its members should increase the expected costs of punishment for each individual. B) Purely unemotional reasons may be responsible for the valuation of other group members or the group. Groups are often built on reciprocity and provide individuals with social capital. In sociology social capital is usually defined as “the ability of actors to secure benefits by virtue of membership in social networks or other social structures”

A Theory of Group-Selective Violence

117

(Portes, 1998, 6). Such a definition highlights that for each individual other members of a group or the group itself can be seen as an asset. Since the group is able to provide benefits an individual has an incentive to protect it. For instance, if an individual derives benefits from being a member of a tribe, e.g., through access to land and resources, a weakening of this tribe is detrimental to the individual as the tribe will be less able to provide those benefits. Thus violence against the tribe depreciates the social capital of the individual. The second aspect concerns the ability to exercise social control within the group. As outlined above, the coercive threat works in a way that may give all individuals an incentive to contribute individually to the public good in order to avoid being targeted. However, some individuals, for idiosyncratic reasons, may not follow this incentive. For instance, an individual may seek revenge even at the cost of his own life. Therefore, in cases where individuals need to be prevented from opposing the perpetrator of group-selective violence, the ability of the community to exercise social control, defined as the ability for self-regulation, will play a role. By now a large literature exists showing that various forms of selfregulation (e.g., of common pool resources) are possible (see Ostrom, 2000). Whenever such regulatory mechanisms exist, the community should be more likely to enforce a uniform group-behavior than in cases when no mechanisms for social control exist. If we again think of concrete examples, in asymmetric wars, it is sometimes found that village leaders or village councils can play such a regulatory role (e.g., Souleimanov & Siroky, 2016). Empirically, research in social psychology and sociology provides much support for the notion that group aspects, often summed up under the term social cohesion, can have an influence on group behavior. For instance, experimental results show that even minimal amounts of social cohesion of a group can increase individual propensity to act according to established norms and overcome the so-called bystander effect, which is a phenomenon in which people are less likely to offer help when other people are present (Rutkowski, Gruder, & Romer, 1983). However, there is one research approach that seems particularly relevant for the study of groupselective violence as it also focuses on in-group policing. Research on the prevention of neighborhood violence emphasizes the importance of social cohesion. Multiple researchers have pointed out that the social fabric of neighborhoods has an influence on the ability of the community to control violent crime. For instance, Sampson, Raudenbush, and Earls (1997) argue that “collective efficacy,” which they define as a combination between social cohesion and the willingness to take action on behalf of the group, can be a key influence in reducing violent crime in specific neighborhoods. The idea behind this approach is that social cohesion

118

What Do We Know About Indiscriminate Violence?

gives the members of a neighborhood the ability to control criminals. Thus, cohesive groups are more able to prevent crime than others. However, as Browning, Feinberg, and Dietz (2004) note, for the prevention of crime, the ability of the groups to control the behavior of potential offenders is conditioned by the social capital of those potential offenders. If for instance, criminals accumulate enough social capital within a group, the group may not control them anymore. In the case of group-selective violence this means that if within a group a critical mass of opponents is reached, who cannot be deterred (e.g., because they seek revenge without regard for the costs), social control will work in the opposite direction. Then, opponents may be able to use their social capital in order to turn the rest of the group against the coercer (Browning et al., 2004, 510). However, such a situation should be rare as it entails the group of offenders valuing neither their own lives nor the survival of the group enough to change their behavior. This means that the drastic punishment that lethal group-selective violence constitutes makes it less likely that the social capital of offenders can actually save them. However, less severe punishments may be less likely to produce the coercive effect. To a certain extent this relationship may be the basis for the oftnoted problem that too low levels of punishment do not work (see e.g., Lake, 2002). In addition to crime, social cohesion of the neighborhood has an effect on a broad variety of social phenomena (Sampson, 2012). This way it seems that communities are able to police themselves, and social cohesion is important in doing so. The important point of this brief discussion of group properties is that the impact of the coercive threat on collective action might be moderated or heightened by the underlying social fabric of the group. Smaller, tightly knit groups like families where affection and group identification are high and information and social control are generally available may experience group dynamics that support the coercive effects on the individual. These group dynamics may in general increase the effectiveness of the coercive threat. However, it has to be noted that these same group properties may also increase some of the provocative effects of group-selective violence when it is used. For instance, the more affection a group member has for his group, the more likely he will experience feelings of revenge that motivate him to fight back against the perpetrator once the group is actually targeted. In addition, social control may work in the opposite direction, once a critical mass of the group opposes the perpetrator of violence. If, for instance, the leadership of a village comes to the conclusion that the perpetrator needs to be fought, they may exercise their influence in order to make more people join them. Browning et al. (2004) make this argument for neighborhood criminality. This means that for both, the positive effects and the neg-

A Theory of Group-Selective Violence

119

ative effects of violence, the strength of the bonds within groups may positively reinforce the effects of violence. In sum, on the individual- and on the group-level, arguments can be made for the potential effectiveness of group-selective violence in coercing a group as a whole. Group-selective violence may lead to a transformation of the aggregation function of the collective action problem. By changing from an additive to a weakest-link function, collective action becomes rational for individuals in those groups who receive the coercive threat. On the group-level, group-selective violence may be more effective against groups with strong bonds and the ability to exercise social control. It is important to note, however, that group-selective violence may have a coercive effect, even in the absence of most of the group properties. The attitudes of an individual towards the group or other group members will increase the coercive effect but the effect may occur even in their absence. Similarly, the ability to exert social control might make it easier to control those with deviant preferences, but again, individuals each have an interest in exercising control, even in the absence of existing control mechanisms. Conclusion Here, I outlined a theory of group-selective violence. Group-selective violence punishes a group collectively for an unwanted behavior in order to deter groups from engaging in that behavior. I distinguished particularly two types of the group-selective targeting strategy: specific and general groupselective violence. Specific group-selective violence is a strategy where an actor targets a group with coercive violence and seeks to change the behavior of that same group. General group-selective violence is a strategy where the targeting of one group is supposed to change the behavior of other groups (see Figure 5.7). In addition, the individual level is different from other forms of violence as it deliberately targets the uninvolved civilians. By this, the perpetrator outsources the obligation to find those who are responsible for an unwanted behavior and circumvents the identification problem. In sum, the theory highlights the need to study particularly those who remained uninvolved, in order to understand the effects of group-selective violence, groups and individuals. Only by focusing on their behavior can we assess how group-selective violence may benefit the perpetrator. The theory of group-selective violence as outlined in this chapter explains the central puzzle of the study of indiscriminate violence: How can indiscriminate violence be effective in coercing? Building on the insights of the terrorism literature (e.g., Abrahms, 2006, 2013; Goodwin, 2006b; Pape, 2003, 2005; Schmid, 2005) and the literature on government violence (Douglass, 2012; Kalyvas, 2006; Schubiger, 2014), the theory points to two

120

What Do We Know About Indiscriminate Violence?

Figure 5.7 The Effects of Specific and General Group-Selective Violence

separate strategies with differing effects: A) General group-selective violence can achieve a coercive effect that outweighs the provocative effects. B) Specific group-selective violence may often not be able to achieve a positive net effect. As a consequence, the theory provides the missing link between attacks on civilians and positive outcomes for the perpetrator as they are found for instance in Schubiger (2014), Douglass (2012), and Johnston (2009). For instance, Schubiger (2014) argues that violence against civilians can lead to mobilization for the perpetrator, Douglass (2012) points to the connection between selective violence and indiscriminate violence, and Johnston (2009) emphasizes the importance of targeting. However, none of the approaches addresses the problem that indiscriminate violence must have both positive and negative effects and how it needs to be targeted in order to be coercive. The approach presented here builds on these previous works and remedies some of the shortcomings, as it explains the conditions for the instrumental value of violence by the perpetrator against innocent civilians. In addition, the theory is able to offer an explanation for the contradictory empirical finding that violence may be effective for the perpetrator in some cases (Arreguin-Toft, 2005; Johnston, 2009; Lyall, 2009), and detrimental in others (Arreguin-Toft, 2003; Kocher et al., 2011; Schutte, 2017b).

A Theory of Group-Selective Violence

121

Since group-selective violence produces provocative as well as coercive effects both findings may be right. It seems to be one of the strongest aspects of this theory that it can provide a solution for this empirical puzzle. Notes 1. Remember that although Kalyvas defined indiscriminate violence as violence against groups his description of the effects characterized violence as random (see Chapter 2). 2. Kalyvas’ theory is concerned with the origins of this type of violence and he argues that denunciation and territorial control are the key to understanding the patterns of violence (Kalyvas 2006, 173–209). 3. As mentioned above, the effectiveness of violence depends also on the feasibility and the desirability of the demanded action (and with it the preferences of the receiver) as well as the interaction between sender and receiver. To judge a specific incident of coercion, it would, therefore, be necessary to take these criteria into account as well. However, in order to assess the differences between types of violence, this is not necessary. 4. The term moral hazard stems from economic theory. It describes a situation in which an information asymmetry between a principal and an agent leads to incentives by the agent not to comply with the demands of the principal (Eisenhardt, 1989). In the case of group-selective violence it is of course ethically not justified to speak of “moral hazard” by the civilian population. 5. For instance Barter (2014) emphasizes the room for maneuver that civilians have. 6. However, Olson’s propositions are found to be untenable in their strong form. For a discussion see, for instance, Sandler (2015). 7. For an example of how altruism is incorporated into an individual cost-benefit calculation see e.g., Levine (1998).

6 Assessing the Effects of Indiscriminate Violence

After having concentrated on the theoretical arguments, I now turn to the empirics of indiscriminate government and nongovernment violence. While studies on the causes of indiscriminate violence against civilians are manifold, I will concentrate only on the effects of violence. Here, I will examine the evidence that is currently available to support the positive and the negative view. The aim of this chapter is twofold. On the one hand, it seeks to evaluate what is the current knowledge about indiscriminate violence. Given that some authors remain ambiguous in their conceptions of indiscriminate violence, I will outline that at least in some cases this has translated into problematic measurement and problematic research strategies. On the other hand, this chapter relates previous research to the theory of group-selective violence outlined in the previous chapter. While authors generally do not focus on group-targeting, some of the evidence can be seen as supporting the theory of group-selective violence. Government Violence Three different research strategies can be distinguished: First, whether actors behave according to a specific logic of violence (Douglass, 2012; Fjelde & Hultman, 2014; Kalyvas, 2006; Schutte, 2017a; C. M. Sullivan, 2012; Valentino et al., 2004, 2006). For instance, if killing civilians in guerrilla wars is more effective than in other wars, governments should be found using more violence in guerrilla wars. Therefore, these researchers investigate whether the determinants of violence are in accordance with its hypothesized effects.

125

126

How Does Indiscriminate Violence Work in Practice?

Second, the impact of indiscriminate violence on the macro level (Arreguin-Toft, 2001, 2003, 2005; Downes & Cochran, 2010; Johnston, 2009; Schubiger, 2014). These approaches seek to show that indiscriminate violence increases the probability of certain outcomes or has an impact on the duration of conflicts. Third, violence on a disaggregated level (Asal, Gill, Rethemeyer, & Horgan, 2015; Benmelech, Berrebi, & Klor, 2015; Kocher et al., 2011; Lyall, 2009; Schubiger, 2014; Schutte, 2017b; Zhukov, 2014). This research seeks to evaluate the local impact of indiscriminate violence. The Determinants of Indiscriminate Violence

To begin with, governments may choose between different forms of violence (selective vs indiscriminate). However, it is generally assumed that indiscriminate violence is less effective than selective violence. Selective violence, Kalyvas argues, is therefore used whenever information on collaborators or supporters of the opponent is available, and he hypothesizes that indiscriminate violence should be used predominantly in areas where less intelligence exists. In particular, selective violence is used in areas highly but not fully controlled by an actor, while indiscriminate violence is used in areas of little or no control (Kalyvas, 2006, 146–209). This hypothesized pattern has been confirmed by a number of studies. In his original study on violence Kalyvas corroborated this pattern for a sample of villages during the German occupation of Greece and in the Greek civil war between 1943 and 1949 (Kalyvas, 2006). In addition, Kalyvas and Kocher (2009) used disaggregated data on the Vietnam War and found that South Vietnamese indiscriminate violence in the form of artillery shelling or aerial bombing was directed predominantly at areas where the opponent enjoyed full control. C. M. Sullivan (2012) found that, in Guatemala between 1959 and 1996, areas with higher rebel control were more often the target of government massacres. Schutte (2017a) provided a study of 11 cases of insurgency and found that indiscriminate violence was used more frequently in places that were further away from the home base of the perpetrator. Finally, Douglass (2012) tested the underlying assumption of the control-collaboration mechanism put forth by Kalyvas. In his empirical investigation Douglass used information from the Hamlet Evaluation System (HES) on the Vietnam War to analyze the determinants of selective and indiscriminate violence. He found that an increase in the provision of information to the counterinsurgent reduced the share of indiscriminate violence, which directly corroborates Kalyvas’ logic and confirms that indiscriminate violence is indeed influenced by the availability of information on collaborators.

Assessing the Effects of Indiscriminate Violence

127

Other researchers find indiscriminate violence to follow additional patterns. For instance, Valentino et al. (2004) found that guerrilla wars have a higher probability of experiencing mass killings1 since here civilians are an asset of guerrilla forces. In addition, they found that governments are more likely to use violence against guerrilla groups that pose a major military threat and that have high support from the civilian population. They interpreted this as evidence for governments trying to violently separate civilians from rebels. Similarly, Fjelde and Hultman (2014) investigated the determinants of violence in Africa and found that both sides in civil wars target the opponent’s ethnic constituency. However, compared to Valentino et al. (2004), Fjelde and Hultman’s measure of violence against civilians also included much smaller amounts of killings. Finally, C. M. Sullivan (2012) investigated government violence in the Guatemalan Civil War with disaggregated data. His dependent variable was the number of massacres, defined as more than five people killed in a given event. He found that attacks seemed to be directed especially at municipalities with a higher amount of indigenous population, which were perceived as supporting the rebellion. However, he also found that violence was directed against those municipalities that previously experienced more rebel attacks and had higher levels of rebel control. All this shows that there is ample support for the argument that indiscriminate violence is used mostly when little information is available on individual opponents or when the population is perceived as supporting the opponent. Selective violence seems to be preferred whenever possible (i.e., in cases where information on collaborators is available) and perpetrators of violence seem to be reluctant to use indiscriminate violence. This can be seen as evidence for indiscriminate violence being less effective than selective violence if information is available. However, indiscriminate violence is still used and it must be assumed that the perpetrators can gain benefits from it if not enough information is available for using violence selectively. Indiscriminate violence against civilians seems to target those groups of civilians that are seen as supporters of the opponent, particularly when the opponent can be defined ethnically. Valentino et al. (2004), Fjelde and Hultman (2014), and C. M. Sullivan (2012) highlight that perceived support for the opponent plays a key role in the targeting. Taken together, both these points confirm that indiscriminate violence is used with relative restraint which indicates that it is used deliberately. Although indiscriminate violence may not have the same effects as selective violence, it is not used randomly but targeted at specific groups. This is very much in line with the previously outlined theory of group-selective violence that highlighted the fact that group targeting is the key to generating coercive effects.

128

How Does Indiscriminate Violence Work in Practice?

Macro Level Effectiveness

Other studies tested directly the effects of indiscriminate violence. On the macro level, numerous studies were conducted to assess whether indiscriminate violence can be effective in achieving military or political goals. However, the results were mixed. On the one hand, several studies concluded that indiscriminate violence was overall counterproductive. To begin with, Arreguin-Toft (2003) investigates a strategy he calls “barbarism” (i.e., the targeting of civilians), focusing on conventional wars. In a bivariate analysis, however, he found mixed results: barbarism increased the chance of a victory for weak actors and decreased the chance for strong actors. In his following case studies on the German occupation of Serbia in World War II and the Kosovo War (i.e., on strong actors) he concluded that barbarism backfired as it increased the number of opponents. Schubiger investigated the influence of indiscriminate violence on conflict duration and conflict outcomes (Schubiger, 2014, 193–241). Her results indicate that government violence against civilians was associated with shorter conflict durations and increased the probability of rebel victories as well as the conflict fading out. She found no effect on negotiated settlements and government victories. Schubiger interpreted this as evidence for government violence being ineffective. On the other hand, several investigations found indiscriminate violence to have positive effects for the perpetrator. Arreguin-Toft (2001, 2005) concluded that indiscriminate violence can be effective when used in asymmetric warfare against guerrillas. He also provided some evidence for the effectiveness of barbarism in several case studies. Arreguin-Toft pointed out that only in Russia’s counterinsurgency in Afghanistan (1979–1989) did barbarism against guerillas not lead to success. He attributes this failure to the rebels’ reliance on external sanctuaries and support. In all other cases the targeting of civilians helped the perpetrator to win (Arreguin-Toft, 2005, 215–216). Johnston (2009) also found indiscriminate violence to increase the probability of winning a war. Using a dataset on 168 counterinsurgent campaigns between 1800 and 1999, he discovered that governments using violence against civilians generally have a higher chance of winning the campaign than others. His results suggested an inverted u-shaped relationship: Up to around 10,000 civilians killed in a campaign, the chances of winning increase. After this threshold, they decrease again. Testing the effects of mass violence (more than 50,000 dead), he found that it is associated with losing a counterinsurgency campaign. Johnston supported this evidence with three case studies, the American-Philippine War 1899–1902, the Vietnam War 1955–1975, and the Darfur War that started in 2003. He found that coercive strategies in the first two wars were largely successful, while in the Darfur War violence was annihilatory and counterproductive.

Assessing the Effects of Indiscriminate Violence

129

To study the effectiveness of violence against civilians in international wars from 1816 to 2003, Downes and Cochran (2010) analyzed whether civilian victimization (i.e., the deliberate targeting of civilians or very indiscriminate use of violence where tens of thousands die) had an impact on the probability that a country would win a military contest. They found that, in general, violence is associated with winning but that the effectiveness of violence decreases in the 20th century. In addition, smaller targets are more vulnerable to violence than larger countries, although democracies are not more vulnerable than autocracies. In their investigation of case studies, they concluded however that association between victimization and winning might stem from a reversed causation. In order to target civilians, an army has to be on enemy territory (at least this is true for most wars before World War I) and might therefore, be in a better position to win. This may be the reason for the finding that violence seemingly lost its effectiveness over time. These findings on the macro level are contradictory and may be seen as support for the effectiveness or counterproductiveness of indiscriminate violence. In particular, the two larger studies of Schubiger and Johnston provide diametrically opposed findings while Ivan Arreguin-Toft presents research that supports both the views that indiscriminate violence does and does not help the perpetrator. A closer look at the differences between these studies is helpful in understanding some of these results. It seems that studies focusing on counterinsurgency or asymmetric warfare, Johnston (2009) and ArreguinToft (2001), tend to report a positive effect, and those focusing either on conventional wars, as Arreguin-Toft (2003) and Downes and Cochran (2010) or including all wars as Schubiger (2014, 193–241), tend to find no strong effects. In addition, although the timeframes differ, there seems to be no trend involved. While Downes and Cochran (2010) report that their positive findings hold only for the 19th century, Johnston’s analysis also provides a positive finding for the post-1945 period (Johnston, 2009, 115). Nevertheless, three problems seem to exist with studies on the macro level, which call into question their ability to assess the coercive effectiveness of indiscriminate violence. First, analyses on the macro level often show an incongruence between the level of analysis and the object of investigation. For the study of coercive violence, it is important to acknowledge that this sort of violence is meant as a threat at a certain point in time in order to achieve some concession. Therefore, using the civil war, war or counterinsurgency campaign as a level of analysis is inappropriate as it obscures the nature of the threat (Arreguin-Toft, 2003, 2005; Johnston, 2009). At this level we only have one observation per war and threats would have to be aggregated over time. This way, we are, for instance, unable to distinguish between one large or several minor threats. However, the impact of a threat considerably hinges on how it is issued. Thus, a macro level research design that does not at least disaggregate the time axis

130

How Does Indiscriminate Violence Work in Practice?

(i.e., include observations for different time periods) is not well suited for studying coercion. Second, and partly related to the first point, none of the studies on the macro level considers which groups are targeted and which are being coerced. As outlined in Chapter 5 it is important to realize that the coercive effect of group-selective violence is not necessarily found in the targeted group. Potentially the coercive effects of violence must be looked for in groups that are not even part of the conflict, yet (see Chapter 8). It is therefore important to have an explicit hypothesis about group targeting and the location of the coercive effect in order to adequately analyze the effectiveness of group-selective violence. Third, most macro level studies measure effectiveness with reference to an outcome like campaign victory. Particularly in Johnston’s analysis, it becomes apparent that campaign victory is a questionable measure of effectiveness. The main problem is that counterinsurgent campaigns are often embedded in larger wars or political circumstances. Failure might therefore reflect a failure in the larger context and not in the counterinsurgency campaign. Johnston realizes this problem himself in his case study on the Vietnam War. This war is counted as a failure in his dataset, although he finds that the US makes great progress after the Tet Offensive. Yet, the Vietnam War is not an exception in Johnston’s data. For instance, the German antipartisan campaigns during World War II, which are all counted as a failure in Johnston’s dataset, are equally problematic. The German campaign in Serbia makes great progress in 1942 to the point that no partisan activity was present anymore. Yet, with troops thinned out due to the ongoing war in the East, the partisans grew stronger in Italian-occupied Yugoslavia and the campaign was ultimately lost (Shepherd, 2010). In many counterinsurgency campaigns the incumbents make military progress, but “lose” due to political or external factors. This raises the question if the political outcome of a campaign is an adequate measure of coercive success. As a consequence, the results of the macro level studies are difficult to use as evidence for the effectiveness or ineffectiveness of indiscriminate violence. At best it could be argued that the results show that indiscriminate violence can sometimes be beneficial for the perpetrator, while at other times it might be detrimental. However, the reviewed studies cannot help us in understanding why or under which circumstances indiscriminate violence may have a particular effect. Micro Level Effectiveness

A growing number of studies investigate the effects of indiscriminate violence on the micro level. Here too results are mixed and studies provide evidence for the effectiveness and counterproductiveness of indiscriminate

Assessing the Effects of Indiscriminate Violence

131

violence. On the one hand, a large number of investigations find that indiscriminate violence has counterproductive effects. Kocher et al. (2011) investigated the effect of indiscriminate bombings in Vietnam with the hamlet as their level of analysis. They consistently found that bombing in one period of time was associated with increases of rebel control in the following period. Consequently, the authors concluded that indiscriminate violence was generally counterproductive. Similarly, the results of Balcells (2010) seem to indicate that indiscriminate violence triggers revenge, although her investigation is not directly concerned with the counterproductiveness of violence. In a study on the Spanish Civil War she found that municipalities that experienced higher levels of direct violence against civilians in a first period also experienced higher amounts of violence later. Balcells interpreted this as evidence for revenge killings. Lyall, Blair, and Imai (2013) sought to find out whether the effects of civilian victimization on civilians’ attitudes were conditional on their ethnic identity. Lyall, Blair, and Imai (2013) analyzed the results of a large survey in Afghanistan, where people were asked about their experiences of being victimized and their attitudes towards both the International Security Assistance Force (ISAF) and the Taliban. They found that Pashtun respondents saw Taliban violence directed against them as less repulsive than ISAF violence. Experiences of being victimized by ISAF decreased support for ISAF and increased support for the Taliban. To the contrary, being victimized by the Taliban did not lead to increased support for ISAF and led only to a small detrimental effect on support for the Taliban. The authors argued that the differing effects of violence were the result of the ethnic identity of the perpetrator. Another approach comes from Asal and his colleagues (2015) who were primarily interested in the probability and lethality of PIRA attacks in Northern Ireland between 1970 and 1998. However, they also analyzed the effects of government-related killings of civilians on terrorist attacks at the county level. Yet, the results were mixed. While British violence against civilians led to reduced probability of attacks the following year, those attacks that did follow tended to be more lethal. For the authors, the findings reflected PIRA’s strategy of seeking economic damage and not civilian casualties. They argued that the increase of lethality might stem from either an emotional reaction or that reduced planning supposedly went into these attacks. Further evidence comes from the study of house demolitions. House demolition is a strategy used by the Israeli Defense Forces to combat and deter terrorism, with two types being used: precautionary and punitive demolitions. Precautionary demolitions seek to prevent attacks from a certain location while punitive demolitions are directed against the houses of terrorists or “individuals suspected, detained, or convicted of involvement in

132

How Does Indiscriminate Violence Work in Practice?

violent acts against Israelis” (Benmelech et al., 2015, 29). Studying their effect on suicide terrorism, Benmelect et al. found that punitive demolitions, which Benmelech et al. interpret as selective violence, led to a reduction of suicide terrorism originating from that same district. By contrast, precautionary demolitions, which they saw as indiscriminate, led to an increase of suicide terrorism from that location. Schutte (2017b) investigated whether indiscriminate violence can lead to either increased support for the adversary or increased support for the perpetrator. He uses data on Afghanistan published by US forces covering the time between 2004 and 2010. As a dependent variable, he used a measure of civilian collaboration with the US forces: the turning in of unexploded ordnance as an indication of civilian support for the US-led coalition. As independent variables, he used a proxy for indiscriminate violence. He measured the use of indirect fire by US forces and the explosion of land mines placed by Taliban forces, which he assumed to target civilians more frequently than either direct fire or direct Taliban attacks. His measure of indiscriminate violence led to a reduction of support for the perpetrator. US violence led to decreases in turning in ordnance to US forces while Taliban violence led to increases. So far, the research presented on the micro level has supported the negative view. On the other hand, an increasing number of authors find support for the positive view. One of the first systematic studies to find a coercive effect of violence was Lyall who investigated the effect of random artillery attacks with data from the Second Chechen-Russian War between 2000 and 2005. The dependent variable was rebel attacks, which comprised “insurgentinitiated attack against Russian or pro-Russian Chechen proxy forces, their local representatives, and civilians at the village level” (Lyall, 2009, 340). Lyall’s results showed that the random artillery shelling of a village was associated with a reduced number of rebel attacks in the 90 days after the artillery attack, compared with the 90 days before the attack. Zhukov (2014) presented two tests on the effectiveness of violence. The first test was a disaggregated analysis of 80 conflicts since 1997 with data from two databases, the Armed Conflict Location and Event Database (ACLED) and the Global Database of Events, Language, and Tone (GDELT). The second test used disaggregated data on the First Chechen War 1994–1996. He used instances of selective and indiscriminate violence as independent variables, but contrary to most other authors he coded battle events (i.e., military fighting) as selective violence and violence against civilians as indiscriminate violence (Zhukov, 2014). He found that government violence (selective and indiscriminate violence combined) had an inverted u-shaped effect on rebel attacks: low levels increased attacks and high levels decreased them. However, indiscriminate violence was effective in suppressing rebel activity only at very high levels.

Assessing the Effects of Indiscriminate Violence

133

Finally, Schubiger (2014) conducted a disaggregated study of the Peruvian Civil War where she evaluated the effect of state indiscriminate violence in one period (1983–1985) on mobilization of self-defense forces in another period (1986–1988). Since mobilization cannot be observed directly, her dependent variable was a binary variable that measured whether a given locality experienced rebel violence against the selfdefense forces of that locality. The variable on state indiscriminate violence was a dummy whether a given locality was affected by state violence in the treatment period. Her measure for mobilization was associated with higher state violence in the treatment period. In sum, research on a subnational level shows similarly contradictory results as at the macro level. It is again helpful to take an in-depth look at the different approaches which these studies take in order to assess their merits. Among some of the studies of the micro level, the concepts of indiscriminate violence, selective violence, and collateral damage are sometimes confused. As outlined in Chapter 2, a problem in the study of violence is the ambiguous definition and description of indiscriminate violence found in the theoretical literature. This seems to translate into problems with the empirical study of the phenomenon. For instance, some authors conflate indiscriminate violence as a targeting strategy (which is generally defined as violence targeting groups) with collateral damage (e.g., Lyall et al., 2013; Schutte, 2017b). Schutte points out that “based on the assumption that attacks are more or less selective dependent on the tactics and weapons systems involved, I focus mainly on direct versus indirect fire and their effects on civilian collaboration” (Schutte, 2017b). Unfortunately, neither measure of indiscriminate violence in Schutte’s paper, US artillery and aerial attacks, nor Taliban-placed land mines, target civilians deliberately as a strategy and the measure is more apt to capture unintentional killings, i.e., collateral damage. However, as discussed in Chapter 5, indiscriminate violence can only work as a coercive strategy, when the pattern and the incentives are clear to the civilian population. Collateral damage is by definition unable to change the behavior of civilians as it does not follow an intentional pattern. As a consequence, it does not make much sense to investigate whether collateral damage can increase support as there is no theoretical reason to believe that it would. A similar problem may exist with the study of Lyall et al. (2013) who compare the effects of ISAF violence with the effects of Taliban violence. Here, they do not account for the fact that ISAF violence did not follow a deliberate strategy of targeting civilians, while Taliban violence did (see e.g., Foundation, 2010). Again, ISAF violence is collateral damage and thus can be assumed to be random. Therefore, it cannot coerce. Taliban violence is likely not to be random and will, therefore, be more likely to produce positive effects. As a consequence, their findings may be the result of different types of targeting.

134

How Does Indiscriminate Violence Work in Practice?

Another conflation concerns selective violence and violence against groups (Benmelech et al., 2015). In their study, Benmelech et al. compare demolitions directed against the families of known terrorists (punitive) with demolitions seeking to clear spaces that are dangerous for Israeli troops (precautionary). They argue that punitive demolitions are selective violence and precautionary demolitions indiscriminate violence. Yet, what they call punitive demolitions is not selective violence by generally accepted definitions, as selective violence targets individuals based on their own behavior. Of course, the authors are right to notice a certain selectivity. However, it is a selectivity on the group level, which I call group-selective violence. Finally, Zhukov (2014) conflates military violence with violence against civilians. Although he is only interested in violence against civilians he uses instances of military violence as a proxy for selective violence. While there may be reasons to do this, his empirical study can only partially speak to the literature on selective and indiscriminate violence against civilians. Only his findings on indiscriminate violence (which comprises all violence against civilians) can be taken as relevant. Disentangling the findings in this way helps to paint an improved picture. In general it can be observed that whenever researchers use measures that go into the direction of collateral damage or random violence (Schutte, 2017b), they find mostly provocative effects. This is further corroborated by studies that focus explicitly on collateral damage (Condra & Shapiro, 2012). By contrast, whenever researchers use measures that capture groupselective violence (Benmelech et al., 2015), they find positive effects. Even those studies that do not generally conflate different forms of violence against civilians often remain unclear about the targeting pattern. For instance, Kocher et al. (2011) and Schubiger (2014) both explicitly define indiscriminate violence as violence which targets groups. However, neither study explains which groups are targeted and what the effect of such targeting is. A peculiar case in this regard is the study by Lyall (2009). On the one hand, Lyall acknowledges that indiscriminate violence against civilians in civil wars is part of a strategic process to shape behavior. On the other hand, he argues that random artillery shelling makes sense as a measurement of indiscriminate violence. This obviously creates a certain tension. Either violence is actually random and we should not expect a positive effect for the perpetrator, or violence is not random and it would be necessary to take the targeting pattern into account. In addition, many of the micro level studies rely on an outcome measure (i.e., a dependent variable) that concerns rebel behavior (Asal et al., 2015; Benmelech et al., 2015; Condra & Shapiro, 2012; Lyall, 2009; Schubiger, 2014; Zhukov, 2014). However, such a measure is highly problematic for two reasons: A) indiscriminate violence in one location does not necessarily influence rebel attacks in that same location, and B)

Assessing the Effects of Indiscriminate Violence

135

a decrease in the amount of attacks is not tantamount to a decrease in rebel strength.2 To begin with, there are two mechanisms that link indiscriminate violence to rebel behavior: on the one hand, it increases recruits for the rebels and, on the other hand, civilians may lobby or pressure for lower rebel activity. However, the mechanism of increasing recruits may only have long-term consequences that are felt somewhere else entirely. Recruits most likely will not stay in their own locality (see e.g., Souleimanov & Siroky, 2016). It is only the second mechanism that would lead to a local response if we assume that the effect on civilians is confined to the area of the attack. However, the deterrence effect on civilians may also not be confined in that way. For instance, civilians of one village can learn from government attacks in another village what kind of behavior the government seeks to punish. Therefore, it is unclear why we would try to establish that indiscriminate violence in one place leads to increases or decreases of attacks in that same place. In addition, attacks are a function of military strategy and tactics of both sides in a conflict (if it is a two-party conflict). This means for a rebel attack to take place an adequate target has to be in the area. Rebels cannot attack government troops when there are no troops and they are less likely to attack if government troops are very strong in an area. As an observable implication, this means that rebel attacks are very unlikely in places with very high and very low government presence. Rebel attacks on government troops are, therefore, a measure that does not capture well rebel strength. This also means that changes of rebel attacks are also subject to changes in government behavior. If, for instance, government troops avoid a certain area, rebels cannot attack anymore. On the positive side, however, others have provided interesting alternatives (Douglass, 2012; Schutte, 2017b). Douglass and Schutte use measurements on civilian collaboration with the government. These alternatives should serve as an example for future research as they are less prone to many of the problems mentioned above. Unfortunately, these studies do not focus on the effects of group-selective violence, yet. The Effectiveness of Government Violence

In summary, the previous sections gave an overview over the most recent studies that provide insights into the effectiveness of indiscriminate violence. However, overall, these findings give a mixed picture. On nearly all levels studies support the positive and the negative view on the effectiveness of indiscriminate violence. In addition, it seems that on all levels some of the studies have taken problematic approaches which force us to interpret their findings with caution.

136

How Does Indiscriminate Violence Work in Practice?

Yet, despite all the problems some tentative conclusions can be drawn. Evidence on the effects of indiscriminate violence can be found in disaggregated studies. On the micro level, some solid evidence can be found on the counterproductive effects of violence that is perceived as random, mainly but not exclusively through collateral damage (Benmelech et al., 2015; Lyall et al., 2013; Schutte, 2017b). Similar findings come from studies that do not specify the targeting pattern (Kocher et al., 2011). Less strong but still interesting are the findings on a potential coercive effect of indiscriminate violence. Some limited evidence shows that indiscriminate violence (Lyall, 2009; Schubiger, 2014) may lead to reduced rebel activity and increased progovernment mobilization, although the targeting patterns are often not clear. Unfortunately these results cannot support the hypotheses of the previous chapter directly. Nevertheless, they highlight that violence that may be perceived as random is particularly provocative. On the macro level evidence also supports the positive and the negative view. Johnston (2009) provides a quantitative study with support for the positive view and Schubiger evidence for the negative view (Schubiger, 2014). The contradictions may be explained by focus on specific types of warfare and levels of aggregation. A possible interpretation is that indiscriminate violence is more useful against guerrillas, which is the argument of Valentino et al. (2004) as well as Arreguin-Toft (2001). However, it seems currently impossible to draw a reliable conclusion. Finally, the studies on the patterns of indiscriminate violence confirm that it is often not used randomly. On the contrary, the targeting of specific groups and specific areas seems to be the basis of a deliberate strategy. On the one hand, governments seem to use indiscriminate violence mostly when they have no access to information. They use it when they are far away from their home base (Schutte, 2017a), have low local control over the civilian population (Douglass, 2012; Kalyvas & Kocher, 2009; C. M. Sullivan, 2012), and when they are desperate (Valentino et al., 2004). On the other hand, indiscriminate violence is used as a response to rebel attacks (C. M. Sullivan, 2012) and specifically targets groups associated with the opponent (Fjelde & Hultman, 2014). All this seems to indicate that indiscriminate violence is less effective than selective violence, but it also confirms that indiscriminate violence is seemingly used as a deliberate and targeted strategy. While we cannot be certain about the targeting patterns there is some evidence that indiscriminate violence is targeted and used to punish certain behaviors. As a result the evidence presented by the reviewed studies on government violence does provide some limited support for the theory of groupselective violence. The only study that focuses on targeting groups finds a coercive effect, while additional evidence corroborates that indiscriminate violence is targeted and restrained. However, overall, it seems that the stud-

Assessing the Effects of Indiscriminate Violence

137

ies mostly did not look in the right locations or used study designs that made detecting the coercive effects difficult. Nongovernment Violence I will now turn to the empirical evidence on nongovernment indiscriminate violence. Similarly to the literature on government violence, studies on nongovernment violence find that indiscriminate violence can be effective in coercing as well as counterproductive. However, the study of nongovernment violence also adds a twist. As outlined in Chapter 3, nongovernment violence can also follow a provocation strategy. This means that I will discuss the intentional provocative effects of indiscriminate nongovernment violence, too. I will again distinguish the macro level and the micro level of analysis. Studies on nongovernment violence mostly remain at the macro level as this literature did not experience a general turn towards the micro level. Yet, the few microstudies that are available produce some of the most important insights. Macro Level Studies on Nongovernment Violence

Studies on the effects of coercion vary mostly in the way they measure the outcome of terrorism. In particular, political concessions, the onset of negotiations and effects on domestic politics have been used as dependent variables. On the one side, a good amount of studies find that nongovernment indiscriminate violence can be effective in coercing. An early study was put forth by Pape (2003), who investigated 11 suicide terrorist campaigns and focused on political concessions as their outcome. Regarding the effects of coercion, he found that terrorism is associated with moderate concessions in about 50% of the cases. However, Pape made two additional observations. On the one hand, he also found a positive temporal trend in the occurrence of suicide terrorism. He argued that rational organizations would only produce this trend if they expected terrorism to be working. On the other hand, he observed restraint among the terrorists’ usage of violence, a finding similar to that on government violence outlined in the previous section. Pape writes: The most important indicator of the strategic orientation of suicide terrorists is the timing of the suspension of campaigns, which most often occurs based on a strategic decision by leaders of the terrorist organizations that further attacks would be counterproductive to their coercive purposes for instance, in response to full or partial concessions by the target state to the terrorists’ political goals. Such suspensions are often accompanied by public explanations that justify the decision to opt for a “cease-fire” (Pape, 2003, 347).

138

How Does Indiscriminate Violence Work in Practice?

As a consequence, Pape’s comparative case study analysis draws a picture of rational organizations that follow the expected patterns of coercion and he finds some evidence for effective coercion. With regards to the effects of terrorism within civil wars, two studies in particular provide evidence that terrorist violence can have an impact on the probability of negotiations commencing or being successfully concluded. Thomas (2014) concluded that the amount of terrorism used by a rebel group in the previous month made negotiations with the government in the following month more likely and increased the probability of receiving concessions during the negotiations. R. M. Wood and Kathman (2014) provided a similar investigation, but focused instead on the successful conclusion of negotiations. They found an inverted u-shaped relationship between violence against civilians and negotiated settlements. Low levels of violence increased the probability, while high levels decreased it; a result similar to Johnston (2009). An additional line of research investigates the influence of terrorism on domestic politics with a focus on the effect of violence on government failure. However, these studies do generally not restrict their measure of terrorism only to violence against civilians, and their findings have to be taken with caution for the purposes of this book. First, Williams et al. (2013) concluded that left-wing governments had a heightened risk of ending prematurely after terrorist attacks while right-wing governments did not experience this effect. Second, Park and Bali (2017) found that terrorism in general led to decreased time in office for government leaders. However, they qualified their results in two ways. On the one hand, they observed that incidents with no fatalities better predicted leadership removal than attacks with fatalities (Park & Bali, 2017). This indicates that killing many people may not necessarily help in removing a leader. On the other hand, terrorism interacts with the political system, as it has a significant impact on leadership removal in autocracies, but not necessarily in democracies. Finally, Aksoy et al. (2015) empirically investigated if terrorism can lead to an increased risk of a coup. They distinguished between coups from insiders (reshuffling coups) and from outsiders (regime-change coups). They found that terrorist events were associated with a higher probability of reshuffling coups, but not regime-change coups. These findings suggest that terrorist attacks have an impact on government stability, but the impact may be contingent on the government in place as well as the attack strategy. Not everyone agrees that terrorist violence can coerce either a government into giving concessions or increase the probability of a government failing. In fact, recently terrorism becomes increasingly associated with losing and thus ultimately counterproductive. One of the most vocal proponents of this view is Max Abrahms. Abrahms (2006) investigated a dataset on 28 terrorist organizations and their

Assessing the Effects of Indiscriminate Violence

139

42 policy objectives, and found that these groups reached their goals only 7% of the time. In contrast, guerrilla tactics were much more effective.3 In a subsequent study and as a reaction to previous criticism, Abrahms (2012) assembled a larger dataset on 125 campaigns waged by 54 groups to see whether the usage of terrorism increased the probability of achieving political goals. Here, again he found that of the 128 terrorist campaigns, only one was successful while 36 campaigns of guerrilla warfare were effective in coercing the enemy into concessions. This holds after controlling for factors such as group strength, enemy strength, and extremeness of goals. Other researchers have found similar results. For instance, Jones and Libicki (2008) observed that terrorist organizations rarely win a campaign and that less than 10% of terrorist campaigns ended because the terrorists’ goals were achieved. However, they also concluded that terrorist campaigns often ended when terrorist groups integrated into the regular political process. This indicates that terrorist groups often seem to be capable of transforming their struggle into a nonviolent form. As a consequence, while terrorists may be unlikely to achieve their goals in full through coercion, they seem to be able to achieve some political progress. Looking at the impact of indiscriminate terrorism on the probability of rebels engaging in negotiations with the government, Fortna found that rebel groups that use terrorism as a strategy are less likely to achieve a negotiated settlement and less likely to win the war (2015). However, the use of terrorism increases the duration of civil wars which, Fortna argues, may indicate that indiscriminate terrorism may increase the chances of survival for terrorist organizations. In addition to these studies that compared larger samples of cases two authors have investigated the effectiveness of terrorism with in-depth case studies (Crenshaw, 1995; English, 2016). In her study of the impact of indiscriminate violence against civilians during the Algerian War of Independence (1954–1962), Crenshaw pointed out that terrorism had mixed successes as it led to military losses, but also to increased support from the population. For instance, following the battle for Algiers (1956–1957), the rebel organization Zone autonome d’Alger (ZAA) was completely destroyed in the wake of the brutal French repression. This same repression pushed the Algerian population more and more into demanding independence and increased support for the rebel organizations (Crenshaw, 1995, 493). Yet, Crenshaw concludes that “as a method of achieving radical political change, terrorism has no inherent efficacy. Indeed, from the strategic perspective it is often counterproductive” (Crenshaw, 1995, 509). She argues that in principle two problems make terrorism ineffective. First, as a communication strategy, terrorism cannot send a clear message. Second, terrorism makes it necessary to build decentralized secretive networks.

140

How Does Indiscriminate Violence Work in Practice?

This makes a strategic campaign of violence difficult and gives way to nonstrategic acts of vengeance. In addition, English investigated the success of terrorism along four case studies (al-Qaeda, the Provisional IRA, Hamas, and ETA). Contrary to other authors, he sought to differentiate between types of success: full strategic success, partial strategic success, tactical success and the inherent rewards of the struggle (English 2016, 30–38). He did not differentiate between targets of violence, arguing instead for the inclusion of both military and civilian targets (English, 2016, 11). Consequently, his results were also only partially relevant for this book. After studying the four cases and considering various other cases in his final chapter, he concludes that at least with regard to the main goals of terrorism it has been remarkably unsuccessful: [...w]hat has emerged is the profound uncertainty of terrorism achieving its central goals, together with a complex pattern of other successes and failures at lower level. What is almost certain (in fact, the historical record suggests it to be certain in all major terrorist campaigns) is that terrible human suffering will ensue from terrorist violence. [...] Every one of the case studies examined sustainedly in this book has involved considerable human suffering being caused; none of them has involved the achievement of the relevant group’s central goals (English, 2016, 265).

Apart from the verdict on reaching their main goal, English gives a very nuanced and detailed account that repeatedly points to the tactical success of terrorists. For instance, most terror organizations seem to have had success in provoking the government (English, 2016, 248–249). While the overall assessment thus may be true and terrorists are often not able to achieve their main stated policy goal through terrorism, English finds that terror can contribute to their struggle in various ways. As a consequence, the case studies indicate that apart from a potential coercive success terrorism is frequently able to achieve a provocative effect. The mechanism behind provocation is similar to the counterproductive effects of government violence. Yet, in the case of terrorism provocation may work in favor of the perpetrator; at least in the long run. Therefore, it is important to give provocation a deeper look. Piazza and Walsh (2009) investigated the impact of international terrorism on the respect for human rights. They found that terrorism increases disappearances and extra-judicial killings, but has no influence on political imprisonment and the practice of torture (although political imprisonment becomes significant when the measure of terrorism captures the number of victims). In addition, terrorism seems to be unrelated to changes in the freedom of speech, political participation rights and religious freedoms. This indicates that governments confronted with terror-

Assessing the Effects of Indiscriminate Violence

141

ism increase lethal repression, but do not generally reduce political rights (Piazza & Walsh, 2009, 144). Similar studies come from Robison (2009), Dreher, Gassebner, and Siemers (2010), and Shor, Charmichael, Nazif Munoz, Shandra, and Schwartz (2014). All find that terrorist violence can lead to human rights abuse, while it does not lead to a general reduction in political freedoms. Blankenship (2016) provides a different approach, testing the effect of terrorism conditional on the circumstances in which governments find themselves. He found that the effect of terrorism was conditional on the bureaucratic capacity of a state. Countries with less bureaucratic capacity seemed to show a stronger increase in rights violations than those with more capacity. In addition, the provocation effect vanishes completely for high capacity states. Finally, there is some evidence of the impact of terrorism on voting patterns that indicates a potential provocative effect of terrorism on the electorate. Studying the impact of terrorist attacks on voting preferences in Israel between 1990 and 2003, Berrebi and Klor (2006) found that increases in the amount of deaths from terrorism led to increases in the relative share for rightwing parties. The results hold regardless of whether a right-wing party is in office or not. Since right-wing parties are generally hawkish towards terrorist challengers, this shift can be seen as evidence of provocation rather than coercion. In sum these macro level studies on the effectiveness of nongovernment violence provide a contradictory picture of the effects of indiscriminate violence. Unfortunately, the study of terrorism is even less aware of the possibility of the countervailing effects of violence than the study of government violence. As a consequence, none of the studies differentiates between provocative effects and coercive effects; they are dealt with only in isolation. The fact that both coercion and provocation are potential effects seems to be at the center of some of the ambiguities of evaluating terrorism. Although this greatly complicates quantitative studies, it becomes especially apparent in the two case study approaches (Crenshaw, 1995; English, 2016), where both authors acknowledge that terrorism is able to provoke and effective in doing so. Still, they argue that terrorism is overall ineffective because it cannot coerce or directly reach the stated policy goals, and that there is no direct link between terrorism and the opponent giving in. Thus, they assume that the only relevant strategy of terrorism is coercion and disregard provocation as a specific strategy that might be successful in the long term. Yet, at least in Algeria, this strategy led to the desired policy goal: the independence of Algeria from France. In addition, on the macro level, in all these studies the authors remain oblivious to the terrorists’ targeting of groups and the specific audience of

142

How Does Indiscriminate Violence Work in Practice?

the coercive messages. This makes the interpretation of the results for the theory of group-selective violence very difficult. While often not explicitly stated the assumption seems to be that terrorism seeks to coerce or provoke the government and the population associated with it. In terms of group-selective violence this means that terrorism is a form of specific group-selective violence, a point I will come back to later. Specific groupselective violence, it was argued earlier, produces provocative and potentially coercive effects in the same group. Therefore, it is particularly difficult to disentangle the countervailing influences if one is to look only at the general outcome. Two issues in particular complicate the findings of terrorism research. First, the measurement of the effects often sets a very high bar for success. Most authors assess success of terrorism by comparing its achievements to its stated policy goals or strategic goals (Abrahms, 2006, 2012; Pape, 2003, 2005). However, this strategy of measuring impact is problematic and, therefore, often criticized (English, 2016; Krause, 2013; Marsden, 2012).4 For instance, Krause (2013) points out that policy goals are only one among many types of goals for terrorist groups. As all organizations do, terrorists will try to focus on organizational survival first and only later will they seek to achieve their strategic goals. In addition to Krause’s argument, at least three problems can be added that make the usage of strategic goals particularly bad for measuring success of coercion. 1. Terrorism can only reach limited goals by coercion. Coercion threatens punishment in the future and exchanges this potential punishment for a concession. This means that the value of the concession can never surpass the negative value of the potential punishment. Whenever terrorists are not overwhelmingly strong terrorism can only reach limited goals. 2. Stated policy goals are not necessarily the true goals. For instance, Abrahms assumes that it is possible to take the policy statements of terrorists as an indication of their true goals (Abrahms, 2012, 367). However, such an assumption seems unwarranted. The stated goals of all organizations serve purposes beyond providing accurate information for observers. Terrorists’ goals are more likely tools of mobilization and negotiation. Therefore, it will always be better publicly to seek high aims like independence than argue right away that one would possibly settle for less. 3. Finally, as outlined above, terrorism may also be meant to provoke, and this effect goes in the exact opposite direction. Studying the effects of terrorism with very aggregate measures may thus obscure its countervailing effects.

Assessing the Effects of Indiscriminate Violence

143

Recent studies have partially improved on these issues, using either the onset of negotiations or the granting of concessions during negotiations as a measure of success (Fortna, 2015; Thomas, 2014; R. M. Wood & Kathman, 2014). This approach has the advantage of evaluating terrorist success not only against the maximalist policy goals. However, in these cases the problem remains that terrorism may not be meant to coerce now. As Lake points out, terrorists that seek to provoke are not interested in negotiations now (Lake, 2002). Nevertheless, if we take a particular look at the three studies using negotiations as an outcome it is possible to conclude that the findings are in line with the predictions of the theory of specific groupselective violence. R. M. Wood and Kathman (2014) directly find the predicted u-shaped relationship for the successful conclusion of negotiations. Thomas (2014) finds a positive linear relationship for the onset of negotiations and thus also a coercive effect. Only the study by Fortna (2015) indicates that violence may lead to a reduced probability of negotiated settlements. However, the problem is that her measure of indiscriminate violence captures only “high-casualty-terrorism”; it is not changing over time and is only a binary variable. As discussed with government violence, binary time-invariant measures cannot possibly capture the difference between a year of low coercive violence and a year of high coercive violence which might lead to negotiations or other concessions. The measure will therefore confuse those rebels using violence with restraint and rebels using violence without control. In addition, the focus on high-casualty-terrorism may lead to finding only the negative results of high amounts of terrorism. This should caution us to take Fortna’s results at face value. In addition, as with the measurement of the outcome, the measurement of terrorism itself is often problematic. Most datasets cannot distinguish between selective violence and indiscriminate violence, yet the two strategies often follow distinct logics. Selective terrorist violence mostly targets the terrorists’ own constituency in order to increase control and decrease civilian collaboration with the enemy (Kydd & Walter, 2006). Indiscriminate violence is used as a strategy to coerce or provoke the enemy. Therefore, to test an argument of coercing the enemy, it seems necessary to focus mostly on indiscriminate violence. A good example comes from C. M. Sullivan (2012) who counted events with more than a certain threshold of deaths as indiscriminate. Another strategy is Fortna’s analysis (2015) of high casualty terrorist campaigns, in which she is able to separate indiscriminate terrorism from selective violence. Unfortunately, this measure is then aggregated over time to organizations using indiscriminate violence and thus time-invariant. Thus, while the efforts are not ideal in capturing indiscriminate violence as a coercive tool they are both an innovative way to deal with the fact that selective and indiscriminate violence are difficult, but necessary, to distinguish.

144

How Does Indiscriminate Violence Work in Practice?

Micro Level Studies on Nongovernment Violence

In comparison with the large amount of studies that seek to investigate the effects of terrorism on the macro level, studies on the effects of terrorism on the micro level are short in supply, but three studies give a very detailed picture of the effects of indiscriminate violence on civilians. Berrebi and Klor (2008) investigated the effects of civilian fatalities from terrorist attacks in Israel on the vote of right-wing parties in elections between 1988 and 2003. They found that every terrorist fatality significantly increases the share of right-wing parties in elections. In addition, the effect seems to stem from closeness to violence since the level of terrorism in Israel overall has a much lower effect on the voting behavior of a given locality. However, when only investigating left-leaning localities the effect changes. Although attacks still increase the rightwing vote the general level of terrorism decreases the relative vote share for the right-wing parties. Gould and Klor (2010) investigated the impact of terrorism on individual preferences of territorial concessions and voting. They found that increases of Israeli fatalities in a given district also increase the inclination towards territorial concessions among Israelis, particularly among rightwing voters. However, this holds only up to a certain point, after which fatalities reduce support. Fatalities also increase the right-wing vote. However, they argue that this does not contradict the previous findings since the increase in right-wing votes stems from a shift of the whole party system to the left. Overall, they conclude that terrorism can work, but that it may backfire once its magnitude reaches a threshold. Getmansky and Zeitzoff (2014) add to this research with a particularly interesting approach. By focusing on rocket attacks in Israel they were able to distinguish between localities that are threatened to be attacked (within the range of rockets) and localities that are not. They found that the threat of a potential rocket attack significantly increased the right-wing vote share. The authors were therefore able to show that it is the threat of violence that has an important influence. These studies circumvent most of the problems described for macro level studies. They focus on indiscriminate attacks against civilians and use outcome measures that are much more capable of capturing effects of provocation and coercion. As a result, these studies seem to find support for provocation as well as coercion. Getmansky and Zeitzoff (2014) provide a way to measure the effect of the coercive threat alone by focusing on the range of missiles fired from Gaza on Israel. Of course it is difficult to interpret voting behavior as provocation or coercion (which Getmansky and Zeitzoff do not do). However, if we argued that increasing the amounts of right-wing votes constitutes provocation, the findings would be at odds with our theoretical expectations. We would expect that a

Assessing the Effects of Indiscriminate Violence

145

threat would coerce and only violence would provoke. Yet the results contradict this idea. Gould and Klor (2010) provide results that may be interpreted as support for the theory of specific group-selective violence. They found that terrorism in a certain district increases the willingness to make concessions to the Palestinians in this same district. However, they also found that this effect may turn into the opposite if terroristic violence increases to very high amounts. This means they found the expected u-shaped effect of specific group-selective violence on the district level. Finally, Berrebi and Klor (2008) found an effect in line with the predictions of general group-selective violence. Terroristic violence in one district increased the share of right-wing votes in that district. This indicates a provocative effect as voters shift to a hard-liner position. However, the effect of terrorism overall (i.e., mostly in other locations) decreases the level of right-wing votes which means that indiscriminate terrorism may produce exactly the effects of general group-specific violence when we look at a disaggregated level of analysis. It increases resistance in targeted districts and decreases it in districts not targeted. The Effectiveness of Nongovernment Violence

The previous sections gave an overview over the most recent studies that provide insights into the effectiveness of indiscriminate nongovernment violence. However, for nongovernment violence as for government violence these findings give a mixed picture. Studies on the macro level are often unable to capture coercive or provocative effects because of their research design. Nevertheless, some observations can be drawn from this.

Macro level. On the macro level the general findings are mixed. Yet, at least part of these results seems to stem from the fact that there is no differentiation between provocative and coercive effects; the outcome measures are often setting too high a bar and the measures of terrorism cannot distinguish between selective and indiscriminate types of violence. However, if we focus on those studies of nongovernment violence that circumvent some of these problems (e.g., negotiation as outcome), the evidence tilts to the side of effective coercion. Given that terrorism can be expected to fall in the category of specific group-selective violence, we would expect a coercive effect of violence only for low amounts of terroristic attacks while higher amounts would lead to more provocative effects. This is in fact what Kathman and Wood (2014) find. Thomas (2014) does not test for a potential curve-linear relationship but finds a positive effect of terrorism on the onset of negotiations between governments and terrorist groups. Only Fortna (2015) does not find such an effect. However, as pointed out, her

146

How Does Indiscriminate Violence Work in Practice?

measure of terrorism is time-invariant and only a binary variable indicating the use of indiscriminate terrorism. This makes it impossible for her to capture the effects of different amounts of terrorism or effects over time. In addition, her focus on high-casualty-terrorism makes it likely that what she is finding is the right-hand part of the curve-linear relationship where more terrorism decreases the chances of a negotiated settlement.

Micro level. Micro level studies provide a clearer picture of the effects of terrorism, although these are not all in line with the expectations of the theory of group-selective violence. Particularly, Getmansky and Zeitzoff (2014) seem to find provocative effects where the theory of group-selective violence would expect coercive effects. To the contrary, the findings of Gould and Klor (2010) and Berrebi and Klor (2008) support the theory of group-selective violence. Gould and Klor (2010) find a curve-linear relationship between fatalities and the willingness for concessions in the same district, as specific group-selective violence would predict. In addition, the findings of Berrebi and Klor (2008) support the predictions on general group-selective violence, as violence seems to have provocative effects in the targeted districts, but overall levels of violence have coercive effects. In sum, these findings of research on nongovernment indiscriminate violence are surprisingly aligned with the theory of group-selective violence. Both the inverted u-shaped relationship of specific group-selective violence and the deterrence effects of general group-selective violence can be found.

Conclusion In this Chapter 1 discussed studies which either analyzed the patterns or the effects of indiscriminate violence. Superficially, these studies give the impression of contradictory results. Findings on government violence as well as on nongovernment violence support both the potential coercive effectiveness of indiscriminate violence as well as its ineffectiveness or counterproductiveness. However, this chapter also discussed some of the potential shortcomings, in order to arrive at a clearer picture. Among others I outlined that the research design of many studies is unable to capture the potential coercive effects. I argued that particularly macro level studies have difficulties in capturing coercive effects because they either do not disaggregate the time axis, or use outcome variables that set an extremely high bar for success. Micro level studies fare better in their ability to find coercive effects but sometimes also suffer from the use of ambiguous measures of indiscriminate violence. Nevertheless, in the light of the theory of group-selective violence several important corroborating results can be identified and these findings

Assessing the Effects of Indiscriminate Violence

147

come particularly from micro level studies. First, most studies focusing on violence that could be interpreted as collateral damage or random violence, found strong evidence for provocative effects. Second, the patterns of violence which were identified are mostly in line with group-selective violence: indiscriminate violence is used to target groups and violence is tied to certain behaviors like rebel attacks. Third, whenever studies are designed to capture the effects of general group-selective violence, the coercive effects can be found for government (Benmelch et al. 2015) and nongovernment violence (e.g., Berrebi and Klor 2008). These studies seem to provide particularly strong support for the theory of general group-selective violence. Finally, the theoretical expectations on specific group-selective violence also found some support. For terrorism, which I consider to constitute mostly a form of specific group-selective violence, studies seem to support the inverted u-shaped effect of increasing amounts of violence. Nevertheless, some findings remain at odds with the theory of group-selective violence. Particularly the results of Getmansky and Zeitzoff (2014) seem to directly contradict the theoretical expectations. Overall, this chapter highlighted that the existing evidence may be less contradictory than a superficial view may suggest. The findings of several studies can be interpreted as corroborating parts of the theory of groupselective violence. Nevertheless, further evidence is needed to increase our confidence in the theory. In the next two chapters I will therefore discuss several cases and provide a quantitative investigation of ethnic wars in order to further corroborate the hypotheses outlined in Chapter 5. Notes 1. Mass killings are defined as 50,000 intentional deaths through government violence in less than 5 years (Valentino et al., 2004, 378). 2. Kocher, Pepinsky, and Kalyvas make a similar point (Kocher et al., 2011, 204). 3. For a critique of Abrahms work see (Chenoweth et al., 2009; Krause, 2013). 4. The criticism on Pape’s work has been especially vocal (Ashworth, Clinton, Meirowitz, & Ramsay, 2008; Krause, 2013; Moghadam, 2006).

7 Group-Selective Violence Across History

As outlined in the previous chapter, current empirical studies do not investigate indiscriminate violence along the lines of group targeting. In particular, general group-selective violence is often overlooked. As a consequence, most studies are unable to capture the full spectrum of coercive effects which group-selective violence can have. In this chapter I will provide anecdotal empirical evidence on the existence of general group-selective violence as a strategy and its effects. I will concentrate particularly on general group-selective violence because the previous theoretical considerations suggest that the coercive effects should be the strongest with this strategy. The case studies I provide in this chapter thus serve as a plausibility probe. With unestablished theories it makes sense to investigate how far they can be expected to work in the way outlined (Eckstein, 2000, 140–142). Although the theory of the effects and the usage of general group-selective violence has some plausibility to it, it needs to be established in how far this strategy is actually used. The cases were selected in order to illustrate best the usage of general group-selective violence. One intention was to gather cases in which the perpetrators were willing to explain the rationale behind their violence. This can seldom be expected to be the case, since the deliberate usage of violence against innocent civilians is under most circumstances deemed to be morally unacceptable. These moral restrictions certainly became even more significant since the targeting of civilians was outlawed by the Hague Convention of 1907 and the Geneva Convention of 1949 (Valentino et al., 2006). However, in some historical episodes, and especially throughout colonial history, we find perpetrators explaining their deeds with confidence.

149

150

How Does Indiscriminate Violence Work in Practice?

In addition, the selection of cases is intended to give an indication of how prevalent this type of strategy was, and still is, in warfare. Cases of general group-selective violence can be found from ancient times to this day and it involves various types of groups. This in turn should also highlight the urgent need to better understand it. Since it is still part of the repertoire of some authoritarian counterinsurgents, it is necessary to realize its benefits for the perpetrator in order to design effective countermeasures. Siege Warfare The logic of both types of group-selective violence is perhaps best encapsulated in the way sieges were fought throughout the ages. 1 Since the first cities were fortified, conquering them was a time-consuming and difficult enterprise. Besieging a city made the army vulnerable to counterattacks and strained the logistic capacities for supplying the army. In order to circumvent such problems, many strategists have thought of ways to shorten sieges and it seems that already early on generals realized that the targeting of the civilian population of a city may be one such instrument. Interestingly, violence against civilians during sieges took two distinct forms that reflect the categorization of specific and general group-selective violence, although these are often not distinguished as separate logics of civilian coercion: 1) violence could be directed against a city to make it surrender quickly and 2) violence could be used against those cities that were captured in order to set a terrifying example for other cities. Decreasing the length of a siege was a primary concern for all besieging armies, and history books are full of tactics on how to do so. One of these was to bombard the whole city indiscriminately. Besiegers then attacked with artillery or whatever weaponry they had at hand in order to make life in the city increasingly intolerable. In addition, this violence was often deliberately directed against the civilians of a city, in order to increase their inclination to surrender. Since coercive violence in this case is directed at one group (in this case the city) in order to make this same group behave in a certain way (i.e., surrender), this type of targeting corresponds to specific group-selective violence. One among many examples for this type of specific coercion is the Siege of Paris from 19 September 1870 to 28 January 1871 during the Franco-Prussian War. After winning several battles, Prussian troops surrounded the remainder of the French army in Paris. With the beginning of the winter, Prussian troops suffered from sickness and limited supplies so that the political and the military administrations sought for ways to bring the war quickly to an end. While Field Marshal Moltke argued for prioritizing fighting French forces outside Paris, Chancellor Bismarck wanted to

Group-Selective Violence Across History

151

increase French suffering in Paris in order to force the Government into submission. Ultimately Bismarck got his way and Paris was indiscriminately shelled from January 1871 on, leading to hundreds of civilian casualties (Wawro, 2005, 278–282). The rationale behind shelling Paris was, therefore, to hasten the surrender and free German troops from a lengthy siege. In the end Paris surrendered but the effect of shelling the city on the outcome is disputed among historians. Another tactic was for the besieging army to target civilians after the siege was won. History is full of gruesome examples in which the attacker took the city by force and subsequently killed either just the male population or even all civilians. For example, the sack of Jerusalem by the Christian army during the first crusade in 1099 as well as the sack of the Protestant city of Magdeburg by imperial troops in 1631 are still remembered today for their extreme brutality. In both cases, however, the killing of civilians did not help in shortening the siege, as the city was already taken. While some of the atrocities may have been committed in order to let the ordinary soldiers loot and satisfy their bloodlust, others served a distinct coercive purpose. By killing civilians of cities that had not surrendered quickly but defied the attacker, the attacker could signal to other cities that defiance would be punished by death (Waxman, 1998, 360–368). As a result, a quick surrender became the only option for surviving when there was no hope for relief. In fact, as Waxman (1998) points out, the customary law of siege warfare from the late Middle Ages on granted immunity to cities that did not resist a conquering army. However, once an army was forced to lay siege to a city, this sort of lenient treatment could no longer be expected. If the city was taken by force, the attacker had the right to do with it as he pleased. As a result, this practice of exercising cruel violence against civilians often had the effect of scaring other cities into a quick surrender.2 One of the earliest traces of this type of targeting civilians can be found in the Bible.3 In the book of Deuteronomy the Hebrews receive two specific orders from god with regard to the cities they conquer. For the cities in the Promised Land the Hebrews should kill everybody since they are supposed to inhabit this land and erase the faiths of all others. However, for cities outside the Promised Land the rule differed: When you draw near to a city to fight against it, offer terms of peace to it. And if its answer to you is peace and it opens to you, then all the people who are found in it shall do forced labor for you and shall serve you. But if it makes no peace with you, but makes war against you, then you shall besiege it; and when the LORD your God gives it into your hands you shall put all its males to the sword, but the women and the little ones, the cattle and everything else in the city, all its spoils, you shall take as booty for yourselves; and you shall enjoy the spoils of your enemies, which the

152

How Does Indiscriminate Violence Work in Practice?

LORD has given to you. Thus, you shall do to all the cities which are far away from you, which are not part of the nations here (Deuteronomy 20.10 cited in Kern, 1999, 62).

Killing civilians after a siege was common practice in the ancient world and part of a strategy of intimidation that was supposed to minimize the amount of time-consuming sieges and make conquests easier. As Saggs points out, Assyrian atrocities were designed as a form of psychological warfare that had the explicit intention of reducing the length of sieges (Saggs, 1963, 149). In addition, the fact that these atrocities can still be found today on bas-relief sculptures tells us that the perpetrators tried to communicate the message directly to the intended audience (Kern, 1999, 69). Violence was not an emotional reaction after a long and demanding siege, but it was meant as a message for others who would potentially oppose them in the future. Also, indications of the effectiveness of this type of coercive communication can be found repeatedly. For instance, in a report on a campaign of the Assyrian king Shalmaneser III (859–824 BC) we learn that when he reached the city of King Giammu, the people were so terrified by the prospects of resistance that they killed Giammu and handed over the city to Shalmanser (Kern, 1999, 69). Another example of killing civilians after sieges comes from the campaign fought by Fernando Alvarez de Toledo, Duke of Alba (Waxman, 1998, 364–368). Alba became the governor of the Netherlands in 1567 and was charged with subduing the revolt that had broken out against the Spanish king, soon before. Until today Alba’s name is remembered mostly for his brutal regime in the Netherlands. However, it is less often remembered that part of the violence followed a strategy of deterring opposition. For instance, Waxman describes Alba’s strategy as very similar to what I have outlined above as general group-selective violence: Alva turned to a strategy of terror, the application of brutal violence against a portion of enemy territory in hope of inducing surrender of the rest. As Alva himself would later plead, following the sacking of one rebellious town, “God grant that the rest may learn from it and that it will not be necessary to carry with them to the end and go from town to town with the army of your Majesty” (Waxman, 1998, 365).

Alba started a campaign against the rebellious cities in Brabant in early 1572 and the first city he arrived at was Mechelen just to the North of Brussels. The city refused to admit his troops after his son’s arrival. Instead, it seems, several shots were fired at the Spanish. The garrison, frightened by the overwhelming Spanish military might, left Mechelen in the following night, rendering the city defenseless. The citizens who saw that all hope was lost for military resistance, approached Don Frederic in

Group-Selective Violence Across History

153

the morning with a procession in the hope of calming the Spanish. However, the city was sacked immediately. The destruction was so absolute that a Spaniard wrote later that “hardly a snail was left standing in the walls” (Motley, 1929, 492). While the annihilation of Mechelen revolted observers, it certainly produced the intended effect. “There could be no doubt that his policy was working. Louvain and Termonde had surrendered on merely hearing of his threats against Mechelen, and when the news of his terrible vengeance spread beyond the confines of Brabant other towns followed suit” (Maltby, 1983, 240). Alba repeated this show of force at Zutphen which led to the same results. “City after city sought to avoid the rape, killing, and arson that had accompanied the sack of brethren cities by taking advantage of the waivable protection afforded them by legal custom” (Waxman, 1998, 366). It is important to stress that the strategies of the Hebrews, the Assyrians, and the Spanish are just a few examples of a very much wider practice. Especially in those wars that saw large numbers of conquests, massacres can be found as a coercion strategy. For instance, Oliver Cromwell seems to have had a strategy of coercion in mind when he sacked Drogheda at the beginning of his conquest of Ireland in 1649 (Burke, 1990, 14). In addition, the destruction of Corinth by the Romans in 146 BC (Derow, 1989), and many of the massacres after sieges in the Peloponnesian War (Seaman, 2013), may be interpreted as exemplary forms of punishment with the aim of coercing. Although cities were a natural target throughout much of ancient, medieval and modern history, under different circumstances other groups would be targeted. Colonial Control and Violence Against Tribes and Ethnic Groups Some of the most pervasive examples of general group-selective violence can be found in colonial conflicts. The reason is twofold. First, colonial powers were frequently confronted with situations where territories lacked a monopoly of power. Often multiple groups, such as tribes or ethnic groups, remained powerful and from time to time rebelled against the colonial government. Second, colonizers were less reluctant to use violence against civilians in their far-flung colonies than at home. For instance, in the Middle East and in North Africa, British and French colonizers had to deal with myriads of Arab and Berber tribes that either kept plundering transportation lines and settler villages or engaged in outright insurgencies (see e.g., Jacobsen, 1991). It is therefore not surprising that one of the best documented cases of general group-selective violence can be found in Algeria in the mid-19th century, early after the French conquest.

154

How Does Indiscriminate Violence Work in Practice?

French Algeria in the 19th Century

The French conquest of Algeria began after the Dey, the regent of Algiers, whipped a French representative with a fly whisk. The French government used this incident as a pretext for a blockade and ultimately for an invasion. Expedition forces landed west of Algiers in June 1830 and conquered the capital in July of that year. The French military campaign against the Dey was won within several months. However, it proved very difficult to fully occupy the countryside outside the major urban centers. Soon after the French arrival, Emir Abd el-Kader was able to organize resistance among the western Algerian tribes and launched a military campaign against the French. After the French won the battle of Sikkak on 6 June 1836 and decisively beat Abd el-Kader’s forces, the Algerian tribes resorted to guerrillastyle attacks. Although Abd el-Kader was captured in 1847 and sent into exile, it took the French several decades to completely subdue the Algerian colony (Bouche, 1991, 21–34). As a consequence, from the mid-1830s on, the French colonizers were involved in a protracted guerrilla war and soon realized that it was necessary to adapt to the circumstances. Irregular warfare was certainly not new to the French Army. Under Napoleon the French had fought a war against partisans on the Iberian Peninsula from 1807 on, and during the French Revolution the army had fought against rebels in the Vendée and against the so-called Chouannerie. Yet, in Algeria, the first years of guerrilla war after the battle of Sikkak were less than satisfactory. It took until the arrival of Maréchal Thomas-Robert Bugeaud as governor-general in order to change the French way of fighting. Bugeaud was a French army general and member of parliament who arrived in Algeria in late 1840 (A. T. Sullivan, 1983). He was charged by the government with the full conquest of Algeria and instantly introduced multiple changes to the military regime. Among others he drastically increased troop size, asked for younger and more flexible recruits and introduced a system of flying columns that were able to rapidly strike if a tribe attacked (d’Ideville, 1882a, 260). However, since it was quite difficult to get hold of the tribal fighters who retreated quickly into the rough terrain, Bugeaud began to institutionalize a tactic of punitive raids which came to be known as the razzia.4 Punitive raids were part of French tactics from the beginning of the colonial conquest of Algeria. As the Algerian horsemen often remained elusive and unwilling to engage in a battle, the raids were supposed to target the interests of the tribes. Much like the raids of the Algerian tribes, the ghaziya, French raids, therefore, aimed at hurting the personal and economic interests of a tribe by appropriating livestock or goods and even killing men, women and children.5 Early reports on such raids can be found immediately after the French conquest and the term razzia is used from

Group-Selective Violence Across History

155

1833 on (Gallois, 2013, 92). However, it was not until the arrival of Bugeaud that the razzia became a systematically employed tactic. Bugeaud set up an encompassing military system that was able to respond with punitive razzias across much of the Algerian territory. The main aim of this punishment tactic was to clearly signal to all tribes that rebellion and raids on French interests could lead to the death of all members of their tribes (Gallois, 2013, 97). The rationale can be found among others in one of Bugeaud’s remarks from 1843: Going against the Zerdeza and the other tribes who lie close to our line of communications between Bône and Constantine and between Constantine and Philippeville, we must destroy [écraser] them through the superiority of our forces in order to make a real example of them, to intimidate those who would go against us, whilst consolidating the loyalty of those who have submitted (cited in Gallois, 2013, 102).

As a consequence, the system of the razzia as it was designed by Bugeaud was a form of general group-selective violence. Punishment was meant to send a coercive message to those tribes who might oppose the French. However, in order to make the razzia effective it was indispensable to have intelligence on who was responsible for a raid, which tribes were joining the rebellion and so forth. This was the role of the so-called bureaux arabes. The bureaux arabes were initiated in 1833, but were made full use of only under Bugeaud. Under Bugeaud’s supervision, the bureaux arabes became liaison offices with the Arab tribes. They gathered intelligence, were responsible for the surveillance of markets and mosques, and were sending off relevant information to the general government. In addition, they regularly had at their disposal a small indigenous cavalry force, the goum, which could be used for quick punishment raids or razzias (Rid, 2009, 625–628). The bureaux arabes and the razzia were thus two sides of the same coin. As Rid writes: “The raids punished uncooperative tribes while the indigenous affairs officers rewarded cooperative ones for those yet undecided, the razzia was meant to deter while the bureaus, designed to increase security and prosperity, provided an incentive” (Rid, 2009, 632). Typically, a razzia was designed to attack the tented camp of a tribe in order to gain prisoners, mostly women and children since the men were normally killed, and appropriate the livestock and possessions of that tribe. While the razzia was not always designed to kill civilians in large numbers it very often led to casualties. For instance, in a rather moderate example of a razzia recounted by one of the participants, Pierre de Castellane, the attack left several dead, among others the mother and father of a small girl who then accompanied the French (Castellane, 1856, 132–134). However, usually the razzias included the killing of all men and at times they could also lead to

156

How Does Indiscriminate Violence Work in Practice?

enormous amounts of violence against women and children. Perhaps the most notorious single incident, took place on June 18th 1845 in Dahra, when Colonel Pelissier chased down a tribal group until they retreated into a cavern in the mountains. Pelissier tried unsuccessfully to persuade the group to surrender and leave the cavern and several of his negotiators were shot dead. As a consequence, he decided to light a fire in front of the cavern with the aim of asphyxiating those inside. In the end, according to one account, this one incident alone left more than seven hundred people dead, many of them women and children (Pitois, 1846, 441). After the incident, the general government made clear that the deaths of the so-called enfumades were in fact in line with the French government policy of relentlessly punishing recalcitrant tribes, and similar events followed.6 Nevertheless, foreshadowing the discussion of French torture in Algeria a hundred years later, the massacre at Dahra led to a huge outcry in Metropolitan France that was highly discrediting to the army. Yet, even this political backlash did not end the practice and the razzias were used until the end of the pacification of Algeria in the 1870s. With regard to the effectiveness of the razzia as a tactic, contemporary observers were convinced that it was very successful in repressing the tribes of Algeria. After Bugeaud set in place his new system in 1841, the number of official submissions by the Algerian tribes rose dramatically. Already by the end of 1842 the success of Bugeaud was visible and the Moniteur Algérien, the official French Algerian newspaper, wrote: We have found a more efficient way than burning crops: waging an incessant war which impacts the population through individuals and in all their interests. The flights, the continuous alarms, the enormous losses inflicted by the razzias and even by mere relocations, the women and children we captured; the old, the women, the children and the herds who perished from fatigue and hunger; the necessity to live the entire winter in the harshest mountains, on summits covered with snow—that is what for better or worse pushed the Arabs into submission (cited in Rid, 2009, 624).

In addition, Bugeaud himself was convinced that even the harshest forms of punishment were effective and necessary in subjugating the Algerian tribes. For instance, after Colonel Pelissier’s enfumade at Dahra caused the highest controversy in Metropolitan France, Bugeaud defended the strategy of the razzia and the killing of civilians in an article in the Moniteur Algérien. There he pointed out that it was of the utmost importance to not give the tribes reason to believe that they could have a safe haven to retreat to. Only in this way could the coercion of other tribes be ensured and only in this way could further bloodshed be avoided in the future (d’Ideville, 1882b, 21–23). On a smaller scale, we can find episodes that show the immediate results of single raids. For instance, in June 1842 Colonel Brabaud reported to Bugeaud about a raid he had undertaken in the valley of Ouerz. Brabaud

Group-Selective Violence Across History

157

reported having seized cattle, burned houses and taken prisoners. While he released the prisoners subsequently he made clear he had sent a strong message to the surrounding tribes (this was an often used euphemism for killing people, see Gallois, 2013): unless they did not submit he would burn down all villages of the tribes around him. As a result, during the following days the tribes in his vicinity sent goods and horses to him as signs of their submission (Gallois, 2013, 114–115). While most of these accounts come from French official and military sources and may be treated with some caution, the same sorts of observations were made by more or less independent observers who joined the French army (Rid, 2009). For instance, Dawson Borrer, a British writer who joined French troops in a campaign against Kabyle tribes in 1847, wrote about a similar situation: “Some chiefs of the neighboring tribes presented themselves at the camp, both this day and the preceding; for the disasters which had befallen the Beni Abbes served somewhat to allay their intentions of resistance” (Borrer, 1848, 140). In addition to the usage of a strategy of group-selective violence, it is possible to find evidence that the French colonial army tried to communicate directly the message to the intended audience. In one example, some of the women that were captured in a previous razzia were sent back with the message that not submitting to the French colonial rule would lead to complete destruction (Gallois, 2013, 122–123). In other instances, Bugeaud sent out proclamations to the Kabyle tribes in order to inform them that future resistance would be met with annihilation (Gallois, 2013, 125). In some cases, the communication could be even more barbaric. In March and April of 1843 General Baraguey d’Hilliers undertook an expedition against the Erdough and killed their leader Zerdoud. In order to demonstrate to the surrounding tribes the French superiority, Baraguey d’Hilliers had the head of Zerdoud cut off and put on display at his camp for some time (Gallois, 2013, 130–132). The massacre at Dahra is indicative of a larger problem that the French army faced in Algeria. For violence against the tribes in Algeria to be effective in creating a coercive effect, it was necessary to publicize instances of violence. It was necessary to communicate to the surrounding tribes that the punishment they could expect for rebelling was harsh. At the same time, it was equally necessary to conceal violence from the Metropolitan public who perceived it as utterly barbaric and irreconcilable with the French mission to civilize (mission civilisatrice, Gallois, 2013, 94). Over time, the razzia seems to have developed into harsher and even exterminatory attacks against these tribes (Gallois, 2013, 118–121). Thus, while the razzias were selective on the tribe level, they were arbitrary and even exterminatory on the level of the individuals of the targeted tribe. This shift may reflect a response to the trade-off outlined above. Violence

158

How Does Indiscriminate Violence Work in Practice?

against one tribe has provocative effects and coercive effects. Completely eliminating the target of an attack reduces the provocative effects to a minimum and, in fact, the French administration was keenly aware of the negative effects of collective punishment. The war minister, Jean-de-Dieu Soult, wrote repeatedly to individual officers that he disapproved especially of burning trees because this created “irreconcilable enemies” (Gallois, 2013, 111). However, as Gallois points out, this disapproval did not mean that he would not support the razzia as a tactic. In sum, the razzia corresponds very well with the theory of groupselective violence outlined in Chapter 5. The razzia was a system of violent punishment, designed to be selective on the tribal level but killing arbitrarily on the individual level. As Bugeaud himself pointed out, its main intention was to communicate with those that would potentially oppose the French in the future. Finally, there are some indications that group-selective violence had the intended effect of making tribes submit to French rule. However, especially with regard to the effects, it has to be clearly stated that what is presented here is not systematic evidence. Few historians would argue that the razzia was effective. For instance, Gallois emphasizes the provocative effects of the razzia and concludes that it was ultimately self-defeating (Gallois, 2013). Potentially this assessment is due to the fallacy of only investigating the targeted groups. For instance, Gallois cites as one example for the futility of the tactic Bugeaud’s expedition to Dahra, where the general burned several villages in 1842. Gallois argues that while Bugeaud wanted to send a strong message the region around Dahra became a hotbed of resistance in the years to come, which ultimately led to the enfumade mentioned earlier (Gallois, 2013, 112–113). This way, Gallois assumes that the aim of the razzia is specific coercion (coercive violence and coercive message have the same target), when in effect the aim was general coercion (coercive violence and coercive message have different targets). As a consequence, serious doubt is in order when considering the assessment of authors like Gallois and more research would be needed to ascertain this claim. Additional Colonial Conflicts

The case of French Algeria is especially well documented. However, even in this case French reports are full of euphemistic descriptions of the attacks the French undertook. For instance, French reports seldom described in detail the violence of a razzia but used terms as “severe punishment” or “serious actions” that had been taken (Gallois, 2013, 97–98). Thus, even when the French military leadership was comfortable in defending acts of violence against the tribes, it is very difficult to know what that actually meant on the ground. This is even more the case in other colonial

Group-Selective Violence Across History

159

conflicts, where the reluctance of the colonizers to describe and explain violence was even greater. Nevertheless, it is possible to find indications that massive violence was used in other conflicts. For instance, the United States fought a guerrilla war against insurgents in the Philippines (1899–1902). As Johnston (2009) describes, the US began with a hearts and minds strategy that sought to convince the local population to support the US. However, after this failed to provide results, the military enacted severe punishment for the civilian population, including collective punishments for civilians living close to the rebellion (Johnston, 2009, 156–158). While collective punishment mostly took the form of property destruction, the war against the Moro insurgency on Mindanao (1899–1913) was more violent, as US Major General Leonard Wood resorted to a similar tactic to that of the French in Algeria. Going on punitive expeditions against resistant Moro clans, he tried to deter others from resisting. These efforts bore some fruit. For instance, Byler notes about Wood: “He and other officers expressed satisfaction with the results of these devastating campaigns. As a result of punishing one group of Moros, other groups that had been ‘lukewarm and hostile’ were inclined to submit to the Americans. Wood’s campaign effectively ended large-scale resistance by the Moros on Mindanao” (Byler, 2005, 43). In addition, British counterinsurgency in the 19th and early 20th centuries seems to demonstrate that the same logic of general group-selective violence was very often used. To begin with, the ideas of targeting groups certainly found praise among military strategists. For instance, Sir Charles Callwell in his Small Wars: Their Principles and Practice explicitly mentions Bugeaud’s campaign and tactics as a role model for fighting guerrilla wars and argues for its effectiveness (Callwell, 1906, 128–129). The book, which was originally published in 1896, remained an important source of information for young officers in the British army until the beginning of World War II (Moreman, 1996, 109–110). Since the principle of group-selective violence was deemed effective against tribal insurgencies, it is not surprising that punishment strategies seemed to have played a large role in British colonial security. For instance, punishment strategies against tribes and ethnic groups were used in policing the empire’s frontier in India. To name just one example, British troops in the Peshawar district undertook 15 punitive expeditions (which were commonly known as “butcher and bolt” expeditions) against tribes across the border between 1849 and 1855 (Moreman, 1998, 8–9). Even Winston Churchill participated in some of these attacks on local tribes (Toye, 2010, 39–40). The same methods were used in other colonial conflicts fought by the British. During the inter-war period, many of the revolts in the British Empire were met with punitive measures against civilians, as in the 1920 revolt in Iraq (Jacobsen, 1991). Particularly interesting is the development

160

How Does Indiscriminate Violence Work in Practice?

of a strategy of air policing. In order to make security as cost effective as possible, the air force was used to enforce control in “uncivilized” areas of the Empire (Killingray, 1984, 432). In one example, the Royal Air Force attacked Nuer tribes for staging a rebellion in Sudan in 1927 and 1928, killing several hundred people and their cattle (Killingray, 1984, 437). Over time this strategy developed into one of using only a minimum of violence. Portal (1937) gives a detailed account of an instance of simultaneous specific and general group-selective violence. In Aden, where a military presence was unjustified, Portal argues that air policing had tremendous effects on the local tribes. For instance, after one tribe attacked a caravan in 1934, the Royal Air Force used a strategy of punishment that relied on aerial bombardment, but sought to minimize casualties. First, the British issued an ultimatum to the responsible tribe, detailing exactly the demand to pay a fine and the threat that the tribe would be forced to leave its village or otherwise be bombarded. This ultimatum was given to the Sheikh personally and then dropped as leaflets on all major villages. After the Sheikh did not respond the main village was bombed although only in a very limited way since the tribe had left for the hills. In this case, as Portal points out, the main threat of air power was to kill if the tribe returned to the village (he calls this an “inverted blockade”), not to kill at all cost. After a period of being blocked for two months from entering their villages, the tribe gave up and agreed to pay a fine, although it had lost only three tribesmen. Portal argues that this kind of punitive action produced a minimum of provocative effects, had the intended coercive effect on the tribe that was punished and had an additional moral effect on other tribes not implicated in the punishment (Portal, 1937, 351–355). This example demonstrates how the threat of violence is the important factor in the effectiveness of coercion and indicates what might be needed for specific group-selective violence to work. The inverted blockade places the decision to be bombed on the tribe and it can avoid doing so. Therefore, the effective punishment is not to bomb but the unpleasant situation where the tribe is unable to return to its village. Punishment is thus (largely) nonlethal and open for immediate reversal by the tribe, should they opt to pay the fine. As a consequence, it seems that the punishment can avoid all provocative effects. However, it seems to describe a very particular situation where such a blockade is possible. Certainly, the tribes in Algeria were not sedentary and were able to move away with their cattle. This would have made such a blockade impossible. Targeting Villages in Counterinsurgency The first two sections discussed examples of targeting cities mostly in premodern times and tribes in colonial wars. More recently in civil wars, and

Group-Selective Violence Across History

161

especially in guerrilla wars, the focus of violence has been villages. In these cases, an actor targets villages that are suspected of collaborating with the enemy. Typically, the government’s side uses violence against whole villages as a reprisal for some previous incident and the severity of reprisals ranges from a few people killed to the total annihilation of the village. German Antipartisan Warfare in World War II

Perhaps the most notorious example of this type of targeting is the Wehrmacht’s antipartisan warfare in World War II.7 Soon after the start of the war in 1939, the German army became confronted with partisan warfare in multiple theaters, including Yugoslavia, Ukraine, Belorussia, France, and Greece. Consequently, the Wehrmacht enacted a harsh policy of reprisal that guided commanders to ruthlessly kill civilians in revenge for every attack on German soldiers. The Wehrmacht carried out two different types of reprisals: the shooting of hostages and reprisals directed specifically against villages. In the first type, hostages were taken from different parts of a given country. For instance, in Yugoslavia in October 1941 the commanding general Bohme ordered that the armed forces “were to round-up nationalists, democrats and communists as hostages and these hostages were to be shot in the event of anti-German action” (Fattig, 1980, 183). The second type originated on the ground directly in response to attacks. Here German officers often ordered a response against villages they suspected of collaborating with the partisans. In terms of group-selective violence, it could be argued that the shooting of hostages constitutes an instance of specific group-selective violence, while the targeting of villages constitutes general group-selective violence. The shooting of hostages threatens to kill specific people from a national group if other people from the same group engage in an unwanted behavior. However, this type of violence contains a great distance between actions and those threatened by a reaction. In addition, the German practice of hostage taking was directed at people that the National Socialists were inclined to kill in any case, as the selected particularly political and ideological opponents. The coercive purpose and effects are therefore questionable. This was different with reprisals of the second type, where German military commanders ordered reprisals often in direct response to partisan activity and with the aim of sending signals to the surrounding villages. As a consequence, these reprisals constituted a form of general group-selective violence. Since the aim of this chapter is to exemplify the usage of general group-selective violence the focus will be on the targeting of villages. A typical case was the reprisal against the Ukrainian village of Baranivka in November 1941 which was briefly introduced in Chapter 1 (T. Anderson, 1999). Starting in June 1941 the German army attacked the Soviet Union and advanced into its territory. By the end of September Kiev had fallen into

162

How Does Indiscriminate Violence Work in Practice?

German hands and with it most of Ukraine. However, not long after the German advance, resistance against the occupiers began to take shape, especially in the eastern part of Ukraine, where the Soviets had had more time to prepare for the ensuing conflict. As a consequence, the rear of the Wehrmacht was confronted with sporadic partisan attacks and several regiments were posted to eastern Ukraine in order to fight the upcoming insurgency. In this situation, on November 4th 1941 a German army vehicle that was on its way to Poltava passed through the district of Myrhorod. With night falling, the commanding officer decided to seek quarters at a house in the village of Baranivka. However, unbeknown to the German patrol, informers from the village reported the arrival of the soldiers to a partisan detachment that had its headquarters in the nearby woods. The partisans decided to take action and during the night a small group attacked the German soldiers in their sleep. They killed three of the four Germans, including the two officers (T. Anderson, 1999, 585–586). Five days later, on November 9 th the 11 th company of the 3 rd battalion of the 190 th infantry regiment arrived in Baranivka to investigate and punish the killing of the German soldiers. When they arrived, they encircled the village and rounded up all villagers. The German officer, allegedly called Hoffmann, interrogated the villagers about the incident and demanded that those with connections to the partisans were handed over to him, threatening that he would kill villagers indiscriminately if he was not given the names of the people. Yet, only one villager came forward and indicated a woman who was the mother of one of the partisans. She was shot instantly by the Germans. However, since Hoffmann was not pleased with the dearth of information coming from the villagers, he selected 10 male inhabitants and had them shot on the edge of the village. Afterwards the troops burned down the village of more than 500 buildings, leaving only a few houses as shelter for the winter (T. Anderson, 1999, 607–609). Subsequently, to make clear that this was not an isolated instance, the German district commander, Lt. General Friderici, proclaimed to the population in his area that the massacre at Baranivka was in line with the German antipartisan policy and that future partisan attacks would be followed by reprisals of the same sort (T. Anderson, 1999, 615). As Anderson points out, the reprisal at Baranivka was restrained in comparison with other attacks of the German antipartisan warfare (T. Anderson, 1999, 609). In other circumstances the Wehrmacht executed hundreds and even thousands of villagers in reprisal. In fact, only five days after the executions in Baranivka the 12th company was ambushed by a partisan detachment at the nearby village of Velyka Obukhivka. In direct reprisal for this attack and as punishment for failing to warn the Germans, the commanding officer ordered the complete elimination of the village which left several hundred dead (T. Anderson, 1999, 611–613).

Group-Selective Violence Across History

163

Although these reprisals are morally revolting, the evidence seems to support the claim that they were not ineffective. In this regard Anderson notes that the two reprisals at Baranivka and Velyka Obukhivka seemed to have a repressive impact on civilians, leading to an increase in information given by the local population. In addition, Lt. Col. Faasch, a battalion commander, noted that in a village some eight kilometers away from Baranivka the villagers were willing to identify seven partisans without any action on his side (T. Anderson, 1999, 615). Also, Anderson interviewed some of the witnesses of the Baranivka shooting fifty years after the incident. He found to his amazement that three out of four interviewees blamed the partisans for the reprisals, and not the Germans. Anderson concludes: “Once word of the reprisals and shootings of partisans began to spread, the villagers seemed to be more willing to cooperate with the Germans […]” (T. Anderson, 1999, 622–623). Descriptions of similar retributive killings of civilians by the Wehrmacht and the German security forces can be found for countries all over Europe. Wherever the German occupation army arrived and encountered partisan activity, it carried out reprisals as a consequence. For instance, in Greece 879 villages were destroyed between 1943 and 1944 according to one report (Mazower, 2001, 190–200), but all over Eastern Europe (Shepherd, 2009), Yugoslavia (Shepherd, 2010) and France (Laub, 2010; Lieb, 2008) similar patterns can be found. Alongside reports of these reprisals, indications of their effectiveness can be found in other parts of Europe. For instance, Mark Mazower (Mazower, 2001) describes an episode during the German occupation of Greece, where in 1941 the Wehrmacht had attacked two villages in the Struma estuary and executed all the men. As a reaction to these atrocities, people from surrounding villages handed over suspects to the Germans in order to save their villages from being attacked in the same way (Mazower, 2001, 163). In another episode in Greece, Greek partisans even went so far as to openly collaborate with the German Wehrmacht (Meyer, 2008). In Greece two partisan groups fought against the Germans, the nationalist EDES and the communist ELAS. After EDES lost ground to ELAS in 1943, the leader of EDES, Napoleon Zervas, approached the Germans for a collaboration against ELAS. Apart from a common interest in fighting communism Zervas wanted an end to the German atrocities. In a message sent to the German commander General Hubert Lanz, he outlined his reasons for the proposed collaboration in December 1943: “The reason [. . .] is the following: As a consequence of every act of sabotage the Germans burn Greek villages, and Zervas wants to prevent Greece being destroyed in this way” (my translation from Meyer, 2008, 541). In addition, reprisals in Yugoslavia and especially the massacre in Kragujevac seem to have convinced Draza Mihailovic, the leader of the Chetniks, that further resistance was going to be detrimental for the civilian

164

How Does Indiscriminate Violence Work in Practice?

population (Tomasevic, 1975, 148). Seemingly, this was also the reason why he refused an alliance with Josip Broz Tito, the leader of the communist partisans. While Tito wanted to increase military action, Mihailovic did not (Pavlowitch, 2007, 163). With regard to these indications of a coercive effect it is important to note that German reprisals were relatively unlikely to produce these effects. Several aspects complicated the ability of the Germans to communicate a coercive message. The German approach to counterinsurgency was driven by a certain “guerrillaphobia” and a paradigm of harsh punishment that made adjustments to a more effective approach often impossible (Pieper, 2015, 639). As a consequence, in many instances reprisals took a form that was unable to produce coercive effects. To mention just one example, Schulte (1989) describes a case of German antipartisan warfare in the territory of the Soviet Union. Here, a German general ordered villagers to be shot in the case of partisan attacks on the railway. According to a report, local German units complained to the military administration that reprisals of this fashion were unwarranted as there were no connections between local civilians and the partisans. Thus, it was unnecessary to pressurize civilian behavior. Still, reprisals were enforced (Schulte, 1989, 141–143). Therefore, the German forces adopted strategies of punishment even when complicity between civilians and partisans was ruled out and when civilians were already inclined to support the Wehrmacht. As a consequence, the highly repressive regime often did not leave an actual alternative to resistance and we would expect provocative effects to be abundant. At least after 1943, but most likely already after the German failure to capture Moscow in 1941, the civilian population was under the impression that the German regime was not there to stay (see e.g., Schulte, 1989, 123). The expectation that the Germans would leave within a couple of years may have discouraged many from collaborating with them. Ideologically-driven violence of extermination and reprisal in counterinsurgency existed alongside each other and sometimes became interwoven (Pieper, 2015, 639). On the one side, hostages were taken preemptively, based on ideological grounds, and killed later in cases of resistance. Here often no connection existed at all between the hostage and the resistance (e.g., Laub, 2010, 115–119). On the other side, reprisals were a direct reaction to resistance. In addition, the parallel structures of the Wehrmacht and the SS/SD often led to contradictory approaches (see e.g., Lieb, 2007). As a consequence, it is likely that counterinsurgent violence often was unable to communicate an understandable coercive threat due to the mixed messages sent by different approaches and actors. It is surprising that in many instances German violence against villages seemed to have produced the expected effects. Given the factors that made

Group-Selective Violence Across History

165

German reprisals unlikely to produce coercive effects, there are an astonishing number of instances where group-selective coercion seems to have worked. Additional Examples for the Targeting of Villages

If we look beyond the German case we find similar targeting patterns and consequences of general group-selective violence. Sometimes whole villages defected to the incumbent side, as for instance during Algeria’s War of Independence (Horne, 2006, 222–224) and the Russian conquest of Chechnya in the 19th century (Gammer, 1994, 286). In addition, recent reevaluations of classical hearts and minds campaigns, such as the British counterinsurgency campaign in Malaysia (1948– 1960) conclude that violence was much less selective on an individual basis than is generally acknowledged. Bennett (2009) points out the British strategy at the beginning of the Malayan Emergency was designed to punish villages as a group. Although the punishment inflicted was deportation, the targeting of villages had the same logic as outlined under general groupselective violence (Bennett, 2009, 430). While Bennett does not argue that the British government deliberately authorized the killing of civilians, he nevertheless points out that it “created a permissive environment by encouraging a hostile attitude towards an entire population, without initially setting out specific guidelines on the use of force” (Bennett, 2009, 431). As a consequence, in the beginning of the counterinsurgency campaign, many civilians died during British operations, which may have added to the coercive effect. In a subsequent book on the Mau Mau rebellion in Kenya (1952–1960), Bennett argues that the attitude of the British army in colonial wars was in general focused on the usage of exemplary violence even after World War II. While the principle of minimal force existed it was not applied in overseas conflict. (Bennett, 2013, 83–107). Even today targeting villages seems to remain in practice. Souleimanov and Siroky (2016) point out that Russian forces, especially during the Second Chechen War (1999–2009), used this type of targeting (Souleimanov & Siroky, 2016, 691–694). Unfortunately, they investigate only the effects on the targeted communities, thereby noting mostly the provocative effects. However, they provide some evidence of the countervailing effects that specific group-selective violence has. On the one hand, they find that indiscriminate violence has provocative effects. Given the strong norm of blood revenge in Chechnya, Souleimanov and Siroky point out that usually attacks led at least to a strong desire for revenge, with retaliatory attacks sometimes carried out years later (Souleimanov & Siroky, 2016, 694–697). On the other hand, they find that attacks by the Russians on a village sometimes convinced the village council that they should ask the rebels to refrain from attacks within the vicinity of the villages (Souleimanov &

166

How Does Indiscriminate Violence Work in Practice?

Siroky, 2016, 699–700). As a consequence, they discover what a theory of specific group-selective violence would predict: provocative and coercive effects within the same group. Targeting Families

Finally, another target of general group-selective violence can be the families of those who engage in an unwanted behavior, meaning mostly the families of rebels or terrorists. As with the previous examples, the strategic rationale is to coerce the families into preventing their sons and daughters (or other relatives) from supporting the opposition. Compared with coercing villages and tribes, this type of targeting may necessitate more intelligence on the individuals who are involved in that behavior. Nevertheless, it seems to be frequently used in recent conflicts. As a first example, in conflicts in the North Caucasus and especially in Chechnya, Russian security forces even recently targeted the families of known insurgents. Although from the Russian side this is not acknowledged as official policy, the practice is neither new for Russian security forces nor specific to the region (International Crisis Group, 2012, 28). Especially, under Chechen President Ramzan Kadyrov, family members were targeted and the Chechen administration even tried to legalize the policy of burning the houses of families under Chechen law in 2014 (International Crisis Group, 2016, 9–10). The logic of targeting families is sometimes pointed out by semiofficial figures. For instance, Kirill V. Kabanov, a member of President Vladimir Putin’s human rights council, explained: “When a person leaves to become a terrorist, he can kill hundreds of innocents. [...] Those are the morals we are talking about. We should understand, the relatives must fight this first. If the relative, before the fact, reported it, he is not guilty. If he did not, he is guilty” (cited in Kramer, 2016). Even Kadyrov himself acknowledged it in a TV interview in 2010, saying: “Next time, father will be held responsible for his son’s actions [...]. If you gave birth to a child you have to be accountable. Father will be held accountable for his son, mother for her daughter” (International Crisis Group, 2016, 9). Seemingly, this tactic was particularly used during the Second Chechen War (1999–2009) where family members of rebels disappeared if the person in question did not switch sides. With regard to the potential coercive success, the targeting of families is often counted as an important factor in reducing the insurgency (International Crisis Group, 2016, 9). In addition, the New York Times wrote: “Chechnya had about 3,000 to 5,000 unresolved disappearances from 2000 to 2005 or so. The policy, executed by the Chechen leader, Ramzan Kadyrov, the scion of a prominent Chechen family that itself switched sides, broke the organized resistance”(Kramer, 2016). Experts frequently note the

Group-Selective Violence Across History

167

counterproductive effects. For instance, Ekaterina Sokirianskaia, from the Crisis Group noted, “There can be short-term results, but I wouldn’t call it success. You can prevent some episodes of violence at the moment, but you are radicalizing whole communities” (cited in Kramer, 2016). Similar tactics can be found in other conflicts, too. Israel has targeted the families of known suicide terrorists by punitively demolishing their houses (see Benmelech et al., 2015). This measure was used mostly during the Second Intifada (2000–2005), and was reinstated in 2014. A government spokesperson explained the rationale of the policy: “There are all too many incentives to encourage people to commit terrorist crimes. [...] People who commit these terrorist crimes are not afraid to die. So the fear that the house that they lived in will be destroyed after they are gone provides an effective deterrence and saves lives” (cited in Hadid, 2015). The case also gives some insights into the potential provocative effects of such a policy. The case of one attacker, Alaa Abu Jamal, suggests that he was carrying out his attack after his cousin’s family’s house was demolished (The Economist, 2015). Similarly, one Iraqi provincial government, the government of Salahuddin, is targeting the families of ISIS supporters for deportation. According to official reports more than 200 families were deported by the beginning of 2017. The provincial government explicitly stated that the aim of these deportations was to raise the personal price that recruits of ISIS had to pay (Zucchino, 2017). The targeting of families seems to then follow the rationale of general group-selective violence. The statements of policy makers explicitly point to the deterrence effect on potential terrorists as a key asset of this type of targeting. Currently, secondary sources on the effects of this type of targeting are scarce. In addition, when journalists and researchers investigate this issue they tend to assume a tactic of specific group-selective violence, thereby disregarding potential deterrence effects. In short, there is room for additional research on this phenomenon, especially since it seems to be used quite often, recently. Conclusion The previous sections described violence against groups of civilians on various levels. Violence was directed against cities, tribes and ethnic groups, villages and families. The breadth of applications of general group-selective violence outlined in this chapter shows that the phenomenon discussed in this book is not marginal. General group-selective violence is not a strategy that was used only in premodern times, colonial wars, or in counterinsurgency. In contrast, it seems to be a generalizable strategy of coercing groups. It is a strategy of trying to scare the many by targeting some on a

168

How Does Indiscriminate Violence Work in Practice?

group level. Given the widespread nature of this phenomenon it is astonishing that it has been overlooked as a general approach to violence so far, and it highlights the necessity for increased research into the nature and the effects of group-selective violence. In addition, the cases presented here were meant to increase the plausibility of the mechanism underlying the theory of group-selective violence outlined in the previous part. The aim here was not to provide systematic evidence on the causes or the effects, but to give anecdotal substantiation to the hypotheses outlined in Chapter 5. In this regard, several observations can be made across the examples discussed: In all the cases, the perpetrators were unable to use selective violence. During sieges, targeting those responsible for resistance is impossible because the city walls shield them from being attacked. In colonial warfare, the colonial administrations often voiced their frustration about the fact that the warriors of the attacking tribes remained elusive and avoided battle even when their villages were approached. Also, in partisan or guerrilla warfare, the partisans are often unknown to the perpetrator of violence and if they are known their location is unknown. As a consequence, all perpetrators sought to indirectly put pressure on potential opponents. They were aware of the potential coercive effect of general group-selective violence. The Duke of Alba, Maréchal Bugeaud, General Friderici, or Ramzan Kadyrov, all articulated in a similar way that violence against one group was intended to coerce the others. Although this cannot be seen as direct support for hypotheses H t1 and Ht3 (group-selective violence is used only if selective violence is not an option and group-selective violence is more likely in cases where multiple groups are present), the cases suggest that all perpetrators turned to the strategy of general group-selective violence because they were seeing it as a second-best option. In several cases, general group-selective violence was put into practice as an institutionalized strategy that established a coherent system of group-selective attacks. This was particularly the case in colonial Algeria. Maréchal Bugeaud explicitly designed the system of the razzia and the bureaux arabes in order to punish tribes that did not fully submit to French rule. However, in some cases the usage of general group-selective violence seemed not always to follow strategic considerations. Most notably some of the reprisals carried out by the German army in partisan warfare were the result of a policy advising brutal violence without much regard for the coercive effectiveness. As a consequence, general group-selective violence was only used in cases where local commanders understood the effects of violence and were not ordered to act otherwise by their superiors. In most cases where detailed descriptions are available, the targeting of the group is in line with the theoretical expectations. Most importantly, while violence is selective on the group-level, it is mostly arbitrary in the selection of its individual victims and deliberately seeks to harm the innocents. A good

Group-Selective Violence Across History

169

example is Baranivka, where the German officer picked the victims randomly from among the villagers. Yet, the targeting of the innocent is common in all cases. For instance, the razzia was explicitly targeting women and children as well, either taking them as hostages or killing them in an attack. Targeting could take even more extreme forms. In siege warfare, colonial wars, and partisan warfare, violence was often designed to exterminate the whole group. Cities were razed, tribes were exterminated and villages were burned to the ground. The gruesome examples of the cases above thus support hypotheses Ht2 and Ht4 that an actor engaging in groupselective violence will deliberately target the innocent civilians within a group and an actor using general group-selective violence has an increased incentive to exterminate attacked groups. The cases also provide limited evidence that the provocative effects of general group-selective violence are mostly found in the targeted groups. Some authors mention the provocative effects particularly for the targeted groups in an effort to support the view that group-selective violence is ineffective. For instance, for Chechen villages as well as for tribes in Algeria, it is mentioned that they experienced provocative effects. However, in other cases it seems that provocative effects were limited. The Baranivka case left at least some people still convinced that the responsibility for the attack was with the partisans. This indicates that the provocative effects may be more limited than hypothesized by the theory of group-selective violence. Yet, these are only very small glimpses into the potential provocative effects which cannot be taken as support for either direction. One reason is that in many of the other examples violence often became extreme, potentially exterminating the targeted group. In these cases a provocative effect is not detectable anymore. Therefore, with regard to He2 it has to be acknowledged that we cannot find support in all case studies that provocative effects will materialize in the targeted groups. Regarding the effects of general group-selective violence it is of course not possible to draw definitive conclusions from these examples. However, as the cases demonstrate, under most circumstances some authors mention episodes of successful coercion. Assyrian kings and the Duke of Alba seem to have coerced cities into surrendering without a long siege, the French attacks in Algeria frequently prompted tribes to submit to the colonial rule, villages across Europe supported the Wehrmacht in the aftermath of German reprisals, and families seem to have reined in on potential rebels and terrorists in Chechnya and Palestine. Therefore, all the described cases conform to the logic of general group-selective violence. The coercive effects are found in the groups surrounding the ones being attacked. This supports He4, the hypothesis that in general group-selective violence the coercive effects will materialize in the surrounding groups. The ubiquity of these stories is truly astonishing. This is even more the case since many authors

170

How Does Indiscriminate Violence Work in Practice?

who report the stories argue against the effectiveness of group-selective violence. Nevertheless, as I mentioned before, most authors miss the coercion effect on other groups than those targeted with violence. History, journalism and in fact social science are often more interested in what happened rather than in things that did not happen. This has the unfortunate consequence that few sources are available on the reaction of those who were coerced. For instance, the microdynamics of collective action within the coerced groups is not described in a single account I have found so far. In this aspect seems to lie the greatest potential for future research. In various instances throughout this chapter it became clear that the perpetrators did not only let the coercive mechanism speak for themselves, but actively communicated their logic to those who were supposed to be coerced. This communication often proved difficult, for it also produced a moral backlash within the constituency. But it can be found repeatedly that the perpetrators sought to increase the coercive effects through additional communication with the targeting audience. Interestingly, this type of communication can already be found in Assyrian reliefs in the ninth century BC. In addition, the French in Algeria, the Germans in partisan warfare or the Chechen government went to great lengths to explain that they would not consider members of the specific groups innocent if they would not support them. This supports He5, the hypothesis that the perpetrator will try to communicate the strategy of group-targeting to the target audience. Some cases highlight a danger of an excess of violence among the perpetrators that was not discussed in Chapter 5. Possibly, there is an effect of barbarization on soldiers. Once soldiers are allowed to use violence and get used to killing innocent people, they may be more likely to kill for purely economic reasons. In the case of French Algeria it seems that regional commanders may have sometimes switched from the strategic use of violence as coercion to purely economic motives for a razzia. Since soldiers became more and more accustomed to getting part of the spoils of the razzia, commanders struggled sometimes to keep them in check (Gallois, 2013, 103– 108). Possibly similar motives were behind some of the German attacks on villages. At least for the case of the Prussian-French war (Wawro, 2005), reports exist of instances where German troops bombarded French villages only to extort more supplies. The instrument of group-selective violence may therefore have an intrinsic tendency to become more and more arbitrary and thus self-defeating (Wawro, 2005, 263–264). This may be an additional reason why many observers have overlooked its coercive impact when it is actually used strategically. Although it was not an explicit focus of this chapter, some cases highlight the usage and effects of specific group-selective violence. The attacks on Chechen villages described by Souleimanov and Siroky (2016) indicate that both types of effect, provocative and coercive, could be found in those

Group-Selective Violence Across History

171

villages attacked, thus giving some support to He1, He2 and He3, the hypotheses that group-selective violence will produce provocative and coercive effects, that provocative effects will materialize in the targeted groups, and that in specific group-selective violence the coercive effects will materialize in the targeted groups, respectively. In addition, the detailed description (Portal, 1937) of a coercive attempt against a tribe in Aden highlights the complications that specific group-selective violence has. In order to avoid provocative effects, the Royal Air Force had to blockade the village rather than attack it. In sum, these episodes are in line with the theoretical expectations of specific group-selective violence. In several instances the counterproductive effects on the political level become apparent. As outlined in Chapter 1, I excluded these effects from my theoretical analysis. This is, however, not to deny that group-selective violence sometimes produces an enormous moral backlash for the perpetrator either domestically or internationally. Importantly the moral backlash is not universal. The targeting of villages by the Wehrmacht and the targeting of families by Russia produced no far-reaching backlash. This may of course be due to several reasons. To begin with, in some cases the political system may not allow for a moral outcry, as for instance in the case of Germany during World War II. Here opposition to the war effort (Wehrkraftzersetzung) was punishable by death and consequentially very risky. In addition, the control of the media and domestic propaganda may distort information on the targeting. In other cases, it seems that the targeting of groups had some legitimacy among the population. Especially the discussion around the targeting of terrorist families shows that a considerable amount of the population may well be willing to accept that families are in fact responsible for the deeds of their members. Finally, these cases already give a first indication of the many possible constellations in which group-selective violence may operate. While the threat is invariably of the same kind (although it may differ in size, i.e., to kill part of the group or even the whole group) the demands made by the perpetrator and the social structure are quite different. While cities and tribes were asked to surrender, villagers and families of potential rebels are demanded to actively prevent either attacks or their relatives from joining an organization. The anecdotal evidence provides support for the effectiveness in all these situations. In addition, the groups differ with regard to their social structure. Cities, tribes, villages and families are all quite different after all. However, if general group-selective violence seeks to coerce primarily the innocent, this presupposes that they can exert influence within the group. In this case more detailed research would be needed, in order to determine how the microdynamics of group-selective violence differs between distinct types of groups.

172

How Does Indiscriminate Violence Work in Practice?

This chapter provides a picture of general group-selective violence that is mostly in line with the theoretical expectations: Perpetrators often understand the logic of violence and apply it accordingly. The targeting pattern is selective on the group-level and arbitrary on the individual level, and general group-selective violence has the expected effects on those groups that are not targeted. Notes 1. In my description of siege warfare, I draw heavily on Waxman (1998) and Kern (1999). Waxman (1998) provides a particularly detailed description of the coercive effects of violence against civilians in siege warfare. 2. Interestingly, as Waxman (1998) points out, both strategies, specific and general group selective violence were deemed moral by the scholastic philosopher Francisco de Vitoria (1483–1546) as long as they served a military purpose (Vitoria, 1991, 119–121). 3. Kern (1999) provides a good overview of the treatment of captured cities in ancient times. 4. For good overviews of the development and evolution of the razzia, see for instance Rid (2009) and Gallois (2013). In particular the account of Rid (2009) emphasizes the coercive effects of the razzia. The term razzia is an adaptation of the raids known among the Algerian tribes as the ghaziya (Rid, 2009). 5. This is however not to say that the razzia could not serve multiple purposes in addition to exercising punishment. Especially the economic benefits of the razzia seem to have motivated a more extensive usage of the tactic than the general strategy would have necessitated (Gallois 2013, 100–121). 6. Gallois argues that the enfumades were a much more common practice even than is generally assumed (Gallois 2013, 132). 7. However, German troops had used similar reprisal tactics as early as in the Franco-Prussian War of 1870 to 1871 (Wawro, 2005, 279). For a useful overview on research concerning German counterinsurgency during World War II see Pieper (2015).

8 Violence Against Civilians in Ethnic Wars

Ethnic civil wars make up a large proportion of armed conflicts today (Denny & Walter, 2014; Cederman, Gleditsch, & Wucherpfennig, 2017) and they tend to be more violent and last longer than other wars (Eck, 2009). Understanding them is, therefore, of particular importance and the theory of group-selective violence is useful to increase our understanding of the dynamics of some ethnic wars. Whenever multiple groups are present in a given country and an ethnic war breaks out between the government and one of these ethnic groups, the government may resort to group-selective violence in order to deter other groups from entering the conflict. Ethnicity and Violence Against Civilians Governments in multiethnic countries may have an incentive to use general group-selective violence in ongoing ethnic civil wars. However, two points need to be clarified upfront. First, the theoretical model discussed below will rely on the assumption that ethnic groups are present. However, it is disputed in how far we can take the presence of groups for granted. For instance, (Brubaker, 2002) argues that it is unwarranted to treat ethnicity per se as a group phenomenon. His main point is that ethnicity as a label can exist independently of the presence of a group. His argument is mostly directed against the notion that groups go to war with each other. For Brubaker, the fact that an ethnic identity exists does not create a group with preferences. However, as he points out, groups tend to form and intensify often under the impression of conflict and violence (Brubaker, 2002, 171). Therefore, situations like the

173

174

How Does Indiscriminate Violence Work in Practice?

one described below may be most conductive for ethnic groups to develop agency. Therefore, it seems plausible to ascribe at least limited agency to these groups. In addition, the argument and the dataset will concentrate on only those ethnic groups that possess a minimum of political organization. It is true that ethnicity itself is insufficient to guarantee even a loose group structure. However, once political institutions form that claim to represent the group, this changes. In these cases the group may still not act as a monolithic bloc but they have some structure to bind them together. Second, the importance of ethnicity in conflict dynamics has produced an ongoing discussion in conflict research1. Several authors argue that ethnic identities influence the dynamics of wars (Kaufman, 2001; Kaufmann, 1996). In their view, wars become either more violent or more protracted since ethnic traits cannot be hidden and thus prohibit the possibility of sideswitching. Others have disputed this view by pointing out that ethnic wars seem to be waged in much the same way as ideological wars and display similar dynamics (Kalyvas & Kocher, 2007a; Kalyvas, 2008; C. King, 2001; J. E. Mueller, 2000). It is particularly pointed out that even in ethnic wars side-switching is possible (Kalyvas, 2008). Therefore, it is important to note that the argument presented below does not rest on the assumption that ethnicity or the presence of ethnic cleavages makes wars fundamentally different from wars with other, e.g., ideological, cleavages. It will merely be assumed that ethnic groups with some political organization are present. Conflict and Violence Against Civilians in a Multiethnic Country

Researchers on ethnic conflicts already realized that countries with multiple ethnic groups may be different from countries with few ethnic groups (Toft, 2003; Walter, 2009). In particular, Toft and Walter both argue in separate studies that governments of countries with multiple ethnic groups may seek to deter groups from an unwanted behavior. Specifically, they outline that whenever a government is confronted with multiple ethnic groups, it will be much less likely to accommodate demands for more autonomy or even secession. Since governments have to fear a further breakup of the country should other groups raise the same demands, they will try to deter demands from all groups. Toft and Walter further argue that governments of multiethnic states may want to deter ethnic groups from raising demands of autonomy by waging civil wars. By fighting for the integrity of the state they raise the costs for other groups who might want to secede as well. The war, thus, has a deterrence effect on other ethnic groups. As a consequence, it is not farfetched to assume that governments may also use group-selective violence against civilians in order to seek a deterrence effect against other ethnic groups. Therefore, I follow Toft and Walter

Violence Against Civilians in Ethnic Wars

175

in that I believe a major concern of governments of multiethnic countries is the potential of additional rebellions and that governments will want to deter ethnic groups from rebelling. However, I depart from their view in that I think that risking a civil war is not an effective deterrent. Instead and in line with my theory, I argue that group-selective violence against civilians is such a deterrent. In addition, while Toft and Walter focus on negotiation before the outbreak of a war, my theoretical considerations concern ongoing wars. So my model concerns the time after potential negotiations have already broken down. In the next section I will discuss the theory of group-selective violence in the context of multiethnic countries. As outlined in the previous chapters, an actor will use violence against groups if he has an incentive to deter other groups from behaving in the same way. In order to demonstrate this, I will outline a model in the following paragraphs that integrates the theory of the previous chapter with situations in multiethnic countries. In doing so, I will build in particular on the previous work of Barbara Walter (2009). Therefore, the model I am proposing rests on similar rational choice assumptions and treats ethnic groups as unitary actors. Adapting the Model of General Group-Selective Violence to Ethnic Conflicts

Suppose a multiethnic country, which is populated by several politicized ethnic groups (A, B, C, D, and E).2 One group, group A, is in a dominant position and can be equated with the government, while other ethnic groups are largely excluded from access to power and economic spoils (B, C, D, and E). If we assume that every ethnic group is trying to maximize their returns, every excluded group will consider ways to change the current political situation, including by violent means. Whether they will try to do so will depend on their expected returns from the war; if the expected returns are positive they will be willing to risk a war, if they are negative they will not. In a very simple model, the expected gains and losses of war can be thought of as a representation of five variables. The potential gains of winning a war can be represented by 1) the probability of winning the war multiplied by 2) the returns of winning the war (e.g., gaining more autonomy). The potential losses can be conceptualized as 3) the probability of losing the war multiplied by 4) the loss incurred of losing the war (e.g., losing an already existing autonomy status) as well as 5) the actual costs of waging war (e.g., loss from destruction) (Fearon, 1995). In cases where the chances of winning the rebellion against the government are high (e.g., the government is relatively weak) groups will be more likely to risk a war, ceteris paribus. Furthermore, when the potential gains from winning are high (e.g., complete independence) or the costs of losing are low

176

How Does Indiscriminate Violence Work in Practice?

(e.g., an already marginalized group may have little to lose), a group will also have an increased probability to risk war, ceteris paribus. Finally, when the expected costs of waging war are high (e.g., human suffering, economic damage, etc.), these groups will be less likely to engage in a war, ceteris paribus. For the purpose of this investigation it will be assumed that a war breaks out between groups A and B. B has decided that it is willing to risk a war and for some reason it was impossible to reach a negotiated settlement.3 The important point in this example is that the newly arisen situation changes the expected returns for the remaining groups, C, D and E. The now ongoing civil war of group B with the government will increase the expected gains of rebellion for all other groups by increasing the probability of winning. Since the government has to use a proportion of its strength for the war with B, the other groups would be able to fight a weakened enemy. Therefore, groups C, D and E will be more likely to risk a war. For ethnic group A, the dominant ethnic group, this situation has profound consequences. As it is the chief beneficiary of the status quo it will want to preserve it. In the case of B rebelling, group A faces two interrelated problems: first, it has to fight a civil war and secondly, the potential opponents, groups C, D, and E have now an increased probability of risking a war themselves. Group A will have to search quickly for ways to reduce this probability again. In principle, the government could do this in a number of ways: first, by lowering the probability of a victory for would-be rebels (e.g., building a bigger army or showing off military strength in the war against B), secondly, increasing the expected costs of losing (e.g., threatening to revoke the rights of rebellious ethnic groups after a defeat or, more subtly, by granting concessions to nonrebellious groups), or, thirdly, increasing the expected costs of a war (e.g., increasing violence and destruction). However, these variables are not equally easy to manipulate. Changing the probability of the rebels winning has only limited relevance, because it is difficult to change in the short run. If quickly building up a bigger and more effective army were an easy task, civil wars would not be as prevalent as they are. Therefore, increasing the probability of winning militarily is not an option in practice. In addition, the costs of losing for the potential rebels depend on the status quo ante bellum. If groups C, D, and E are already excluded and the government is not willing to grant concessions to them (this is most likely one of the reasons why the war with B might have broken out in the first place) the government has no leverage to manipulate this variable (e.g., Cederman, Gleditsch, & Buhaug, 2013; Cederman, Weidmann, & Gleditsch, 2011; Cederman, Wimmer, & Min, 2010). Finally, there is the possibility of increasing the expected cost of war. This variable is open to manipulation by the government and for the sake of argument here, it will be assumed that it is the only variable the government has influence over. Therefore, the government may want to signal to

Violence Against Civilians in Ethnic Wars

177

groups C, D, and E that it will make any war as costly as possible in order to prevent them from entering into a rebellion. The easiest way to signal this is by making the existing war an example. In the terms of the previous section, the government can use coercive violence against one group in a way to send a coercive message to wouldbe opponents. This is what I termed general group-selective violence. In comparison with purely military or other substitute strategies, from the government’s point of view, violence against civilians has two important advantages. On the one hand, violence against civilians of one ethnic group comes at relatively low costs, as the risk for the government troops is minimal. On the other hand, violence against civilians is very visible and is often immediately reported. Thus, the government can be sure that other groups will receive this signal of how far the government is willing to go. However, government violence will still have provocative effects. As outlined in the previous chapters, the individuals of the targeted group are killed arbitrarily. Therefore, these individuals should experience strong provocative effects, most notably via the security mechanism (see Chapter 4). To the contrary, the coercive effects on group B should be minimal as violence does not set an incentive to change the behavior for group B. As a consequence, the conflict between groups A and B will intensify as the rebels connected with B see their ranks swell and receive increased support from the population. Therefore, on the one side, the government will have to deal with a strengthened rebellion from group B, but, on the other side these costs make the threat against other groups credible and therefore increase the deterrent effect against C, D, and E. In the theory chapter this latter connection was thoroughly discussed. Since using violence is costly, verbal communication is often not enough to communicate the willingness to use it. Therefore, violence targeting one group transports the coercive message based on an action-reaction pattern and provides the credible commitment to use violence in the future. In sum, violence against B has two separate effects for actual and potential rebels. The actual rebellious ethnic group B experiences no deterrence effect, and its members will be driven more heavily into the conflict. As a consequence civilians of B will be more likely to join the rebellion against the government. Members of C, D and E will experience a deterrence effect in the way that they perceive violence as punishment for rebelling. By not letting their ethnic groups enter into the conflict they can avoid such punishment. From the government’s point of view, these two effects work in different directions and produce a tradeoff.4 The deterrence effect reduces the probability of additional conflicts, while the counterproductive effect of violence increases the salience of the existing conflict. Therefore, the government has to weigh these two effects in order to decide whether the use

178

How Does Indiscriminate Violence Work in Practice?

of violence is in its interest, and the decision will critically depend on the number and size of the groups involved. The more groups a government needs to deter (excluded groups that have a reason to rebel but do not do so, yet) the more useful will it be to raise the costs of the ongoing wars. The Onset of Ethnic War

The previous argument on violence against civilians in multiethnic states is related and similar to recent research on the onset of ethnic civil wars. As mentioned above, especially Monica Toft (2003) and Barbara Walter (2009) have investigated the circumstances under which ethnic groups demand more freedom and governments accommodate or reject these demands. Both scholars argue that governments confronted with multiple ethnic groups are more likely to go to war than governments facing just one group. Walter in particular points out that in such situations governments find it rational to fight wars in order to deter other ethnic groups from seeking the same political concessions. However, Toft and Walter’s approaches did not take into account that an ongoing war may present an excellent opportunity for other ethnic groups to rebel themselves. The existing war changes the expected returns from rebelling for other groups, making them more likely to risk a war themselves. So governments do not face only the actual cost of wars but also an increasing risk of war. An ongoing civil war thus may raise the specter of a cascade effect, with the first war (with group B) triggering the second with group C, which was less inclined to go to war, the second war triggering the third with group D, and so on. Therefore, ethnic wars per se may not serve as a deterrent. In contrast, we would expect the outbreak of a second civil war to be much more likely if the government is already involved in a conflict and, in fact, recent research provides some support for this view. For instance, Forsberg (2013) finds exactly that ongoing civil wars make it much more likely that an ethnic group will rebel itself. In addition, K. G. Cunningham finds that civil war onsets increase the probability of concessions to self-determination movements (K. G. Cunningham, 2011, 288). Interestingly, neither Toft nor Walter explicitly test the deterrence logic of civil war onsets. What they actually analyze is the influence of previous concessions on civil war onsets. Whether civil war can deter others from entering into a conflict themselves they do not address empirically. In sum, theoretical considerations as well as the empirical record seem to stand against Toft and Walter. It seems unlikely that a government may deter ethnic groups by just fighting a war. To the contrary, it may need to find means of deterrence that counter the detrimental effects of ongoing fighting and raise the cost for potential rebels even further.

Violence Against Civilians in Ethnic Wars

179

Alas, violence against civilians may be such a means and the next section will provide evidence for this. In addition, the nature of violence against civilians as a deterrence tool may also provide the explanation for the outbreak of civil wars in the first place. Rational choice theories point out that wars start among others because of the asymmetric distribution of information (Fearon, 1995). The intended use of group-selective violence may exactly lead to such an information asymmetry. In general, rational choice theories argue that a negotiated settlement should be possible when both sides know the costs and benefits for each other. However, the intention to use group-selective violence may be difficult to communicate in a credible way beforehand. Most perpetrators will not want to admit that they are willing to use group-selective (or indiscriminate) violence as they may seek to attribute violence later to collateral damage in order to minimize the loss of international reputation. Communicating the willingness beforehand thus drastically increases the costs for the perpetrator. In addition, as discussed before, mere verbal communication may still be insufficient to change the opponent’s behavior. Since using violence against civilians has potentially high costs for the perpetrator (either through the provocative effects or international backlash) the opponent may not believe that the potential perpetrator is willing to use it. This may give a potential rebel group the impression that the costs of war are lower than they actually are and, therefore, they may risk war even if it is overall detrimental to them. Generating Hypotheses

Governments will use violence against civilians in ongoing ethnic conflicts whenever there is a need to send a strong signal to other ethnic groups that might challenge the government as well. Following the hypotheses derived in Chapter 5, two sets of tests can be designed. First, the targeting pattern and the determinants of violence can be investigated. As the above model explains under what conditions governments have an incentive to use violence against ethnic groups, it is possible to investigate if governments follow this logic. In the case of multiethnic countries this means that governments with more potentially rebellious ethnic groups have a higher benefit from using violence against civilians as a deterrence tool. Therefore, the first hypothesis is formulated as: The more nonrebellious, excluded ethnic groups are living in a country, the more likely a government will be to use violence against civilians in an ongoing ethnic civil war (Hypothesis 1 [H1])

It is possible to evaluate the effects of violence directly. If violence has a coercive effect we should be able to observe a distinct behavior by those

180

How Does Indiscriminate Violence Work in Practice?

groups coerced. As argued above in the case of ethnic rebellion, governments will in particular seek to deter additional rebellions. This leads to the second hypothesis: The use of violence against civilians by the government will decrease the probability of an additional ethnic group rebelling (Hypothesis 2 [H2])

Empirical Tests In order to test the above hypotheses I compiled two separate datasets, one on government behavior in ethnic wars to test Hypothesis 1 where the level of analysis is the country-year (test 1), and another on nonrebellious ethnic groups to test Hypothesis 2 where the level of analysis is the group-year (test 2). In both cases the scope conditions are that an ethnic armed conflict is ongoing. For both analyses data on ethnicity and violence against civilians are key. Data on the ethnic composition of a country and ethnic groups are taken from EPR Data Family 2019 (Vogt et al., 2015). The EPR Data Family is a conglomerate of several datasets on ethnic groups and conflict, which comprises the most up-to-date and comprehensive records covering 814 ethnic groups. At its core is the Ethnic Power Relations (EPR) that gathered information on ethnic groups on the basis of an online expert survey (Cederman et al., 2010; Wimmer, Cederman, & Min, 2009). Compared with the older Minorities at Risk (MAR) Project (Gurr, 2000; Gurr, Harff, Marshall, & Scarritt, 1993), the EPR does not cover only groups at risk but all politicized groups. As a consequence, EPR includes around three times more ethnic groups than MAR. The EPR dataset also has an additional advantage for this study, as it only includes politically relevant ethnic groups. These are defined as ethnic groups for which at least one organization is claiming to represent the group or where members of the group are systematically being discriminated against (Vogt et al., 2015, 3). Since the theory predicts a groupbehavior as a reaction to group-selective violence it seems advisable to focus only on those groups that have some form of political organization or are at least perceived by the government as distinct groups. Ethnic groups with stronger political organizations can be expected to have a higher probability of successful collective action (Hechter, Friedman, & Appelbaum, 1982). As a consequence, it might be better to focus on politically relevant ethnic groups rather than on socially or culturally relevant groups. The second key data source is the Uppsala Conflict Data Program’s (UCDP) One-sided Violence Dataset (v. 19.1, 1989–2018), which includes

Violence Against Civilians in Ethnic Wars

181

direct and intentional violence against civilians for every actor (governments and rebel groups) in civil wars (Peterson et al., 2019; Eck & Hultman, 2007). To be incorporated a threshold of 25 civilian deaths per actor and year has to be reached. The dataset on one-sided violence is mainly built on newspaper articles and NGO reports and thus may inhibit potential reporting biases (see e.g., Davenport & Ball, 2002; Weidmann, 2015, 2016). However, for the time being it is the most elaborate source available on violence against civilians on a cross-country basis. While certainly not all incidents of violence against civilians are included, it can, nevertheless, be seen as a reliable minimum estimate (Eck & Hultman, 2007, 236). Test 1: The Determinants of Violence Against Civilians

I first investigated whether the amount of nonrebellious excluded ethnic groups in a country can make a government more likely to use violence against civilians in an ongoing ethnic conflict. In the following I will, first, introduce the data and the methods used for the test, secondly, discuss the findings and, thirdly, discuss several robustness checks. Data and research design. The dataset assembled for test 1 covers all

ethnic civil wars and armed conflicts from 1989 to 2014. In order to identify the universe of cases several datasets were consulted. To identify conflict cases the UCDP/PRIO Armed Conflict Dataset (19.1, 1946–2018) is used (see Gleditsch, Wallensteen, Eriksson, Sollenberg, & Strand, 2002; Pettersson & Wallensteen, 2015; Peterson et al., 2019) and the identification of ethnic conflicts is taken from the sub-dataset ACD2EPR v.2014 (Wucherpfennig, Metternich, Cederman, & Gleditsch, 2012). This dataset links data on rebel organizations to ethnic group data of the EPR. The advantage of this dataset is its focus on rebel groups rather than on the war itself in order to define an ethnic war. The two necessary criteria for inclusion into the dataset are first, that the rebel organization recruits along ethnic lines, and second that it claims to represent the ethnic group. If both conditions are met the rebel group and the conflict are coded as ethnic (Wucherpfennig et al., 2012, 95). In addition, data on the amount of excluded ethnic groups are taken from the EPR core dataset (Cederman et al., 2010; Wimmer et al., 2009). The EPR distinguishes between several different categories of political inclusion or exclusion. Groups that are exercising central state authority alone are coded as either monopoly or dominant. If groups share power with others they are coded as either senior partners or junior partners. If groups are excluded from power they are coded as powerless, discriminated or self-excluded. The category “powerless” indicates that a group is not represented at the national level. In contrast, the category “discriminated”

182

How Does Indiscriminate Violence Work in Practice?

highlights that a group is actively discriminated against. The self-excluded category applies only to those groups that have retreated from the central state and declared independence for their own territory. The main independent variable is an ordinal measure of excluded, nonrebellious ethnic groups in a country. These are the groups that can be seen as potential rebels as they have reasons to seek a change in their political position. To determine the number of excluded nonrebellious groups the overall number of excluded groups is taken as a basis. From them I subtracted the amount of excluded but already rebellious ethnic groups in a country. The number of excluded and rebellious ethnic groups is taken from the EPR group-year dataset and aggregated into country-year format. The measure was than split into an ordinal scale of 5 ascending levels. Level 1 contains country years with zero or one group, level 2 country years with two and three groups, level 3 contains country years with four and five groups, level 4 contains country years with six and seven groups, and level 5 contains country years with all higher number of groups. This measure is supposed to smooth the data for the higher numbers of groups where Russia and China are huge outliers with more than 50 groups. However, in the robustness tests I will also use different operationalizations for this variable to demonstrate that this categorization does not lead to a substantive change of the results. The variable is lagged by one year. The main dependent variable for this test is a binary variable that indicates whether civilians were killed by the government in any given year. Data on this variable are taken from the UCDP One-sided Violence Dataset (Eck & Hultman, 2007). However, one important caveat remains with this source. Although it distinguishes different perpetrators (government and rebel groups), it does not distinguish between victims. Therefore, the data only indicate that the government uses violence against civilians, but not if it is used against the rebellious ethnic group(s). Yet, it is plausible to assume that the target is actually the rebellious ethnic group, based on recent findings already discussed in the previous chapters. Particularly, Fjelde and Hultman (2014), using the disaggregated UCDP data, find for Africa that governments are much more likely to target civilians in the territory of the rebellious ethnic groups. Therefore, this analysis will for now proceed under the assumption that violence is directed at the rebellious groups. Furthermore, I will be able to improve this by using geospatial data to determine which area the government targeted in the robustness checks for test 2. Then only massacres perpetrated in the settlement area of the rebelling groups will be counted. In addition, several control variables are included and all are lagged for one year. In choosing variables I restricted myself to a limited amount of the most important influences (Achen, 2005). First, three country-level variables are included that constitute potentially confounding factors. These are variables that may have an impact

Violence Against Civilians in Ethnic Wars

183

on both the dependent and the independent variable and, therefore, might be responsible for a spurious relationship (see e.g., Ray, 2003, 2005). These potential confounding factors are population size, GDP per capita and the political system. Population size is important because it is likely that the number of ethnic groups and the amount of violence increase with the size of the population. On the one hand larger countries generally have more groups to accommodate. In addition, among large populations governments may use more violence to communicate a coercive threat effectively. Data on population size are taken from the Expanded Trade and GDP Data Set (Gleditsch 2002) version 6.0 and were log-transformed as well as lagged by one year. GDP per capita is included as a proxy for economic development. It can be argued that the number of politically relevant ethnic groups tends to be lower in more developed countries. Many studies find a close association between ethnic diversity and lower economic development (Easterly, 2001; Montalvo & Reynal Querol, 2005). In addition, the economic development of a country may change the government’s costs for using violence against civilians. As mentioned in the introduction, violence against civilians may produce international reactions to which an economically more developed country will be more vulnerable. Data on GDP are also taken from the Expanded Trade and GDP Data Set (Gleditsch 2002) version 6.0 and log-transformed as well as lagged by one year. In addition, the political system of a country can have an impact on government violence against civilians and the amount of ethnic groups. Most authors argue that more democratic governments will use less violence because leaders are either constrained by norms or by democratic institutions (Merom, 2003; Rummel, 1995; Valentino et al., 2004). However, a minority of researchers also argues that democracies may be more willing to use violence against civilians (Downes, 2008, 2006; Valentino et al., 2006). In addition, the political system can have a large impact on the political saliency of different ethnic identities and thus the number of groups active in the political sphere (Posner, 2005). Therefore, the polity2 variable from the Polity IV dataset (Marshall, Gurr, & Jaggers, 2014) is included, which is a composite of two measures for autocracy and of democracy. Both measures range from 0 to 10 and count several characteristics of autocracy and democracy respectively. The polity2 variable is constructed by subtracting the autocracy variable from the democracy variable and, thus, depicts a range from -10 (highly autocratic) to 10 (highly democratic). Secondly, three variables are included that characterize the ongoing conflict: intensity of conflict, rebel strength and rebel violence against civilians. I expect these variables to potentially have an impact on the

184

How Does Indiscriminate Violence Work in Practice?

dependent variable as they may change the government’s perception of the necessity to deter potential rebels. The intensity of war is likely to influence the amount of violence against civilians. In more intense conflicts both actors field higher numbers of soldiers with an increased capacity to do harm. Since my dataset includes armed conflicts (at least 25 deaths per year) as well as civil wars (at least 1000 deaths per year) I include a variable that indicates how intense the conflict is. It shows the value of 2 when at least 1000 battle deaths were recorded the previous year, 1 when between 1000 and 25 battle deaths occurred and 0 if the amount of battle deaths was below 25 (since the variable is lagged this is the case when no conflict was present the previous year). This data is taken from the UCDP/PRIO Armed Conflict Dataset version 19.1 (Pettersson & Wallensteen, 2015). The strength of the rebel groups fighting the government is another important variable. It has been argued that governments will be more likely to use violence when they are facing a rebellion that poses a serious threat (Valentino et al., 2004). Data on rebel strength are taken from the Non-State Actor Data which provides data on the traits of rebel groups (D. Cunningham, Gleditsch, & Salehyan, 2009; D. Cunningham et al., 2013). In this dataset the measure of rebel strength is a categorical variable capturing relative strength towards the government. It has five categories ranging from very weak to very strong. These categories are transformed into an ordinal measure ranging from one to five, with 1 being very weak rebels and 5 being very strong rebels. In addition, the rebel group’s usage of violence against civilians has an influence on the government violence. As outlined before, Kalyvas and Kocher (2007b) argue that the interaction between rebel violence and government violence is important and rebel violence may reduce the effectiveness of government violence. This means that rebel violence may also lead to increases in government violence. As a consequence, I included a variable that indicates five different levels of rebel violence against civilians in the previous period. The variable was constructed in a way that all categories include an amount of observations that is as equal as possible. Here level 1 indicates no violence against civilians by rebels, level 2 up to 51 civilians killed per year, level 3 includes up to 172 civilians killed, level 4 up to 350 and level 5 all amounts of violence above. This variable also uses the UCDP data on One-Sided Violence. Finally, the estimation method used here is a logit regression to account for the fact that the dependent variable is a binary variable taking on the value of 1 for years in which the government did use violence against civilians and 0 for years in which the government did not use violence against civilians. Since binary variables violate the model assumptions of the ordinary least squares (OLS) estimation it is necessary to use a logit link function (Menard, 2001).

Violence Against Civilians in Ethnic Wars

185

Results and discussion. In order to get an impression of what the data look

like Figure 8.1 plots the percentage of war-years during which violence against civilians was used by the government for different ordinal levels of nonrebellious ethnic groups. This plot can of course only give a first impression of the data but there is a seeming increase in the usage of violence observable for countries that host more groups. Only countries in category 3 seem to not follow the trend. Nevertheless, we need to wait for the systematic analysis of the data before we can draw any conclusions5. Moving beyond descriptive statistics, the logit regression analysis supports the first impression. Table 8.1 shows the results for several model specifications. In the first model only the main independent variable is included. In the following models the control variables are gradually incorporated and the fourth model shows the full model specification. The regression results give support to the hypothesis that more nonrebellious, excluded ethnic groups are living in a country, the more likely a government will be to use violence against civilians in an ongoing ethnic civil war. The number of potentially rebellious ethnic groups in the previous year is strongly associated with the usage of violence against civilians in the current year. In all three models the variable is significant at the 0.1% level and positive. This finding indicates that in countries with more nonrebellious

Figure 8.1 Ethnic Groups and the Usage of Violence Against Civilians

186

How Does Indiscriminate Violence Work in Practice?

Table 8.1 Logit Regression Results for the Determinants of Violence Dependent Variable

Nonrebellious Groups (ordinal) LN Population (lagged)

Violence Against Civilians (binary) (1) (2) (3)

0.437*** (0.058)

0.371*** (0.078) 0.278* (0.123) -0.454*** (0.110) -0.104*** (0.019)

-1.757*** (0.180)

0.828 (0.757)

LN GDP per capita (lagged) Polity2 (lagged)

Intensity of Conflict (lagged) Rebel Strength (lagged)

Rebel Violence (lagged) Constant

Observations Log Likelihood Akaike Inf. Crit.

690 -419.147 842.295

Note: *p