VIOLENT ORDER: REBEL ORGANIZATION AND LIBERIA’S CIVIL …cd347ss0802/Lidow_DissFinal... ·...

VIOLENT ORDER:

REBEL ORGANIZATION AND LIBERIA’S CIVIL WAR

A DISSERTATION

SUBMITTED TO THE DEPARTMENT OF POLITICAL SCIENCE

AND THE COMMITTEE ON GRADUATE STUDIES

OF STANFORD UNIVERSITY

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

Nicholai Hart Lidow

August 2011

http://creativecommons.org/licenses/by-nc/3.0/us/

This dissertation is online at: http://purl.stanford.edu/cd347ss0802

© 2011 by Nicholai Hart Lidow. All Rights Reserved.

Re-distributed by Stanford University under license with the author.

This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.

ii



http://purl.stanford.edu/cd347ss0802

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

David Laitin, Primary Adviser


James Fearon


Jeremy Weinstein

Approved for the Stanford University Committee on Graduate Studies.

Patricia J. Gumport, Vice Provost Graduate Education

This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.

iii

Abstract

Rebel groups exhibit significant variation in their treatment of civilians, with profound

humanitarian consequences. This dissertation proposes a new theory of rebel behavior

and cohesion based on the internal dynamics of rebel groups. Rebel leaders have

incentives to provide security, but are often unable to prevent group members from

abusing civilians. Leaders exert effective control over their troops when they can offer

cash payments and credible future rewards to their top commanders. Leaders who

cannot offer these incentives allow their forces to loot locals in exchange for a minimal

level of loyalty.

The leader’s ability to offer incentives that allow local security to prevail de-

pends on partnerships with external actors such as diaspora communities and foreign

governments. When these patrons have a stake in the group’s success, they are moti-

vated to supply financial resources to qualified, trusted leaders. Other patrons have

goals that conflict with those of the rebel client. These patrons exert leverage over

the rebels by supporting low quality leaders and withholding resources that could

strengthen leader control, resulting in more abusive, faction-prone groups. The type

of partnership available depends on factors beyond the group’s control.

The dissertation formalizes this theory and tests the implications in various ways,

using an original cross-national dataset of rebel groups, 1980-2003, as well as a micro-

level dataset of crop area during Liberia’s war, derived through remote sensing. The

dissertation also traces the model’s logic through a detailed analysis of Liberia’s rebel

groups, based on fourteen months of fieldwork and interviews with nearly all surviving

rebel leaders and top commanders who participated in Liberia’s war, 1989-2003.

v

Preface

My first experience in a conflict zone was in western Cote d’Ivoire, in August 2003. I

was traveling with Malayee Cheyard, a Liberian refugee I had gotten to know while

completing an internship in Accra, Ghana. Malayee was searching for his mother

and siblings, whom he hadn’t seen in thirteen years, and I, as a naive Sophomore in

college, wanted to gain experience. We rode public buses for two days and crossed

dozens of military roadblocks to reach the Nicla refugee camp, 450km northwest of

Abidjan.

The Liberians referred to the camp as “Peace Town”, but the name was more

hopeful than descriptive. Only a few miles from the fighting in Liberia, the camp

received hundreds of new refugees each day and served as a base of operations for

Liberia’s MODEL rebels. The camp was also on the frontline of Cote d’Ivoire’s civil

war. The refugees lived in fear of forced conscription and sporadic attacks by the

multitude of armed groups in the area. But despite the political turmoil, life carried

on for the Liberians in the camp and for the Ivorians outside it. People bought and

sold goods in the local market. Men gathered to watch European football matches

broadcast live via satellite.

This experience challenged my preconceived notions of conflict. Rather than

chaotic violence, I realized that security—or the lack of security—is the defining

feature of civil war for civilians. While civilians in some conflicts are able to remain

in their communities and continue their livelihoods, civilians in other conflicts, facing

an acute risk of looting and abuse, are forced to flee their homes for the shelter of

a refugee camp. Fighting and killing, while tragic and undoubtedly significant, are

only a small part of the conflict experience.

vi

These insights led me to consider two puzzles, which are the focus of the disserta-

tion. First, why do some rebel groups provide security and promote local production,

while others loot and prey on the population? And second, why do some rebel groups

break into competing, violent factions, while others remain unified? These puzzles

stayed with me over the following years, as I visited the conflict areas of Sri Lanka,

DR Congo, and Somalia, witnessed electoral violence in Zanzibar, and made extended

trips to Liberia.

My core insight, based on these experiences, is that rebel group discipline de-

pends on the interaction between rebel leaders and their top commanders. Rebel

leaders generally have an incentive to provide security in their territories, but rely

on commanders to do so. When leaders provide strong incentives for cooperation,

including both immediate material rewards and credible promise of future benefits,

commanders are motivated to train, feed, and discipline their forces. The result is

greater security for civilians and a lower risk that the group will break into competing

factions. When leaders fail to offer these incentives, commanders and their troops

devote their energy towards personal gain at the expense of group goals. Looting,

civilian abuse, and a high risk of factionalization are the result.

The leader’s ability to offer the incentives necessary for security to prevail depends

to a large extent on external patrons. Without access to credit markets or externally

enforced contracts, external patrons such as foreign governments or diaspora commu-

nities are the best source for financing and weapons. These patrons, therefore, exert a

strong influence on the group, not only deciding how many resources to provide, but

also influencing the leader’s credibility among the troops. When the patrons’ goals

align with those of the rebel group, patrons are motivated to provide ample resources

to qualified, charismatic leaders who are trusted by their troops. These leaders in

turn provide material incentives to motivate commanders and use their credibility to

make believable promises of future rewards.

Other patrons, however, have goals that are not in their rebel client’s best in-

terest. Foreign governments, for example, may support rebellion in a neighboring

state to gain leverage in international negotiations. These patrons maximize their

leverage over the rebel group’s activities by supporting relatively weak leaders and

vii

withholding financial resources that could strengthen leader control. Unable to offer

cash incentives or credible future rewards, these leaders allow their troops to loot

locals in exchange for a minimal level of loyalty and are thus less able to prevent

factionalization.

In the pages that follow, I formalize this argument into a model and test the

implications in various ways. I also trace the model’s logic through a detailed analysis

of Liberia’s rebel groups, based on fourteen months of fieldwork and interviews with

nearly all surviving rebel leaders and top commanders who participated in Liberia’s

war, 1989-2003. The argument is not meant to be comprehensive. It ignores many

features of war that deserve attention, such as ideology, and perhaps devotes too little

space to communicating the violence and terror that civilians experience. But I hope

to show that even a narrow look at the economic and organizational challenges of

rebel groups can explain a significant part of the bewildering complexity that is civil

war.

viii

Acknowledgements

First and foremost, my thanks go out to my friends and colleagues in Liberia. Little

of this dissertation would have been possible without their knowledge, patience, and

generosity. I am especially grateful to Malayee Cheyard, Maurice James Davis, Bill

Roberts, Alfred Lomax, and Musa S. Kamara, who taught me everything I know

about “street smarts”, in Liberia and beyond. More than one hundred rebel leaders,

commanders, and ordinary people throughout the country opened their doors and

shared their often painful stories to help me better understand their country. I owe

these informants, who will remain anonymous, a great debt.

My dissertation committee provided inspiration and invaluable advice throughout

my graduate career. David Laitin encouraged me to ask big questions without giving

up the rigor of scientific inquiry. David understood my need to be “in the field”

to generate ideas and encouraged me to travel, even when these trips threatened to

distract and delay my graduate studies. Jim Fearon is unparalleled in his ability

to see through the jumbled thoughts of an emerging researcher and identify the core

insights—and inconsistencies—of a theory or model. Jim’s attention to detail and per-

sonal support have made a big impact on both this project and my development as a

scholar. Jeremy Weinstein encouraged me to focus on research design and uncovering

data, even when I doubted my ability to accomplish these tasks. That this disserta-

tion contains any empirical data at all is due to Jeremy. These three professors—each

an incredible scholar and person—will continue to inspire me throughout my career.

I have been fortunate to benefit from the support of professors beyond my disser-

tation committee. The first twenty years of my life were devoted to physics. But then

I took courses from David Abernethy, Larry Diamond, and Steve Stedman. These

ix

professors inspired me to shift my attention to the “harder” science of politics—and

changed my life. Over the years I have also received valuable advice and encourage-

ment from Jonathan Bendor, Eli Berman, Lisa Blaydes, James Fowler, Justin Grim-

mer, Macartan Humphreys, Ted Miguel, Dan Posner, Jonathan Rodden, Jonathan

Wand, and Barry Weingast. I have also benefited from the wonderful staff in the

Political Science office, especially Chandelle Arambula, Jackie Sargent, and Eliana

Vasquez. My graduate research was supported by grants from Stanford’s Graduate

Research Opportunity program and the National Science Foundation (SES-1023712).

I attended graduate school with an exceptional generation of students. I am

extremely lucky to count a few of them—Tim Johnson and Amanda Robinson, in

particular—as close friends. Advice from the graduate students at Stanford has

taught me more than any book or classroom ever could, and the camaraderie has

made graduate school especially meaningful. In addition to Tim and Amanda, I am

grateful to Andrea Abel, Claire Adida, Rob Barlow, Rachel Brule, Dara Kay Cohen,

Jesse Driscoll, Roy Ellis, Desha Girod, Jessica Gottlieb, Paul Gowder, Scott Handler,

Danielle Harlan, Sangick Jeon, Kimuli Kasara, Ruth Kritchelli, Avital Livny, Aila

Matanock, Jake Shapiro, Rachel Stein, and Ali Valenzuela. Some of these colleagues

have already made their mark on the profession, and all continually amaze me with

their ideas.

My family supported me despite the extended journeys to places that are unde-

niably dangerous (though I made my best attempts at denial). My parents, Janet

Hart and Alex Lidow, my sister Haile, and my “Didi” Estela Paiz have always been a

remarkable source of strength, inspiration, and encouragement. The companionship

of my fiancee, Heidi Moseson, transformed these past few years into the happiest of

my life. What could have been a lonely year in Liberia became instead a wonderful

adventure together. Heidi’s intelligence, sense of adventure, and enthusiasm—even

after 1,200 miles on the back of a motorbike—are the guiding lights of my life. This

is for her.

x

Contents

Abstract v

Preface vi

Acknowledgements ix

I Rebel Organization 1

1 Understanding Rebel Organizations 3

1.1 A Tale of Two Commanders . . . . . . . . . . . . . . . . . . . . . . . 3

1.2 Three Themes of the Dissertation . . . . . . . . . . . . . . . . . . . . 5

1.3 The Argument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.4 The Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.5 Theories of Rebellion . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.6 Organizational Behavior . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.7 Research Design, Data, and Results . . . . . . . . . . . . . . . . . . . 18

1.8 Scope Conditions and Limitations . . . . . . . . . . . . . . . . . . . . 22

1.9 Overview of the Dissertation . . . . . . . . . . . . . . . . . . . . . . . 26

2 Leaders, Commanders, and Soldiers 30

2.1 Spot Payments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.2 Future Rewards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.3 Lootable Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.4 Punishment, Status, and Ideology as Motivators . . . . . . . . . . . . 40

xi

2.5 Controlling Soldiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.6 Basic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2.6.1 Solving the Basic Model . . . . . . . . . . . . . . . . . . . . . 48

2.6.2 Commitment Problems and Cash Constraints . . . . . . . . . 51

2.7 Extending the Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

2.7.1 Affinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

2.7.2 Imperfect Monitoring . . . . . . . . . . . . . . . . . . . . . . . 53

2.7.3 Private Information . . . . . . . . . . . . . . . . . . . . . . . . 57

2.7.4 Multiple Commanders . . . . . . . . . . . . . . . . . . . . . . 59

2.8 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

2.A Mathematical Appendix . . . . . . . . . . . . . . . . . . . . . . . . . 62

2.A.1 Proof of Proposition 1 . . . . . . . . . . . . . . . . . . . . . . 62



3 External Patrons 68

3.1 The Motives of External Patrons . . . . . . . . . . . . . . . . . . . . 68

3.1.1 State Sponsors . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.1.2 Corporations . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.1.3 Diaspora Communities . . . . . . . . . . . . . . . . . . . . . . 73

3.1.4 NGOs and Other Patrons . . . . . . . . . . . . . . . . . . . . 75

3.2 The Importance of External Support . . . . . . . . . . . . . . . . . . 77

3.3 A Model of External Support and Influence . . . . . . . . . . . . . . 80

3.3.1 Type 1: The “We’re-in-this-Together” Patron . . . . . . . . . 83

3.3.2 Type 2: The “More-for-Less” Patron . . . . . . . . . . . . . . 84

3.3.3 External Patrons and Commitment Problems . . . . . . . . . 86

3.3.4 Multiple Patrons . . . . . . . . . . . . . . . . . . . . . . . . . 87

3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

3.A Mathematical Appendix . . . . . . . . . . . . . . . . . . . . . . . . . 89



xii

4 Rebel Abusiveness and Factionalization 94

4.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

4.1.1 Dependent Variables . . . . . . . . . . . . . . . . . . . . . . . 96

4.1.2 Explanatory Variables for Commander Incentives . . . . . . . 100

4.1.3 Explanatory Variables for External Patrons . . . . . . . . . . 103

4.1.4 Exogenous Proxy Variables . . . . . . . . . . . . . . . . . . . 107

4.1.5 Competing Theories and Control Variables . . . . . . . . . . . 108

4.2 Civilian Abuse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

4.2.1 Civilian Abuse and Commander Incentives . . . . . . . . . . . 119

4.2.2 External Patrons and Leader Resources . . . . . . . . . . . . . 121

4.2.3 External Patrons and Rebel Abusiveness . . . . . . . . . . . . 124

4.2.4 Exogenous Proxies and Rebel Abusiveness . . . . . . . . . . . 125

4.3 Factionalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

4.3.1 Factionalization and Commander Incentives . . . . . . . . . . 128

4.3.2 Factionalization and External Patrons . . . . . . . . . . . . . 129

4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

II Liberia’s Civil War 140

5 An Overview of Liberia 143

5.1 A Settler Republic, 1822-1980 . . . . . . . . . . . . . . . . . . . . . . 143

5.2 The Military Years, 1980-1989 . . . . . . . . . . . . . . . . . . . . . . 147

5.3 The War Begins, 1989-1992 . . . . . . . . . . . . . . . . . . . . . . . 151

5.4 A Failed State, 1992-1997 . . . . . . . . . . . . . . . . . . . . . . . . 155

5.5 Return to War, 1997-2005 . . . . . . . . . . . . . . . . . . . . . . . . 161

5.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

5.A Timeline of Liberia’s War . . . . . . . . . . . . . . . . . . . . . . . . 168

6 Inside Liberia’s Rebel Organizations 170

6.1 NPFL: Open for Business . . . . . . . . . . . . . . . . . . . . . . . . 170

6.1.1 Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171

xiii

6.1.2 Resources and Commander Incentives . . . . . . . . . . . . . . 172

6.1.3 Factionalization . . . . . . . . . . . . . . . . . . . . . . . . . . 178

6.1.4 Controlling Soldiers . . . . . . . . . . . . . . . . . . . . . . . . 178

6.1.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

6.2 ULIMO and its Offspring: Self-Destruction . . . . . . . . . . . . . . . 184

6.2.1 Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184




6.2.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192

6.3 LURD: Distrust and Dysfunction . . . . . . . . . . . . . . . . . . . . 193

6.3.1 Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193




6.3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201

6.4 MODEL: Almost Doesn’t Count . . . . . . . . . . . . . . . . . . . . . 203

6.4.1 Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203




6.4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

6.5 Assessing the Theoretical Model . . . . . . . . . . . . . . . . . . . . . 211

7 Civilian Security during Liberia’s War 214

7.1 Life in Rebel Territory . . . . . . . . . . . . . . . . . . . . . . . . . . 214

7.1.1 Life in NPFL Territory: Better Stay in Town . . . . . . . . . . 215

7.1.2 Life in ULIMO Territory: Run for the Hills . . . . . . . . . . . 218

7.1.3 Life in LURD Territory: Carrying a Heavy Load . . . . . . . . 221

7.1.4 Life in MODEL Territory: Ethnic Ties Help . . . . . . . . . . 223

7.1.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225

xiv

7.2 Remote Sensing and Local Food Production . . . . . . . . . . . . . . 225

7.3 Group-Level Comparisons . . . . . . . . . . . . . . . . . . . . . . . . 230

7.4 Within-Group Spatial Variation . . . . . . . . . . . . . . . . . . . . . 232

7.4.1 Spatial Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 237

7.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242

7.4.3 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245

7.5 Changes Over Time and the Endogeneity of Rebel Bases . . . . . . . 248

7.6 Assessing the Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . 249

7.A Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252

8 Conclusion 254

8.1 Contributions of the Dissertation . . . . . . . . . . . . . . . . . . . . 254

8.2 Policy Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258

8.3 Looking Forward . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265

A Fieldwork in Liberia 266

B Interviews 274

B.1 Civilian Security, Local Production, Rebel Governance . . . . . . . . 274

B.2 Rebel Organization, External Patrons, Commander Incentives . . . . 276

B.3 Miscellaneous Interviews . . . . . . . . . . . . . . . . . . . . . . . . . 277

B.4 TRC Testimonies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278

Bibliography 279

xv

List of Tables

4.1 Average civilian abuse by rebel group, in order of descending abuse. . 97

4.2 Group lifespan (years) and factionalization. . . . . . . . . . . . . . . . 99

4.3 Leader financing and lootable resources, by rebel group. . . . . . . . . 102

4.4 External patron identity and motives, by rebel group. . . . . . . . . . 106

4.5 Descriptive statistics for the RAFD dataset. . . . . . . . . . . . . . . 109

4.6 Competing predictions of the effects of explanatory variables on civilian

abuse. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

4.7 Regional descriptive statistics. . . . . . . . . . . . . . . . . . . . . . . 111

4.8 Correlations among key variables. . . . . . . . . . . . . . . . . . . . . 133

4.9 OLS models of civilian abuse with variables for commander incentives. 134

4.10 Models of leader resources among rebels receiving external support,

with variables for patron identity and motives. . . . . . . . . . . . . . 135

4.11 Models of civilian abuse with variables for patron identity and motives. 136

4.12 Reduced-form model of civilian abuse with exogenous proxies for leader

financing and lootable resources. . . . . . . . . . . . . . . . . . . . . . 137

4.13 Cox proportional hazard model of group factionalization with variables

for commander incentives. . . . . . . . . . . . . . . . . . . . . . . . . 138

4.14 Cox proportional hazard models of group factionalization with vari-

ables for patron identity and motives. . . . . . . . . . . . . . . . . . . 139

7.1 Competing predictions of rebel behavior in Liberia. . . . . . . . . . . 234

7.2 Village-level security outcomes in rebel territory. . . . . . . . . . . . . 234

7.3 Average percent change in crop area among villages, according to dis-

tance from rebel headquarters or presence of rubber. . . . . . . . . . 234

xvi

7.4 Descriptive statistics. . . . . . . . . . . . . . . . . . . . . . . . . . . . 243

7.5 Spatial model of changes in crop area for NPFL (1986 v. 1991) and

LURD (2001 v. 2003). . . . . . . . . . . . . . . . . . . . . . . . . . . 244

7.6 Changes in crop area within 100km of rebel headquarters for NPFL

(1986 v. 1991) and LURD (2001 v. 2003). . . . . . . . . . . . . . . . 247

7.7 Change over time in a section of NPFL territory and spatial variation

within a section of LPC territory. . . . . . . . . . . . . . . . . . . . . 250

7.8 Balance statistics for the matched samples. . . . . . . . . . . . . . . . 252

A.1 List of field sites in Liberia. . . . . . . . . . . . . . . . . . . . . . . . 270

xvii

List of Figures

1.1 A principal-agent model of rebel organization. . . . . . . . . . . . . . 29

2.1 Basic model of rebel organization . . . . . . . . . . . . . . . . . . . . 49

2.2 Commander’s share of group resources in a high-affinity rebel group. . 54

3.1 A model of patron-leader interaction. . . . . . . . . . . . . . . . . . . 82

4.1 Average civilian abuse over time, by region. . . . . . . . . . . . . . . 114

4.2 Average value of leader financing (solid) and lootable resources (dashed)

over time, by region. . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

4.3 Distribution of coefficients for within-group trends in civilian abuse

over time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

4.4 Percentage of groups receiving support from a foreign state (solid),

corporation (dashed), and diaspora community (dotted) over time, by

region. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

4.5 Risk of factionalization over time, by region. . . . . . . . . . . . . . . 127

5.1 NPFL’s advance across Liberia. . . . . . . . . . . . . . . . . . . . . . 152

5.2 ULIMO’s advance against the NPFL. . . . . . . . . . . . . . . . . . . 157

5.3 Territorial control in Liberia, 1995. . . . . . . . . . . . . . . . . . . . 160

5.4 LURD’s advance across Liberia. . . . . . . . . . . . . . . . . . . . . . 165

5.5 MODEL’s advance across Liberia. . . . . . . . . . . . . . . . . . . . . 167

6.1 NPFL command structure. . . . . . . . . . . . . . . . . . . . . . . . . 175

xviii

6.2 NPFL recruits undergo military training, supervised by Charles Taylor

(center, in military fatigues). Gborplay, March 1990. . . . . . . . . . 180

6.3 NPFL recruits wait to receive weapons after three weeks of training

with sticks. Gborplay, March 1990. . . . . . . . . . . . . . . . . . . . 180

6.4 ULIMO command structure. . . . . . . . . . . . . . . . . . . . . . . . 197

6.5 LURD command structure. . . . . . . . . . . . . . . . . . . . . . . . 197

6.6 A LURD platoon chants a song while on the move, 2003. . . . . . . . 202

6.7 A squad of LURD soldiers poses for the camera, 2003. . . . . . . . . . 202

6.8 MODEL command structure. . . . . . . . . . . . . . . . . . . . . . . 213

7.1 Change in crop area in NPFL and LURD territories (shaded) and

neighboring regions of Sierra Leone and Guinea. . . . . . . . . . . . . 228

7.2 Change in crop area, 1991 v. 1995, in territory that remained under

NPFL control (shaded) and territory captured by LPC (dotted). . . . 229

7.3 Percent change in crop area at village level in NPFL and LURD terri-

tories. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235

7.4 Changes in crop area according to distance to rebel headquarters. . . 238

7.5 Effect of rubber on changes in crop area. . . . . . . . . . . . . . . . . 239

7.6 Natural resources in Liberia. . . . . . . . . . . . . . . . . . . . . . . . 241

7.7 Number and location of battles. . . . . . . . . . . . . . . . . . . . . . 243

7.8 GAM plots of the effect of distance from rebel headquarters on changes

in crop area. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253

A.1 Map of field sites in Liberia. . . . . . . . . . . . . . . . . . . . . . . . 269

xix

Part I

Rebel Organization

1

2

History knows many more armies ruined by want and

disorder than by the efforts of their enemies.

Cardinal Richelieu, Political Testament (1624)

Chapter 1

Understanding Rebel

Organizations

1.1 A Tale of Two Commanders

By the time the National Patriotic Front of Liberia (NPFL) launched its invasion of

Liberia on Christmas Eve, 1989, General Johnson T. Leamah1 had been a member

of the group for three years. Leamah was once a soldier in Liberia’s national army

but fled the country when the dictator, Samuel Doe, purged members of Leamah’s

Gio ethnic group from the military. Languishing in the refugee camps of neighboring

Cote d’Ivoire, Leamah heard that Charles Taylor, a former government official, had

secured support from Libya and other patrons to overthrow the brutal Doe regime.

Leamah was initially skeptical but decided to join Taylor when he learned that he

was related to Taylor’s wife, Agnes—she shared Leamah’s Gio ethnicity and grew up

in a village close to his home area. As far as Leamah could tell, Charles Taylor had

connections to every government and businessman in the region and offered the best

chance for one day returning home.

After two years of fighting, Leamah commanded nearly four thousand soldiers

1To preserve the anonymity and confidentiality of the informants, these vignettes are based oncommanders who died during the war. Their experiences are similar to those of other high-levelcommanders in their respective rebel groups, many of whom were interviewed for this project.

3

4 CHAPTER 1. UNDERSTANDING REBEL ORGANIZATIONS

from his base in the southeastern town of Buchanan and had authority over several

hundred square miles of Liberian territory. His primary responsibility was to secure

the area from attacks, but Leamah was also expected to protect civilians and provide

security for the multinational corporations that operated in NPFL territory. The job

was not easy. In addition to planning military operations, Leamah deployed escorts

for logging trucks and constructed a training base for new recruits. He organized a

system of taxation to collect food without looting civilians and punished soldiers who

misbehaved. Because of these actions, civilians in many areas were protected from

abuse and continued to farm and engage in market activity.2

Leamah’s efforts were rewarded by Charles Taylor. Taylor made frequent, im-

promptu visits to Leamah’s territory and gave him $15,000 in cash or more at each

visit. Leamah was grateful for the rewards and knew that other, less successful com-

manders had been transferred to the frontlines or killed as punishment for failure. He

also expected to benefit after the war. Although Leamah died from sickness before

the 1997 elections brought Charles Taylor to power, Taylor appointed Leamah’s suc-

cessor, Roland Duo, as head of security for a major timber corporation, a position

that opened the door to lucrative commercial opportunities. The other high-level

commanders in the group were also rewarded for their service.

Two years after the NPFL invaded Liberia a new rebel group, the United Liber-

ation Movement for Democracy (ULIMO), emerged to challenge the NPFL’s domi-

nance of Liberia’s countryside. ULIMO’s forces were led by Thomas Karla who, like

Leamah, began his career in Libera’s military. When the government collapsed in the

face of the NPFL invasion, Thomas Karla fled to the refugee camps in Sierra Leone,

where he joined a self-defense group that eventually grew into an insurgency. Thomas

Karla distrusted the group’s leader, Alhaji Kromah, but Kromah seemed to offer the

only chance for securing weapons and support for the struggling soldiers.

After invading Liberia, Thomas Karla commanded ULIMO’s forces from his base

in the northwest diamond fields. Leader Alhaji Kromah largely failed to deliver on

his promises of securing support for the group, and Commander Karla struggled to

2While civilians in NPFL-controlled towns were largely protected, civilians in remote or unstableareas, and civilians from particular ethnic groups, faced intense abuse at the hands of NPFL forces.These patterns of abuse are discussed at length in later chapters.

1.2. THREE THEMES OF THE DISSERTATION 5

keep his forces supplied with ammunition and food. Although he initially set up

training camps and organized food distribution for his troops, he soon abandoned

these efforts. Karla didn’t expect to be rewarded by Alhaji Kromah after the war, so

he decided to profit while he still had the chance. Rather than focusing on military

training and discipline, Karla and his troops increasingly turned their attention to

mining diamonds and looting civilians.

Tensions between Karla and Alhaji Kromah erupted into armed confrontation in

1994 over the allocation of positions in a transitional government. Thomas Karla

was wounded in the fighting, and broke off with other high-ranking commanders to

form a new faction. Over the next year, the two ULIMO factions battled each other

fiercely and preyed on the civilians caught in-between. Villages in ULIMO territory

were depopulated as civilians hid in the forest or fled to displacement camps. While

people in NPFL territory were able to continue farming and trading, civilians in

ULIMO territory foraged for food in the jungle and lived in constant fear.

1.2 Three Themes of the Dissertation

These two vignettes illustrate the three principal themes of the dissertation. The first

theme is the vast differences in the quality of life for civilians and the coherence of

rebel armies during civil war that remain unexplained in standard accounts of civil

war violence. In the past half-century, some 18 million people lost their lives in vio-

lent conflict; many more died from war-related malnutrition and disease (Ghobarath,

Huth, and Russett 2003). Existing theories argue that civilians suffer most when con-

trol over territory is unstable (Olson 1993; Kalyvas 2006); when rebels have access

to economic resources (Weinstein 2007); or when rebel groups are ethnically diverse

(Humphreys and Weinstein 2006).

But these existing theories cannot explain why the ULIMO rebels preyed on civil-

ians in even the most stable areas of control, or why the NPFL rebels, with access to

tremendous economic resources, provided security for civilians in many parts of their

territory. Theories of ethnic composition cannot explain the variation in behavior

between these two ethnically diverse groups, or the high levels of abuse committed


by other, ethnically homogeneous rebel groups in Liberia’s war.

The literature also struggles to explain why some rebel groups break into compet-

ing, violent factions while other groups remain unified throughout the war. Using an

original dataset, I find that more than 32% of rebel groups in the last three decades

factionalized at some point in their history.3 Understanding rebel factionalization

holds more than theoretical significance. In Liberia, ULIMO’s split unleashed a wave

of looting, massacres, and abuse. Across conflicts, factionalization makes civil wars

longer, more violent, and more difficult for international intervention (Cunningham

2006, Stedman 2002).

The second theme of the dissertation is the agency problem that exists within

rebel organizations, and especially between rebel leaders and their top commanders.

Leaders delegate weapons and manpower to commanders who have authority over

a territory and its resources. Inducing commanders to follow orders is a challenge

for leaders who cannot rely on externally enforced contracts. Rebel leaders achieve

cooperation through on-the-spot cash payments and promises of future rewards—but

these promises are only credible if personal bonds create trust between the leader and

commander.

In the vignettes, Johnson T. Leamah made great efforts to implement Charles

Taylor’s policies because he was handsomely rewarded by Tayor during the war and

also expected to benefit afterwards. Thomas Karla, by contrast, devoted his efforts to

diamond mining and looting because he distrusted ULIMO’s leader Alhaji Kromah

and received no financial support from him. The contracts between leader and com-

mander, while crucial for the well-being of the population, is subject to a range of

agency issues that have not yet been systematically analyzed.

The third theme of the dissertation is the role of external patrons in supporting

and influencing rebel armies. External patrons have access to the financial resources,

military supplies, and logistics necessary for large-scale rebellion. While many ac-

counts of insurgent success point to the importance of foreign support, the varied

3Factionalization statistics based on the RAFD dataset compiled by the author. Chapter 4provides a detailed description of the coding methods and analyzes the data.

1.3. THE ARGUMENT 7

interests in the civil war outcomes by foreign patrons have not been sufficiently rec-

ognized. Some patrons have an incentive to support qualified, trustworthy leaders

and provide resources to strengthen leader control, while others benefit by installing

weak leaders who can be more easily manipulated to serve the patron’s interests.

1.3 The Argument

I argue that the level of security provided by a rebel group and the group’s risk of

factionalization both depend on the leader’s ability to offer incentives to top comman-

ders, and that the leader’s ability to offer these incentives depends on the motives of

external patrons that supply resources to the group. Leaders exert effective control

over their commanders when they can offer on-the-spot cash payments and credible

promises of future rewards. Effective control induces commanders to maintain a dis-

ciplined fighting force and work towards achieving group goals, rather than personal

profit. Control is weakened in rebel groups when commanders have access to valuable

lootable resources or when the commander’s actions are not easily observed by the

rebel leader. These commanders are less likely to train and discipline their forces and

have greater incentives to use group resources for opportunistic ends.

The relationship between the rebel leader and top commanders is shaped by the

influence of external patrons. Some external actors, such as diaspora communities,

benefit from disciplined rebel groups with strong leaders. These actors have prefer-

ences that align with those of the rebels and are motivated to provide ample financial

and military resources to qualified leaders who are trusted by their troops. The re-

sult is a more cohesive, disciplined rebel army. Other patrons, whose preferences

conflict with those of the group, maximize their leverage over the group’s activities

by supporting relatively weak leaders and withholding financial resources that could

strengthen leader control. Unable to offer cash incentives or credible future rewards,

these leaders allow their troops to loot locals in exchange for a minimal level of loyalty

and are thus less able to prevent factionalization. Figure 1.1 depicts the theoretical

model.


1.4 The Mechanisms

The fundamental challenge of rebel organization is the lack of externally enforced

contracts among rebel members. This challenge distinguishes rebel groups from other

organizations such as corporations or government agencies. Consider the owner of a

factory. The owner relies on managers and workers to produce goods, and these

employees follow orders in exchange for a salary. If the owner fails to pay the salaries,

the workers can sue for damages. Similarly, if the manager steals from the factory,

the owner can call the police and have the manager arrested.

In some ways, a rebel leader is similar to the factory owner. The rebel leader has

certain goals and must recruit commanders and soldiers to achieve those goals. But

unlike the factory owner, a rebel leader cannot write contracts. Commanders and

soldiers cannot sue the rebel leader if they fail to receive their promised salaries or

rewards. And the rebel leader cannot call on the police if commanders decide to steal

group resources or refuse to follow orders.

Rather than formal contracts, rebel groups rely on social networks and on-the-spot

rewards. Social bonds between the leader and group members, such as family ties or

a history of pre-war interaction, create trust and allow for credible promises of future

rewards. Leaders also use cash payments to convince members to implement orders

or to reward them for good behavior. The most effective leaders are both trusted by

their troops and possess substantial cash resources.

My core insight, based on several years of careful observations in civil war zones,

is that civilian welfare and rebel discipline depend on the interaction between rebel

leaders and their top commanders. Rebel leaders generally have an incentive to cre-

ate security in their territory because it allows for reliable access to food, military

intelligence, and new recruits. Security also facilitates large-scale resource extraction,

humanitarian aid operations, and other activities. The provision of security in rebel

territory depends on the rebel commanders, who have access to significant weapons

and resources and directly monitor the behavior of soldiers. Although security pro-

vides benefits to the rebel leader, commanders might find more benefit in looting or

exploiting resources than in the complex administrative tasks required to maintain

1.4. THE MECHANISMS 9

order.

Rebel leaders exert effective control over their commanders when they can offer

on-the-spot cash payments and credible promises of future rewards. When properly

motivated in these ways, commanders have incentives to train and discipline their

soldiers and organize systems of food distribution. When leaders cannot offer these

incentives, commanders are more likely to allow soldiers to loot and use group re-

sources for personal enrichment at the expense of group goals. Leader control is

further weakened when commanders have access to valuable lootable resources such

as alluvial diamonds, which increase the opportunity cost of following orders. The

leader has little power to curb this predatory behavior unless she4 can offer strong in-

centives for cooperation; attempts to punish the troops could trigger factionalization

or large-scale defection from the group.

This interaction between leader and commander is an example of a principal-

agent relationship, similar to standard economic models and previously analyzed in

the context of rebellion by Scott Gates (2002).5 The assumptions of the theory are

simply that rebel members are rational and respond to incentives. When rebel leaders

provide greater incentives, commanders are more willing to undertake costly actions

on behalf of the leader; when the opportunity costs of following orders become steeper,

commanders are less willing to implement the leader’s orders.

But this relationship, by itself, is only a partial explanation for rebel behavior.

The deeper question is: Why do some leaders have access to financial resources and

credible promises, while other leaders do not?

To become a force capable of challenging the state, a rebel group needs access

to weapons, ammunition, food, and manpower. To some extent, rebel groups can

generate these resources on their own. Family members and friends can form the core

group of supporters. A daring raid on a police station or army barracks can yield

some weapons and ammunition. And a network of sympathizers can provide food

4In the dissertation, I refer to generic rebel leaders as “she”, commanders as “he”, and externalpatrons as “it”.

5The theory presented here differs from Gates’ model by adding an expectation of future rewardsand endogenizing the rebel leader’s access to resources by including an external patron as a strategicactor. Gates, however, also considers the effects of competition with government forces, a factorthat is not considered here.


and safe houses. For decades, small bands of fighters can evade capture and remain

a thorn in the side of the government.

Expanding beyond a small-scale insurgency, however, usually requires the assis-

tance of an external patron. In the dataset compiled for this study, I find that more

than 77% of the 74 rebel groups that were powerful enough to control territory be-

tween 1980 and 2003 depended on an external patron to supply money or military

equipment.6 Large-scale military offensives require complicated logistics and regular

shipments of ammunition. To acquire these supplies, rebel groups either need access

to cash and arms dealers, or the support of a foreign government. Without external

support, few rebel groups emerge from obscurity.

The importance of external support allows patrons to influence the rebel group

and its activities. In the most extreme cases, external patrons select the rebel leader

and decide where the forces will be deployed. Even when the patron does not play

such an active role, rebel groups have an incentive to select leaders who are acceptable

to external actors and thus most likely to attract outside support.

Patrons provide support to rebel groups in exchange for access to resources and

commercial opportunities, or to achieve policy goals. These motives have important

implications for what kind of leader the patron will support and the amount of re-

sources the patron will provide. Patrons that benefit from disciplined rebel groups are

more inclined to support well-qualified leaders, trusted by their men, and are more

likely to supply money and resources to strengthen the leader’s control in the group.

These patrons also have an incentive to pressure the leader to maintain discipline

among the rebel forces. A timber company, for example, would not tolerate looting

of its machinery by the rebels it supports. And a diaspora community might recon-

sider whether to send money to rebels that abuse their family members. In these

situations, the incentives of the patron and the leader are aligned, and they work

together to create a cohesive, committed rebel force.

6Other studies also highlight the importance of external support for rebellion. Foreign govern-ments provided military support to 44.1% of the 331 non-state armed groups that were active during1945-2003, excluding coups, according to the EACD dataset (Cunningham et al 2009). RAND’s(2001) survey of insurgent movements finds that 44 of the 74 rebel groups (59.5%) active during1991-2000 received state support that was crucial for their survival. These figures ignore the role ofcorporations, arms dealers, and other economic actors in civil war.

1.4. THE MECHANISMS 11

Other patrons have goals that are not in line with those of the rebels. A foreign

government, for example, might support rebels to gain leverage over a neighboring

state. To use the rebels as a bargaining chip, the patron needs to be able to influence

the group’s activities. If a rebel leader becomes too powerful, the group might deviate

from the patron’s plans and any political gains for the patron would be lost. In such

cases, the patron has an incentive to support relatively weak, untrustworthy leaders

who could not maintain power in the group without the patron’s assistance. The

patron provides just enough support to keep the leader in power, but withholds

resources that the leader could use to solidify her position. If the leader ever disobeys

orders, the patron can cut off support and the leader will be overthrown. With few

resources and no credible promises, the leader lets her commanders and soldiers fend

for themselves to acquire food and the spoils of war. These groups prey on civilians

and face a high risk of factionalization.

The few rebel groups that form without external support are able to do so either

because of tight bonds and credible promises among their core members or because

enough weapons and resources are up for grabs in the conflict area. In the first case,

groups built on trust are likely to be disciplined and provide security to civilians

even without access to financial resources. To risk their lives to attack government

positions with few military supplies, the members of these home-grown insurgencies

must be committed to the cause and believe that their actions are not simply enriching

the rebel leader at their expense.

When groups form based on available resources, the behavior of rebel groups

depends on the lootability of those resources. During the collapse of the Soviet

Union, for example, a flood of weapons from large arms caches in Georgia provided the

supplies necessary to form dozens of militias that preyed on civilians (Driscoll 2009).

Unable to control access to weapons or other resources, militia leaders could not

prevent their men from defecting to rival militias or abusing civilians. By contrast,

when Slovenia seceded from Yugoslavia, the rebel leader Milan Kucan maintained

control over the arms caches and used his access to the semi-autonomous Slovenian

government’s finances to provide salaries and maintain discipline among his forces.


The behavior of rebel groups follows from the dynamic interaction between re-

sources and leadership. The leader’s authority in the group depends on her control

of the group’s financial resources and her credible promises of future rewards—and

both are influenced by the leader’s partnerships with external patrons. Leader control

declines when commanders and troops can profit from lootable resources or acquire

their own weapons and ammunition in the conflict area. Rebel group behavior can

change over time if new resources are captured or if patrons reconsider their support.

The decision of patrons to support rebellion owes more to geography and geopolitics

than to the characteristics of the group itself. While better leaders can produce more

disciplined and effective forces, ultimately even the best leaders are at the mercy of

international forces beyond their control.

1.5 Theories of Rebellion

Much of the early literature on 20th century insurgencies was produced by the revo-

lutionaries themselves. These writers, including Mao Tse-tung (1937), Che Guevara

([1961] 1998), and Vo Nguyen Giap (1970), emphasize the importance of civilian

support and cooperation. According to these practitioners, creating systems of gov-

ernance and maintaining popular support are central to the survival of rebel organi-

zations. For decades, scholars assumed that rebel groups would behave with restraint

towards civilians (Laqueur [1976] 2004: 335-7; Rice 1990: 59).

The end of the Cold War challenged this romanticized notion of rebellion and

led some observers to proclaim the beginning of a new era, one in which many rebel

groups are simply criminal enterprises rather than political movements with legiti-

mate grievances and ideologies (e.g. Kaplan 1994). This faulty distinction, however,

emerged from incomplete and biased information, rather than a substantive change

in rebel group behavior (Kalyvas 2001). Throughout history, rebel groups have been

known to abuse and prey on civilians.7

7For example, civilian abuse was rampant among certain militias during the American Civil War(Kalyvas 2006: 108) as well as among royalist forces during the English Civil War (Carlton 1998).

1.5. THEORIES OF REBELLION 13

Recent theories focus on territorial competition, costly signaling, ethnic compo-

sition, and member selection to explain how rebel groups treat civilians. Territorial

competition shapes the economic incentives for rebels to provide order and security.

Mancur Olson (1993) argues that armed groups have an incentive to provide security

for civilians in their areas if the expected revenue from taxation exceeds the expected

value of looting. In areas of stable control, the groups can be considered “stationary

bandits” and will behave much like a normal government, providing public goods in

exchange for taxes. By contrast, when territory is at risk of capture, rebels have little

incentive to create order because this restraint simply creates looting opportunities

for rival groups.8 Groups in unstable areas become “roving bandits,” looting what-

ever they can before moving on to the next target. Further, territorial competition

may lead to a strategy in which the rebel group loots and abuses civilians as a way

of attracting recruits. Faced with multiple predatory groups, civilians become more

inclined to join the rebels to achieve some degree of security and access to food (Azam

2006).

Territorial competition also affects the rebels’ military and security calculations.

Stathis Kalyvas (2006) argues that rebel groups depend on networks of civilian sup-

porters for cooperation and military intelligence. In stable areas without competition,

rebels provide security for the general population and only punish individuals who

refuse to cooperate or who collaborate with rival groups. Territorial competition pre-

vents rebels from acquiring needed information on government collaborators and other

threats while also making cooperation less attractive for the civilians. Mass violence,

rather than selective punishment, is used to deter civilian defection in areas where

rebels are vulnerable to government incursion.9 Theories of territorial competition

imply the following hypothesis regarding rebel behavior:

Competing Hypothesis 1. Civilian abuse is higher in contested areas than in

stable areas.

Rebel groups may also commit abuse as a costly signal to the government or

8Jean-Paul Azam (2002) presents a model of competitive looting that captures this dynamic.9This summary is perhaps an oversimplification of Kalyvas’ theory, which also includes the bal-

ance of power between governments and insurgents as well as local feuds among the civilian popu-lation.


external patrons. Jessica Stanton (2009) argues that rebel groups are more likely

to abuse civilians when they face democratic governments, which are sensitive to

civilian losses. This predation signals that the war will be costly for the government,

and therefore increases the government’s willingness to offer concessions.10 Civilian

abuse can also be used by rebel groups to signal their commitment to external patrons

that supply weapons and financing. By destabilizing areas and causing population

displacement, the rebel group demonstrates its ability to make use of the patron’s

resources (Hovil and Werker 2005). Rebel groups may also intentionally create a

humanitarian emergency in an effort to gain international attention for their cause

(Bob 2005). According to these views, civilian abuse is costly to the rebels, who would

benefit from greater civilian support, but is instrumental for achieving the group’s

larger objectives. Theories of costly signaling imply the following hypotheses:

Competing Hypothesis 2. Rebel groups commit more civilian abuse when they

face democratic regimes.

Competing Hypothesis 3. Rebel groups commit more civilian abuse when they

receive support from an external patron.

Ethnic composition influences rebel behavior by shaping how rebel members in-

teract with each other and with civilians. Ethnic groups embody pre-existing social

and authority relations. By defining who is eligible to receive benefits, ethnicity

helps create patron-client networks that secure stable bases of support (Bates 1983;

Fearon 1999). Ethnic homogeneity also facilitates the spread of information and

serves as a coordination device (Fearon and Laitin 1996; Habyarimana et al 2009).

As a result, ethnic homogeneity facilitates monitoring and control within a rebel

group. In their study of Sierra Leone, Macartan Humphreys and Jeremy Weinstein

(2006) find that ethnically homogenous rebel units were less likely to abuse civilians

than ethnically-mixed units. Ethnic ties with the local communities also strengthen

a group’s incentives to protect civilians through these social mechanisms. Ethnicity-

based theories imply that more homogeneous rebel groups—or more homogeneous

10Robert Pape (2003) proposes a similar argument in the context of suicide terrorism. Suicideterrorism, Pape argues, is an especially effective weapon against modern liberal democracies becausetheir citizens have a low tolerance for civilian casualties and a great influence on government policy.

1.5. THEORIES OF REBELLION 15

units within diverse groups—provide more security for civilians. Also, groups are

more likely to provide security when they share ethnic ties with the local community.

While some ethnically-based groups may seek to cleanse their territory of non-coethnic

civilians, such abuse is much more common among government forces and paramil-

itaries (Valentino 2004; Valentino et al 2004). In general, ethnicity-based theories

imply the following hypotheses:

Competing Hypothesis 4. Ethnically homogeneous groups commit less civilian

abuse than diverse groups.

Competing Hypothesis 5. Groups with ethnic ties to local communities com-

mit less civilian abuse than groups that lack such

ties.

The selection of rebel members also shapes group behavior. Jeremy Weinstein

(2007) argues that rebel groups with access to economic resources attract oppor-

tunistic recruits who are driven by short-term goals. These recruits are difficult to

control and are more likely to prey on civilians. Rebel groups that lack resources rely

instead on appeals to shared identity and attract committed, disciplined recruits.

The initial endowments of the rebel group—both economic and social—shape the

character of the group from the earliest days of the conflict. Groups that acquire bad

reputations find it difficult to change their behavior. And groups that behave with

discipline early on must constantly reaffirm this commitment to civilians and their

own members. This theory implies that greater access to economic resources leads

to an adverse selection of opportunistic members, which results in more predatory

behavior by the rebel group, summarized in the following hypothesis:

Competing Hypothesis 6. Groups with access to economic resources commit

more civilian abuse than resource-poor groups.

The topic of rebel factionalization has received less attention in the literature,

despite its implications for civil war duration and the feasibility of international in-

tervention. Paul Staniland (2010) proposes the most sophisticated explanation for

armed group factionalization. Armed groups are most likely to factionalize when

they form around fragmented social networks and have access to material resources.


Material resources increase the temptation for group members to disobey the leader’s

orders or break from the group entirely when members are not linked to the leader

through strong social networks. When strong networks exist, material resources can

strengthen control and cohesion in the group. Explaining ex ante why some groups

form around fragmented networks while others have access to tightly-knit social bases,

however, remains a challenge and prevents the specification of a testable hypothesis.

1.6 Organizational Behavior

Existing theories add some insight into the logic of rebel activities, but do not consider

the ways in which organization and hierarchy shape the behavior of rebel members.

A fundamental assumption of neoclassical economics is that individuals respond to

incentives. These incentives could be monetary benefits, social rewards, or some

other form of compensation that is valued by the rational actor. Organizations create

incentives and shape individual behavior in two ways, through delegation and through

socialization.

Models of delegation concern themselves with a principal and at least one agent.

The principal decides which tasks to assign and the level of compensation. The

agent then decides how much effort to spend on the task, i.e. whether to work or

shirk.11 Faced with this delegation problem, economic theories of the firm focus on

how to construct incentive contracts for controlling subordinates who are assumed to

be “self-interest seeking with guile” (Williamson 1983: 354).

The most obvious incentive contracts involve cash payments or salaries. Ceteris

paribus, individuals take actions that increase their access to money and decrease

the risk of fines and other losses. The payments provided by the principal determine

the benefits of a given task, which the agent weighs against the cost of effort and

the value of other opportunities. Higher compensation generally leads to more effort

and better performance (Gneezy and Rustichini 2000). But salaries and performance-

based bonuses can sometimes be counter-productive, resulting in lower performance

11For a treatment of the foundational principal-agent model, see Mas-Collell, Whinston, and Green(1995). For a review of delegation theories see Bendor, Glazer, and Hammond (2001).

1.6. ORGANIZATIONAL BEHAVIOR 17

by crowding out intrinsic motivation (Frey and Jegen 2001).

Much stronger than monetary incentives is the desire to stay alive. Individuals

will do almost anything to secure access to food or avoid physical harm. For example,

a soldier might engage in looting in the hopes of finding food and other vital supplies.

The same soldier, at a later time, might follow orders to avoid being beaten by his

commander. In this case, behavior that appears contradictory—opportunistic one

moment, disciplined the next—results from the soldier’s need for self-preservation.

Desperation and fear are powerful motivators.

Social incentives also shape behavior in organizations. A person might work long

hours at a job she dislikes to earn the respect of her family and community. Another

person might rush into a burning building to save a friend or neighbor. The desire

for social approval, and fear of social sanctions, are often more powerful motivators

than monetary incentives. Piece-rate workers, for example, have been known to re-

strict their output and accept lower pay to avoid social sanctions from their coworkers

(Elster 1989). The intensity of these social incentives depends on the relationship be-

tween the individual and others in the organization. Social motivations are strongest

when an individual hopes to live up to the expectations of an admired boss or a peer

group (Ellingsen and Johannesson 2008) or when the individual is dependent on the

group and has few outside options (Hechter 1987).

Social motivations are especially important during conflict. Scholars have long

considered the paradox of soldiers risking their lives in battle. Rational actors should

not take such great risks for the modest salaries that most soldiers receive. Stouffer et

al’s (1949) study of combat soldiers in World War II finds that soldiers were largely

motivated by a desire to live up to the expectations of their combat unit. Recent

studies of combat motivation continue to emphasize the central role of social incentives

in the lives of soldiers (Wong et al 2003; MacCoun et al 2006).

This need for social acceptance makes comprehensible some of the massacres,

gang rapes, and mutilations that characterize many wars. For example, Christopher

Browning (1992) argues that many of the atrocities of the Holocaust were committed

by “ordinary men” who were motivated by the desire to fulfill the collective obligations

of tightly knit groups. In an intriguing formal model, Laurence Iannacone (1992)


shows how requiring members to undertake costly or painful actions can increase the

social benefits of group membership. Indeed, Dara Cohen (2010) argues that some

rebel groups require their members to participate in gang rape as a means of building

social cohesion, thereby increasing social motivation.

The individual’s position in the hierarchy determines the relative effectiveness of

monetary versus social incentives. People at the bottom of the organization, soldiers

for example, tend to operate in small groups, which create opportunities for strong

social ties. Effective organizations implement policies to strengthen these social mo-

tivators and can inspire high effort from their members with few material rewards

(Akerlof and Kranton 2005). Higher up the organizational chain of command, social

motivators become less important relative to individual incentives such as salaries

and bonuses. A top manager, for example, might oversee hundreds of employees but

interacts with few equals in the organization. Unlike her employees, the manager

does not have the opportunity to form strong social bonds with her coworkers. In

rebel groups, top commanders must be motivated through individual incentives such

as monetary payments and the promise of future rewards, while soldiers can be mo-

tivated through social incentives, the provision of basic necessities such as food, and

the threat of punishment.

In rebel groups, commanders have many tools for controlling soldiers, but leaders

have few options for controlling their commanders. Once commanders have been

given weapons and manpower, the leader is constrained in her ability to influence

their actions. Only when the leader controls access to substantial financial resources

or valuable future rewards will commanders take costly actions on the leader’s behalf.

Understanding how leaders motivate commanders to follow orders, and why some

leaders are more successful than others, is the key to understanding rebel group

behavior.

1.7 Research Design, Data, and Results

The roots of this theory of commander incentives grew out of several years of inter-

views and careful observations in the conflict zones of Sri Lanka, Liberia, and DR

1.7. RESEARCH DESIGN, DATA, AND RESULTS 19

Congo. These insights are formalized into a model, which produces three main ob-

servable implications that can be tested. First, the rebel leader’s access to financial

resources and credible promises drives the level of civilian abuse in the group’s ter-

ritory as well as the group’s risk of factionalization. Second, civilian abuse follows

a spatial pattern that depends on factors such as distance to rebel bases and the

presence of lootable resources. Third, external patrons with commercial or social in-

terests in a conflict encourage the formation of disciplined rebel groups, while patrons

with conflicting interests support more predatory groups.12 The theory also produces

other insights into civil war. For example, the theory predicts a greater risk of group

factionalization during prolonged peace talks because rebel leaders must decide how

to allocate rewards to group members based on imperfect information.

I employ qualitative and quantitative tools to test this theory. For a cross-national

analysis, I compiled a dataset of all rebel groups that controlled territory at some point

during 1980-2003. Using news archives, historical texts, U.N. reports, and declassified

documents I uncovered the resources available to each rebel group during each year

of the conflict. I also include information on the identities and motives of external

patrons, the timing of peace negotiations, post-conflict rewards, factionalization, and

civilian abuse. The result is the Resources, Abuse, and Factionalization Dataset

(RAFD), which includes 73 armed groups in 41 countries and 768 group years.

In the cross-national study using the RAFD dataset, I show that the leader’s

access to financial resources and the credibility of her promises are strongly associated

with lower levels of civilian abuse and a lower probability of group factionalization.

Valuable lootable resources correspond to higher levels of civilian abuse and more

likely factionalization, and peace talks also increase the risk of factionalization over

time. The influence of external patrons on group behavior is also significant. Patrons

with ethnic or social motives are more likely to support trustworthy leaders and supply

them with greater financial resources. These groups in turn provide greater security

to civilians and are less prone to factionalization than groups supported by foreign

governments or other actors that seek to manipulate the group. To strengthen the

inference, I use exogenous variables related to pre-war natural resources as proxies

12Social interests may include ethnic ties, ideological motivations, or humanitarian objectives.


for leader financing and valuable lootable resources while excluding the confounding

effects of unobservable group characteristics.

A detailed analysis of Liberia’s civil war, 1989-2003, allows for an exploration of

the effects of variation in access to financing and external support among four main

rebel organizations by holding constant factors such as geography and culture. This

sub-national, controlled comparison of groups that vary on key independent variables

strengthens inference and accounts, to some extent, for selection bias in the case study

(Collier and Mahoney 1996; Snyder 2001). Through this comparison, the relationships

between external patrons, rebel leaders, and commanders can be examined, and the

effects of rebel organization can be traced down to the ground level, where civilians

are either protected or preyed upon.

To tell Liberia’s story and assess the value of the theory, I spent fourteen months

reconstructing the day-to-day lives of civilians during the war and the internal politics

of the rebel groups involved. Such qualitative research carries a risk of bias for the

researcher. Some informants, by virtue of their charisma or social connections, can

have an undue influence on the researcher’s image of the conflict. To mitigate this

bias I interviewed civilians in 30 locations in Liberia and reconstructed the local

history of the conflict in each location. Ten of these locations were randomly selected

in northwestern Liberia, which, at various points, was controlled by NPFL, ULIMO,

and LURD. Through this process I also interviewed dozens of rebel soldiers. By

journeying to remote villages, large towns, and key strategic locations I developed a

qualitative picture of civilian security and abuse during Liberia’s war, as well as an

understanding of civilian coping mechanisms.

In addition to this village-level fieldwork, I interviewed 50 rebel leaders, high-level

commanders, and key economic actors involved in Liberia’s civil war. These infor-

mants were recruited through social networks and contacts that I developed over five

years of experience in Liberia. By reconstructing the organizational charts of each

rebel group, I compiled a comprehensive list of high-level rebel members, their per-

sonal characteristics, locations during the war, and post-war positions. Working from

these organizational charts, I managed to locate and interview 75% of all surviving

rebel leaders and top commanders. Of the 25% of informants that I was not able

1.7. RESEARCH DESIGN, DATA, AND RESULTS 21

to interview, nearly all have testified to the Truth and Reconciliation Commission of

Liberia or the Special Court of Sierra Leone. Taken together, this information pro-

vides a comprehensive picture of the internal politics of Liberia’s rebel organizations.

The appendix provides additional information on the fieldwork, as well as a map of

fieldwork locations.

Over the course of these interviews I heard many contradictory stories and mul-

tiple interpretations of the same event. For significant events, I tracked down all the

key people involved to understand each side of the story. Reconciling these various

perspectives would be impossible, but I often found common ground among the in-

formants’ stories. Throughout the dissertation I make an effort to present what I

believe is the most accurate version of an event and acknowledge any ambiguities

in footnotes. Interviews are cited by letter and number and interested readers can

consult the appendix for a description of each informant.

To complement the qualitative fieldwork, I use satellite images from various stages

of Liberia’s war to measure changes in crop land as a proxy for insecurity. When rebel

groups are predatory, civilians abandon their farms and seek shelter in the forest or in

refugee camps. Using remote sensing and spatial statistics, I compare changes in crop

area in rebel territories and uncover the spatial patterns of security and predation. In

Liberia, the theory predicts that the NPFL rebels provided security to civilians due

to Charles Taylor’s partnerships with foreign corporations that demanded security,

and his ability to offer cash incentives to his troops. The other rebels—ULIMO, LPC,

LURD, and MODEL—are predicted to be predatory because the leaders lacked the

ability to create incentives for their troops. Through remote sensing methods, I show

that Liberia’s NPFL rebels provided enough security to maintain farming at pre-war

levels in many areas, while the LURD rebels triggered declines in crop land of up to

66% and the LPC rebels triggered declines of over 18%.

This theory also predicts a spatial pattern to security in conflict zones. When a

rebel leader is able to offer cash incentives to enforce discipline, the effectiveness of

these incentives depends on the ability to monitor troop behavior. Rebel members

will be more disciplined and provide more security near the rebel headquarters, under

the watchful eye of the leader. By contrast, if the leader cannot offer cash incentives,


monitoring is ineffective and rebel members prey on civilians. The most intense

looting will occur near rebel bases, where the highest concentrations of armed men

are located. For all groups, valuable lootable resources raise the opportunity costs of

following the leader’s orders and result in less security for civilians.

Through various methods, I show that the LURD and LPC rebels were more

predatory near their bases, while the NPFL provided a higher degree of security.

Holding constant other factors, if a farmer 100km away relocated to LURD head-

quarters, he would suffer from a 70% or greater decline in crop area. By contrast,

a farmer relocating to the NPFL headquarters from 100km away would not lose any

crop land and may even benefit. According to some estimates and controlling for

other factors, the farmer’s crop area would increase by more than 200% due to the

move. For all groups, lootable resources, most notably rubber, are associated with

greater declines in crop area.

Finally, to overcome the identification problem caused by the endogenous selection

of rebel bases I examine changes over time in food production near Buchanan, a major

port in Liberia’s southeast. Buchanan was the secondary headquarters of the NPFL

from 1990 until 1994, when the town became the secondary headquarters of the LPC

rebels. Consistent with the theory’s predictions, food production benefitted from

proximity to Buchanan during NPFL rule and suffered from this proximity after the

town fell to the LPC. By holding geographic factors constant, this shift in security

isolates the effect of rebel behavior from potential omitted variables. The results of

these analyses provide strong support for the theory.

1.8 Scope Conditions and Limitations

This theory is bounded by two important scope conditions. First, the theory applies

to armed actors who operate openly. The theory does not apply to secretive organi-

zations such as gangs, mafias, or terrorist cells. The reason for this condition is that

the model does not account for the challenge of maintaining secrecy or the risks of

denunciation. Terrorist organizations face a trade-off between secrecy and control of

their operatives (Shapiro 2007) and small-scale insurgents are easily apprehended if

1.8. SCOPE CONDITIONS AND LIMITATIONS 23

civilians inform the government of their identities and whereabouts (Fearon 2008a;

Kalyvas 2006). In large-scale rebellions, by contrast, the location of the rebel forces

and even the identity of rebel members are well known by government officials, but

the state is unable to arrest them.

Second, the theory only applies to armed groups that control territory. The central

mechanism of the model concerns the delegation problem that exists between rebel

leaders and their top commanders. This delegation problem depends on some spatial

distance between these actors. The rebel leader delegates weapons and manpower

to a commander who then has some degree of autonomy over a given territory. The

rebel leader cannot physically force the commander to follow orders. By contrast, if

the rebel leader and commanders share a forest hide-out, the leader would be better

able to monitor and pressure for commander compliance.

It can be argued that the theory suffers from other limitations as well. The the-

ory’s focus on economic incentives seems to exclude important, non-economic aspects

of rebellion such as the group’s ties with local communities and its ideology. But de-

spite the focus on economic resources, the theory provides insight into the ideological

and social aspects of rebellion by emphasizing the leader’s ability to offer incentives

to group members. Ideologically-motivated groups form from small, tight-knit or-

ganizations that depend on the mutual aid and sacrifice of their members (Berman

2009). To inspire such sacrifice, these groups depend on charismatic leaders who es-

tablish credibility and trust among group members. By offering credible promises,

these leaders establish control over their troops during war.

Ideology is important for establishing leader credibility, but the theory also implies

that material incentives are important for sustaining ideology. Implementing an ide-

ology requires coordination and discipline among the forces. As groups expand, they

accept new recruits who were not original members of the tight-knit club and thus

may not have the same attachment to the group’s ideology. Without leader control,

predatory behavior by some individuals would undermine the ideological program

of the group as a whole. When rebel leaders cannot offer strong incentives, rebel

members fend for themselves and demand food from an already beleaguered popu-

lation, thereby straining relations with the communities. Even if many of the rebels


are committed to an ideology, they could not prevent abuses by their less-committed

comrades. Although the theory focuses on economic incentives, it adds insight into

the role and limitations of ideology in rebel organizations.

An important assumption of the theory is that the rebel leader has an incentive

to provide security for civilians in the conflict areas. This assumption might seem

too strong, given the ethnic cleansing and genocide that characterized conflicts in

countries such as Bosnia and Sudan. I argue, however, that leaders almost always

have an incentive to provide security for civilians in their territory. Rebel leaders

benefit from security in three ways. First, security is vital for local food production.

Food is expensive to transport and perishable. If rebel groups rely on outside food

shipments, even minor disruptions in the delivery schedule would threaten the survival

of the group and jeopardize its military operations. Acquiring food locally is the

most effective method of supplying the troops. Local food production also benefits

civilians, who would otherwise flee the conflict areas, depriving the leader of military

intelligence and reducing international support for the rebels’ cause.

Second, security allows rebel groups to generate revenue by taxing local trade. The

charcoal trade in eastern DR Congo, for example, produced several million dollars in

tax revenue for local armed groups during 2007-8 (UNSC 2008: 8, 28). Narcotics are

even more valuable local products. The 2007 poppy harvest in Afghanistan yielded

an estimated 8,200 tons of opium, valued at $4 billion (UNODS 2007); during the

same year in Colombia, an estimated 600 tons of cocaine were produced, valued at

approximately $1.3 billion (UNODS 2009). Protecting local producers and trade

routes allows rebels to take a share of these lucrative markets.

Third, security allows for corporations and humanitarian organizations to operate

in rebel territory. Rebel leaders can earn millions through partnerships with foreign

corporations, but these businesses cannot operate in the conflict areas if their equip-

ment might be looted or their staff kidnapped or killed. Humanitarian organizations

also depend on security for their operations, and benefit rebels by providing access

to medical supplies, food, and other resources. Ethiopia’s TPLF rebels, for example,

generated millions in revenue during the mid-1980s from humanitarian organizations

responding to the famine (BBC 2010; CIA 1985).

1.8. SCOPE CONDITIONS AND LIMITATIONS 25

In certain circumstances, large-scale abuse such as systematic looting, forced labor,

or ethnic cleansing can be useful tools for increasing political pressure, redistributing

rewards to supporters, or as emergency sources of revenue. But these strategies are

rarely optimal for the rebel leader. Systematic looting, for example, may be a sensible

economic strategy in areas of unstable control or when the group needs emergency

funds, but quickly threatens the group’s food supply as stores are depleted and civil-

ians flee the area. Ethnic cleansing might be a useful tool for strengthening support

among a core constituency, but only when rebels can target an easily identifiable and

wealthy minority group. The majority of organized, large-scale violence is carried out

by government forces, not rebels, and usually occurs only after the leader exhausts

other strategic options (Valentino 2004; Valentino et al 2004). My argument does not

deny the possibility that some rebel leaders plan and encourage large-scale violence.

The leader’s preference for security is an assumption of the model, not a conclusion.

If the model has value, its predictions will account for much—but not all—of the

variation in rebel behavior.

Further, the theory does not attempt to explain a rebel group’s chances of success.

Although rebel group cohesion and discipline likely play some role in its ability to

capture the state, other factors such as the group’s access to arms and the capacity

of the government to suppress rebellion also play a major role. Even if a rebel leader

exerts only minimal control over the troops, the group might have access to enough

weapons and logistical support to reach the capital city of a weak state. The military

success of rebel groups depends on factors beyond the scope of this theory.

Finally, there is a potential omitted variable bias. The theory predicts that trust-

worthy, high-quality leaders will tend to attract support from patrons that have a

stake in the group’s success and provide financial resources, resulting in a disciplined

fighting force. Group behavior, according to the theory, arises from the leader’s cred-

ibility as well as her access to financial resources. But suppose some other factor

creates rebel members with an inherent preference for discipline, and that this same

factor creates rebel leaders who are better able to attract financing from external pa-

trons. The result would be a disciplined group with access to financial resources, but

there would be no causal effect between these two characteristics. This dissertation


cannot conclusively address this potential bias. But through statistical techniques

and a careful analysis of Liberia’s rebel groups, the dissertation provides compelling

evidence to support to the theory’s claim that the leader’s credibility and access to

finances shape group behavior independent of the underlying preferences of group

members.

1.9 Overview of the Dissertation

The dissertation is organized into eight chapters. Chapter 2 looks within rebel orga-

nizations and analyzes the interactions between leaders, commanders, and soldiers.

The chapter argues that rebel group behavior is driven by the leader’s ability to offer

spot payments and credible promises to top commanders. These incentives encour-

age commanders to train, feed, and discipline their forces while closely following the

leader’s orders, resulting in lower levels of abuse and less risk of group factional-

ization. The chapter then formalizes the relationship in a model involving a leader

and commander. The leader sets a policy and offers a spot payment to the com-

mander, who also has some expectation of a future reward. The commander decides

whether to accept or reject the leader’s offer, given his access to lootable resources.

The model is extended in various ways to consider the effects of leader-commander

affinity, imperfect monitoring, private information, and multiple commanders.

Chapter 3 examines the role of external patrons in rebellion. The chapter discusses

the multitude of actors that become involved in civil war, with special emphasis on

states, corporations, and diaspora communities. These patrons can be characterized

by the extent to which their interests align with the goals of the rebel group. The

formal model from the previous chapter is extended to consider how the motives

of these patrons shape the characteristics of the rebel group. The game allows the

patron to transfer resources to a rebel leader, who then uses the resources to provide

incentives to the troops. The analysis shows that patrons with aligned preferences

support more trustworthy leaders while patrons whose interests conflict with those

of the group support less trustworthy leaders. The model is extended to consider

commitment problems between patrons and rebel leaders as well as multiple patrons.

1.9. OVERVIEW OF THE DISSERTATION 27

Chapter 4 assesses the theory’s predictions using a dataset of all rebel groups that

controlled territory at some point during 1980-2003. The dataset provides insight

into both civilian abuse committed by rebel groups, as well as a group’s risk of

factionalization. The results provide strong support for the theory. Groups in which

the leader has greater access to financial resources and/or credible promises commit

less civilian abuse and face a lower risk of factionalization, while lootable resources

lead to higher abuse and greater risk of factionalization. External patrons shape

the group’s behavior through the leader’s credibility and the provision of financial

resources, and their motives are systematically related to group behavior.

Part II of the dissertation traces the theory’s causal mechanisms and tests its

predictions through a detailed study of the rebel groups that participated in Liberia’s

civil war, 1989-2003. Chapter 5 presents a historical overview of Liberia, from its

founding in 1822 by freed American slaves to the 2005 post-conflict elections. This

concise summary attempts to synthesize the best available primary and secondary

sources. The chapter also draws on information from interviews with key actors to

fill gaps in the existing literature.

Chapter 6 presents a narrative of the internal politics of Liberia’s rebel organizations—

NPFL, ULIMO, LURD, and MODEL—and traces the causal mechanisms of the

model through the groups’ histories. The chapter details how each group formed

and the role of external patrons in their early expansion. The chapter continues with

a discussion of the resources available to the rebel leaders and the leaders’ attempts

to control their commanders. Key challenges in the groups’ histories, such as fac-

tionalization, receive special attention. Also discussed are the leader’s incentives to

provide security to civilians and the commanders’ efforts to control their troops.

Chapter 7 turns to the situation for civilians during Liberia’s war. The chap-

ter discusses food security, domestic trade, and humanitarian aid, outlining general

trends as well as specific anecdotes from villages where fieldwork was conducted. The

chapter then turns to remote sensing to measure changes in crop land for a spatial,

quantitative analysis of security during the war. The patterns of changing crop land

are consistent with the predictions of the theory. In NPFL territory, security pre-

vailed in the vicinity of the rebel headquarters, but declining crop land is observed


near the front lines and in areas that contain lootable resources. In LURD territory,

by contrast, catastrophic declines in crop area are observed throughout the group’s

territory and the situation was especially severe near the rebel bases, due to the high

concentration of armed and predatory soldiers in those areas.

Chapter 8 concludes the dissertation with a summary of the main contributions

and some policy implications for how international actors can mitigate the humani-

tarian costs of civil war and help create durable peace agreements.

1.9. OVERVIEW OF THE DISSERTATION 29

Leader

Spot Payments

Credible

Future Rewards

Civilian Security

Risk of Factionalization

CommandersLootable

Resources

PatronSelection Pressure

Financial Resources

Soldiers

Training

Discipline

Food Distribution

Figure 1.1: A principal-agent model of rebel organization.

Chapter 2

Leaders, Commanders, and

Soldiers

Rebel group discipline depends on the interaction between leaders and their top com-

manders. When rebel leaders can provide on-the-spot cash payments and credible

promises of future rewards, they motivate commanders to work towards achieving

group goals. Motivated commanders train, monitor, and feed their troops and provide

security for civilians. Commanders, however, might be tempted to use their weapons

and manpower for personal profit, or simply let the soldiers fend for themselves. This

chapter discusses how leaders incentivize commanders through spot payments and

future rewards and considers the effects of lootable resources as well as motivations

such as punishment, status, and ideology. The chapter then discusses the tools used

by commanders to control their soldiers. The second half of the chapter develops a

formal model of the theory and draws out testable predictions of rebel group behavior.

2.1 Spot Payments

Spot payments operate in the same way as salaries or performance-based bonuses in

business firms. By providing money to commanders, rebel leaders encourage com-

manders to deploy their weapons and manpower towards achieving group ends. The

efficacy of spot payments depends on both the size of the transfer and the leader’s

30

2.1. SPOT PAYMENTS 31

ability to monitor the commander’s actions. As with standard economic models,

the more money the leader can offer the commander, the greater the commander’s

willingness to exert effort and implement the leader’s orders.

The key feature of spot payments is that the leader provides rewards greater than

those the commander could acquire on his own. The money for spot payments must

come from resources controlled by the rebel leader and inaccessible to commanders

on the battlefield. Revenues from civilian taxation or alluvial diamonds, for example,

cannot form the basis for spot payments because the leader must rely on commanders

to collect those resources. Without externally-enforced contracts, commanders might

be tempted to keep the revenues for themselves, rather than send a share to the rebel

leader.

Because spot payments must be accessible only to the rebel leader, these funds

almost always come from partnerships with external actors.1 Some of these actors,

such as diaspora communities, international organizations, or foreign governments,

transfer money directly to the rebel leader. For example, the leader of Sri Lanka’s

Tamil Tigers, Velupillai Prabhakaran, received some $50 million annually from Tamil

diaspora groups and other international partnerships (Byman et al 2001: 50). These

financiers are not present in the conflict area and thus cannot be looted or extorted

by troublesome commanders. Commanders profit from these partnerships by earning

rewards from the rebel leader.

Resource extraction corporations provide another source of spot payments. Large

capital investments and specialized knowledge are required to extract and bring to

market resources such as oil, timber, and minerals. Kimberlite diamonds, for example,

are embedded in pipes of igneous rock. To extract the diamonds, explosions loosen

the hard rock, allowing hydraulic shovels to load the ore onto trucks, which then

transport the ore to specialized facilities for processing (AMNH 1997). This process

requires skills and equipment that are not possessed by military organizations. Other

resources are easy to extract but difficult to transport. Timber, for example, can

1Unless the leader is independently wealthy. For example, Fikret Abdic led a rebellion in Bosniathat was financed through his ownership of Agrokomerc, the largest company in Bosnia’s Bihacregion. In 1987 Abdic was accused of issuing worthless promissory notes totaling approximately $1billion (BBC 2001).

32 CHAPTER 2. LEADERS, COMMANDERS, AND SOLDIERS

be harvested with relatively inexpensive equipment, but bringing the hardwood to

market requires a fleet of trucks, an accessible port, and connections to international

markets.

By forming partnerships with multinational firms, rebel leaders earn revenue in

exchange for providing security in the conflict areas. These companies are susceptible

to extortion by rebel members, but generate revenues that exceed the money that

commanders could earn through looting the company’s assets. Rebel leaders create

incentives for discipline and protect the firms’ operations by providing payments to

group members. The UNITA rebels, for example, controlled nearly all of Angola’s

Kimberlite diamond deposits during 1992-7. The largest mining operation in UNITA

territory was run by the Cuango Mining Corporation and used teams of specially

trained divers to strip the Cuango river of its underwater deposits. UNITA provided

security for this and other operations in exchange for a 50% share of the revenue,

which amounted to hundreds of millions of dollars per year (UNSC 2000a: 40-1).

Renegade UNITA commanders could have earned some money by looting the mining

equipment, but this revenue would have paled in comparison to the finances controlled

directly by Jonas Savimbi, the group’s leader.

Rebel leaders can also generate revenue through partnerships with large-scale

traders. During the war in DR Congo’s Ituri district, the leader of the FAPC

rebels, Jerome Kakwavu, allowed James Nyakuni, a Ugandan businessman, to im-

port cigarettes, sugar, petrol, and other commodities into FAPC areas in exchange

for cash payments, gold exports, and vehicle maintenance (UNSC 2005a: 29). Leaders

can personally control the import and export revenue when these goods flow through

a compact geographic area, which facilitates monitoring.2 In Colombia, for example,

cocaine is often exported from remote jungle air strips by private plane. With few lo-

cations to monitor, a rebel leader can observe air traffic and collect payments directly

from the exporters. Rebel commanders might loot the raw cocaine, but they would

have no access to international markets for its sale. Indeed, the FARC leadership

used these strategies to centralize control over annual revenues estimated at $60-100

million and enforced discipline among its troops (Weinstein 2007: 291-2).

2Philippe Le Billon (2001) would define these trades as “point resources”.

2.1. SPOT PAYMENTS 33

Although systematic data on transfers between rebel leaders and their comman-

ders are not available, there is anecdotal evidence that rebel leaders use these funds

to control their commanders. In the West Bank, for example, Fatah leader Yasser

Arafat provided some $50,000 per month to Zakaria Zubaydi, commander of al-Aqsa

Martyrs’ Brigade, to influence the commander’s tactics (BBC 2003). Anecdotal evi-

dence also suggests that spot payments have a significant effect on group discipline. In

Ituri, FAPC leader Jerome Kakwavu generated an estimated $100,000 per month in

revenue by personally overseeing the Ariwara gold market.3 Kakwavu used the gold

revenue to pay salaries to his commanders and successfully maintained discipline

among his ethnically-diverse forces, many of whom were foreign mercenaries (UNSC

2005a: 28). In 2005 Kakwavu left Ariwara to accept a position in the Congolese

military and ceased payments to his troops, who responded by plundering market

shops and civilian homes. The weeks-long looting spree ended only after a series of

violent confrontations with civilian self-defense groups (I25, I26). Neither theories of

ethnic composition nor member selection can explain the changing behavior of the

FAPC’s forces, whose discipline depended on spot payments, not on ethnic networks

or individual preferences.

In the RAFD dataset, rebel leaders had some ability to offer spot payments in 40

of the 73 groups (55%) that controlled territory between 1980 and 2003, although the

amount of revenue available to leaders varies widely. The richest rebel leaders had

access to hundreds of millions of dollars per year. The leaders of El Salvador’s FMLN,

for example, received tremendous financing from communist organizations around the

world (Bracamonte et al 1995). Other rebel leaders generated more modest revenues.

In Uganda, Yoweri Musevi robbed banks to supply the initial financing for the NRA

rebels (Weinstein 2007: 268). Many rebel leaders, however, possess few resources

beyond what is needed to keep themselves alive and cannot offer cash incentives to

their supporters.

The effectiveness of spot payments depends on the leader’s ability to monitor

the commander’s actions. When monitoring is difficult, commanders can behave

3This figure can be estimated from customs revenue data provided in UNSC (2006: par. 52)or from gold trade data provided in HRW (2005: 104); The estimate of $100,000 per month alsoappears in Vlassenroot and Raeymaekers (2004: 398).


opportunistically with less risk of sanctions. The information asymmetry created

by imperfect monitoring means that the rebel leader cannot design optimal incentive

contracts purely based on spot payments. Unless the leader can offer long-term incen-

tives for cooperation, some inefficiencies in group production will always exist. Gary

Miller’s (1992: 198) study of hierarchy reaches a similar conclusion: “Information

asymmetries...make it impossible for managers to realize the full efficiency poten-

tial of team production processes through the manipulation of short-term economic

incentives alone.” Cash transfers cannot fully substitute for the leader’s credibility

and the promise of future rewards when convincing the commander to work towards

achieving group goals.

2.2 Future Rewards

Civil wars end either through military victory or negotiated settlement. Unless de-

feated, a rebel leader will have some ability to allocate post-conflict rewards. A top

rebel commander, therefore, might expect to earn a position in the government or

military, or gain access to lucrative commercial opportunities in the post-conflict pe-

riod; alternatively the commander might hope that the leader implements policies

consistent with the commander’s ideological beliefs. The leader’s ability to inspire

cooperation based on future rewards is the best indicator of leader quality. Credible,

charismatic leaders inspire their followers to make sacrifices and work hard to achieve

group goals.

The most common type of reward for commanders is a position in the post-

conflict government. After the TPLF rebels captured the Ethiopian government, the

top commander, Seye Abraha, became Minister of Defense. In Iraq, the KDP’s top

commander, Sami Abdul Rahman, became Deputy Prime Minister of Kurdistan. In

Kosovo, the UCK’s top commander, Agim Ceku, became head of the Kosovo Protec-

tion Corps. When the EPLF rebels secured independence for Eritrea, commanders

assumed valuable positions as Governor of Asmara, Minister of Health, and Minister

of Defense.

Not all commanders, however, are rewarded for their service. A top commander

2.2. FUTURE REWARDS 35

might hope to become Minister of Defense, but the rebel leader could instead ap-

point a family member or political crony to the position. Even worse, a powerful

commander might be viewed as a threat by a rebel leader who hopes to solidify her

power after capturing the government. Shortly before the AFDL rebels captured the

government of Zaire (now DR Congo), the group’s top commander, Kisase Ngandu,

died under “mysterious circumstances,” allegedly because the soon-to-be-president

Laurent-Desire Kabila feared him as a rival (Turner 2002: 88; Turner 2007: 4). In-

stead of rewarding his commanders, Kabila installed his son Joseph as chief of staff

of the new Congolese armed forces and stacked key positions with his relatives (ICG

1998: fn. 7).

In the RAFD dataset, rebel leaders earned post-conflict rewards in 41 of the 73

groups (56%) that controlled territory between 1980 and 2003. In 24 of these groups

(59%), commanders received a share of those rewards. Commanders are more likely

to put their efforts towards achieving group goals when they believe they will receive

a significant share of the group’s post-war spoils. When commanders do not expect

significant future rewards, they are more likely to pay themselves in the short-term

by looting and preying on civilians.

Without externally-enforced contracts, the commander’s expectation of future

reward depends on the leader’s credibility and personal relationship with the com-

mander. In the uncertain, dangerous environment of civil war, trust often depends

on family ties and a long history of pre-war interactions. Nearly all of the top com-

manders in Sri Lanka’s LTTE, for example, shared family ties with leader Velupillai

Prabhakaran. When the group formed in 1983, Prabhakaran’s cousin, Sadasivam Kr-

ishnakuman, alias Major Kittu, led the group’s military activities until his death in

1992, at which point the top military position passed to Vaithilingam Sornalingam,

another of Prabhakaran’s relatives. When leaders cannot recruit commanders based

on these strong ties, leaders may rely instead on broader networks, such as ethnic

groups, religious communities, or civil society organizations to recruit commanders

who can be trusted—and who will trust the leader. Although these social connections

cannot predict the level of trust between two individuals, they correspond to repeated

interactions and are useful proxy indicators for affinity and credibility.


More difficult to explain or predict is leader charisma. Some leaders are able

to inspire sacrifice and commitment from followers, even without a history of pre-

war interaction. Max Weber (1947: 358) describes charisma as a “certain quality of

an individual personality by virtue of which he is set apart from ordinary men and

treated as endowed with supernatural, superhuman, or at least specifically exceptional

powers or qualities.” Individuals with these qualities incite an emotional reaction from

their followers, and exert their authority without systems of formal rules or abstract

principles. Charisma, in Weber’s view, is an almost mystical quality, in direct conflict

with rationality and bureaucracy; it is also transitory. While charismatic authority

may suffice during the early days of a group, it must eventually be bolstered by a

more rational system of rules and incentives.

Unless sufficiently rewarded, commanders might use their access to weapons and

manpower to split from the group and attempt to secure a better outcome for them-

selves. Among the 73 groups covered in the RAFD dataset, commanders attempted

to break from the leader at least 53 times, and succeeded on 43 occasions. To prevent

factionalization, leaders have some incentive to reward commanders for their service,

even when leaders do not share an affinity with them. In the breakaway republic of

Abkhazia, rebel leader Vladislav Ardzinba appointed his top commander, Sultan Sos-

naliev, as Minister of Defense of the de facto independent government, even though

the commander was Russian, not Abkhaz, and had no personal ties to the leader.

Peace talks are especially challenging for rebel organizations because the leader

must make concrete the value of the post-conflict rewards for group members. Twenty-

one factionalization attempts (49% of all attempts) occurred during peace talks, even

though peace talks took place in only 24% of the group-years covered in the RAFD

dataset. In DR Congo, for example, Wamba dia Wamba, leader of the RCD-K rebels,

secured himself a place in a transitional government but did not reward the faction’s

top commander Mbusa Nyamwisi, whom Wamba had earlier tried to dismiss from

the group (UNHCHR 2010: 218). Nyamwisi split to form RCD-ML and used this

new faction as leverage to secure his own seat in government.

The effects of leader charisma and credibility also trickle down to the soldiers.

When soldiers are inspired by the leader and believe in her long-term promises, their

2.3. LOOTABLE RESOURCES 37

commanders feel increased pressure to follow the leader’s orders. And when soldiers

feel a personal connection with a charismatic leader, commanders are less able to

split from the group because their soldiers might not follow. Credibility and charisma

create positive incentives for discipline at every level of the organization, and decrease

the opportunities for defection among the top commanders.

The leader’s credibility is shaped by the external patrons that support her. In

some cases, the leader is installed as head of the group by these patrons. Rwanda

and Uganda, for example, recruited Laurent-Desire Kabila in 1996 to lead a rebellion

against Mobutu in Zaire (now DR Congo). Even when patrons do not play such

an active role in the formation of the rebel group, groups might select their leaders

based on the individual’s ability to attract outside support, rather than her leadership

qualities. Sudan’s SPLA rebels, for example, selected John Garang as their leader

because of his links with the Ethiopian government, even though other candidates

had stronger ideological credentials and more support from group members (Arop

2006: 67-75). Rebel groups may have to compromise their choice of leader to secure

valuable external support.

2.3 Lootable Resources

Lootable resources undermine discipline in a rebel group by increasing the opportunity

cost of following leader orders and creating incentives to seek personal wealth at the

expense of group goals. Lootable resources also exacerbate tensions within the group

and increase the risk of factionalization. The defining feature of lootable resources is

that they can be extracted by unskilled or semi-skilled labor using only basic tools,

or captured through the use of force. For example, alluvial diamonds—unlike their

Kimberlite cousins—can be extracted from river banks with only a shovel and a

rudimentary sifting device.

Lootable natural resources have gained notoriety after media reports of “blood di-

amonds” in Sierra Leone and gold, coltan, and cassiterite in the Democratic Republic

of Congo. The revenues generated from these resources can be substantial. The min-

eral trade in eastern DR Congo is valued in the hundreds of millions of dollars per


year (UNSC 2008: 31). Rebel groups operating in resource-rich areas have a strong

incentive to tax and control this trade. Sierra Leone’s RUF rebels mobilized armies

of diggers, often by force, to work in the diamond fields, generating between $25-125

million per year (UNSC 2000b: 17).

Although lucrative, lootable natural resources create acute challenges for rebel

groups because nearly any civilian or rebel soldier can become involved in the resource

trade. If spread over a large area, the resources become nearly impossible for rebel

leaders to monitor and control.4 Smuggling routes may form to avoid rebel taxation,

or rebel members themselves may profit from the trade without sending money up

the chain of command. Rebel members also become reluctant to follow orders in

resource-rich areas. Tempted by the possibility of quick profits, commanders and

soldiers might prefer to spend time in the mines rather than on a risky military

offensive. The goals, cohesion, and motivation of the group become diluted by the

pursuit of personal wealth.

The difficulty of monitoring lootable resources strains relationships and trust

within rebel groups. Rebel members might grow suspicious that others in the group

are earning greater profits, or that their personal wealth encourages others to conspire

against them. The RUF, for example, was plagued by accusations of embezzlement

among its top members. Tensions finally reached a breaking point in 1999 when

RUF leader Foday Sankoh accused his top commander, Sam Bockarie, of stealing

diamonds. These allegations sparked an armed confrontation that killed several RUF

members and caused Bockarie to flee to Liberia, where he remained in exile for the

remainder of the war (UNSC 2000b: 17).

Civilian assets and domestic trade can also be considered lootable resources. Rebel

commanders and soldiers can use their weapons to loot from civilians and traders,

or collect taxes from them. Although rarely as valuable as natural resources, rebel

groups can generate substantial sums from the domestic economy. In eastern DR

Congo, for example, the CNDP rebels collected 10kg of beans, sorghum, or corn

from each household to feed their soldiers. The group also collected annual taxes

4In Philippe Le Billon’s (2001) formulation, lootable resources are “diffuse”, rather than “point”,resources.

2.3. LOOTABLE RESOURCES 39

of $30-50 from small businesses, generating hundreds of thousands of dollars in tax

revenue. The group also levied a $5 tax on every 30kg bag of charcoal, worth $18

in the market. With some 50,000 tons of charcoal produced each year in the area,

the CNDP generated several million dollars during 2007-8 from the charcoal trade

(UNSC 2008: 8).

Like lootable natural resources, the revenues generated from civilians are difficult

for the rebel leader to monitor and control.5 The leader nearly always has an incen-

tive to promote security in the group’s territory, but group members can often profit

more by looting. The incentive for looting increases when the leader tries to take a

greater share of the tax revenue. Even if the leader is using the taxes to purchase

ammunition and other supplies, rebel members will see little benefit from taxation

and will be more inclined to keep the revenues for themselves. Taxation also requires

a bureaucracy to collect and keep track of civilian payments. Creating such adminis-

tration requires effort from commanders and rebel members who could collect ad hoc

payments instead.

The strongest incentives for taxation exist when the leader allows commanders

to keep all tax revenues and use the funds as they see fit. Commanders in the

CNDP rebels, for example, maintained possession of all food and taxes collected in

their areas of control, and received additional resources from the rebel leader when

needed (N07, N08). Such a system was possible because the CNDP leader, Laurent

Nkunda, benefitted from his partnerships with large-scale businessmen and other

patrons and did not depend on tax revenue to purchase munitions or pay his own

expenses (UNSC 2008: 7-8). Rebel leaders create stronger incentives for taxation and

security when they can provide credible promises of future rewards and have access

to financial resources other than civilian taxation. Indeed, the most sophisticated

systems of taxation were created by rebel groups such as Sri Lanka’s LTTE that

earned substantial revenue from other sources. Rebel groups create more effective,

5Taxation poses fewer challenges for small rebel groups. In localized rebellions, the leader hasdirect oversight over civilians and the assets being taxed and can judge whether commanders arecollecting revenue responsibly and handing over an appropriate share. The situation is much differentfor larger rebel groups that control a large territory, which limits the leader’s ability to estimate taxrevenues and oversee tax collection.


and less predatory, systems of taxation when commanders are motivated to create

administrative systems through the incentives offered by the rebel leader.

2.4 Punishment, Status, and Ideology as Motiva-

tors

Commanders might be driven by motivations other than a desire for spot payments,

future political office, or lootable resources. A commander might be motivated by a

fear of punishment, the desire for social status, or commitment to an ideology. Even if

the leader cannot compensate the commander, one could argue, a commander might

have an incentive to follow orders and behave with discipline because of these other,

non-pecuniary motivations.

Punishment can be a powerful motivator; individuals will go to great lengths to

avoid physical harm. When dealing with rebel commanders, however, punishment has

only limited value. Rebel leaders must convince commanders to join the group and

provide them with weapons and manpower. Unlike soldiers, commanders cannot be

abducted and controlled by force because they operate with a high degree of autonomy

on the battlefield. Leaders might execute or expel troublesome commanders, but

these punishments must be in concert with positive inducements. If commanders

are unsatisfied with their rewards, they can split from the group or stage a coup

against the leadership. Punishment is only effective when the leader can rely on

other, positive incentives to encourage commander cooperation.

Unlike punishment, a desire for social status could act as a positive motivator

for commanders to follow orders and protect civilians. A rebel commander, however,

gains status from society only if the rebel group as a whole has a reputation for

discipline and beneficial behavior towards civilians. Status as a motivation depends

on the commander’s expectation that the leader can control other commanders in the

group. A single bad commander can destroy the social rewards for everyone in the

group. Status, therefore, depends on the rebel group’s ability to induce cooperation

and enforce discipline among its members. Status is more likely to motivate individual

2.4. PUNISHMENT, STATUS, AND IDEOLOGY AS MOTIVATORS 41

behavior when the rebel leader can create incentives for cooperation and discipline

through spot payments and credible promises.

Ideology is, in some ways, similar to status. An ideologically-motivated comman-

der must expect that other commanders will not betray group goals for personal gain

and that the leader will implement policies consistent with the ideology in the future.

These expectations are more likely when the leader can create incentives to control

other commanders and can make credible promises of future rewards and policies.

But ideology is not simply a product of effective control within an organization: ide-

ology can strengthen control. In the context of rebellion, ideology is similar to David

Kreps’ (1990) concept of “corporate culture”. By defining common norms and clear

standards of behavior, ideology can serve as a coordination mechanism that facilitates

monitoring and increases the credibility of commitments within the group. Ideology

also provides guidelines that help coordinate actions in complex situations—such as

battle—where communication is difficult.

For ideology to affect group behavior, rebel members must understand the es-

tablished norms of the group and expect that other group members will act in line

with those norms (Bicchieri 1990). Creating these common norms and expectations

is a challenge for any organization but is best achieved through formal, collective

socialization processes (Van Maanen and Schein 1977; Jones 1986). In rebel groups,

creating these shared expectations requires standardized training for new recruits

and consistent, reaffirming behavior from the group’s commanders. To maintain

an ideology, commanders must constantly police bad behavior among their soldiers

and, more importantly, live up to the expectations of the ideology themselves (Kreps

1990). Whether or not commanders are motivated to undertake these actions de-

pends to a large extent on the incentives offered by the rebel leader and her credibil-

ity. The formal incentive structures of the group—spot payments and credible future

rewards—shape the extent to which these non-pecuniary motivations can affect indi-

vidual behavior.


2.5 Controlling Soldiers

To survive and succeed, rebel groups must convince their soldiers to follow orders

and work together effectively. Successful groups discourage opportunistic behavior,

as this distracts soldiers from their mission, and take steps to improve soldiers’ tac-

tical abilities. Controlling soldiers depends on the actions of commanders, who are

responsible for training, monitoring, and feeding the troops. When properly moti-

vated, commanders take actions that create social incentives for soldier discipline,

punish misbehavior, and implement systems of food distribution that do not depend

on looting. When commanders do not fulfill these tasks, soldiers are left to fend for

themselves and may rely on brutal, sadistic behavior to socialize their comrades and

establish standing in the group.

Fundamentally, a soldier follows an order if the benefits outweigh the costs of

obedience. Benefits include not only monetary incentives but also social rewards and

perhaps the intrinsic satisfaction of fulfilling one’s duties. Punishment can include

fines or extra chores, as well as physical abuse, expulsion from the group, or death.

Few material incentives are required to maintain a disciplined fighting force. Sri

Lanka’s LTTE rebels, for example, displayed remarkable discipline: despite twenty

years of conflict, no reports exist of LTTE soldiers committing rape or looting.6

Yet LTTE soldiers received no salaries from the organization, only food, shelter, and

clothing (S12, S18, S35). Rather than material incentives, rebel groups—and military

organizations in general—rely on social incentives to motivate their fighters. Estab-

lishing these social bonds is central to the philosophies of highly effective military

organizations (e.g. CAL 2004).

Training is fundamental for creating discipline among soldiers. Military organiza-

tions such as the United States Marines can take nearly any young man off the street

and transform him into a soldier, with all the right reflexes and attitudes, within

a matter of weeks (Dyer 2005).7 Effective training involves systematically stripping

6Elisabeth Wood’s (2006) research into sexual violence uncovered no allegations of rape by LTTEsoldiers, even against vulnerable female combatants. Although organizations such as Human RightsWatch document numerous human rights violations by the LTTE, no allegations of looting appearin these reports (e.g. HRW 2008).

7Young men generally make better recruits than women. Young men combine physical strength

2.5. CONTROLLING SOLDIERS 43

new recruits of their previous identity and forcing them to become dependent on their

comrades and their commanding officers. Training, more than personal characteris-

tics, determines how individuals behave as soldiers.8 Many rebel groups attempt to

replicate the basic training programs used by advanced militaries. The groups com-

bine weapons instruction with physical exercise and group activities. The recruits,

separated from their families and friends, form bonds with their new comrades in

much the same way as in traditional militaries. As Che Guevara ([1961] 1998: 108)

explains, “The fundamental training of the soldier of liberation is the life itself with

the guerrilla band.”

Training also encourages discipline by acting as a signal to rebel soldiers of their

value to the organization. The sunk costs involved in training imply that the rebel

group has long-term plans for the soldier and will not carelessly risk the soldier’s

life. A rebel soldier might hope to receive future employment in the civil service, the

military, or the police, but the soldier cannot be sure that the rebel leadership will

actually deliver on its promises. Investing in training makes the rebel group’s promises

more credible to the soldier. And when promises are more credible, the soldier is more

willing to sacrifice short-term gains and work towards long-term group goals.

This argument is essentially the flip-side of Michael Spence’s (1973) classic work

on job-market signaling, in which employees invest in education as a costly signal of

their quality. In Spence’s model, the employers are able to make credible promises but

face the challenge of selecting employees based on unobservable attributes. It is up

to the employees to signal their long-term commitment through education or other

investments. Similar to Spence, Jeremy Weinstein (2005, 2007) views training in

rebel groups as a screening mechanism whereby new recruits demonstrate their level

of commitment to the group. Rebel groups, however, face an additional challenge

with personality traits that are easily manipulated by military organizations, such as aggression, atendency to hang out in groups, and a desperate desire to fit in (Dyer 2005: 52).

8The United States Army provides an interesting example. In 2007, during the height of theIraq war, only 79% of army recruits had a high school diploma, far short of the 90% target. Inaddition, the Army granted 8,259 waivers for recruits who had committed “serious misdemeanors”(up from 3,002 in 2005) and 5,985 waivers for recruits with medical conditions (up from 4,348in 2005). Nevertheless, recruits who required waivers were slightly more likely to earn “valorousawards” during their time of service than recruits who met the standard requirements (NPR 2008).


because the rebel leader cannot write contracts regarding long-term policies and future

rewards. Rebel soldiers must form their own expectations of the leader’s credibility,

and training provides a useful signal of that credibility.

Commanders who cannot provide careful, intensive military training rely on other

tools for indoctrinating new recruits. Beatings and violence can replace the calcu-

lated progress and indoctrination of standard basic training programs. For example,

Blattman and Annan (2010) find that 57% of people abducted by Uganda’s LRA

rebels were severely beaten within 3 months of joining the group, while 68% were

tied up or imprisoned. In the same three month period, 19% of LRA abductees were

forced to abuse a dead body and 11% were forced to kill a family member or friend.

Forcing recruits to commit atrocities severs ties with their former lives and increases

dependence on the group. Such dependence increases group cohesion and the com-

mander’s ability to control soldier behavior (Hechter 1987). With no home to return

to, recruits have little choice but to follow orders and remain in good standing with

the group.

A story about the life of Sun Tzu, the author of The Art of War, illustrates the

ability of commanders to control soldiers through fear (quoted in Giles 1910). After

the publication of his book, Sun Tzu was called into the palace of Ho Lu, the King

of Wu, and given command of 180 ladies of the palace to test the book’s advice.

Sun Tzu divided the women into two companies, each headed by one of the King’s

favorite concubines as a captain. Sun Tzu provided instructions on how to respond to

military commands such as “Left Turn” and “About Face”, but upon giving an order

the women simply giggled. Sun Tzu restated the order, and when the women again

failed to obey, Sun Tzu declared that the two concubine-captains would be beheaded.

The two women were beheaded and new captains chosen from among the remaining

women. Again Sun Tzu gave an order to the troops, and this time the women snapped

to attention and executed the order without delay. Sun Tzu informed the King, “Your

soldiers, Sire, are now properly drilled and disciplined... They can be put to any use

that their sovereign may desire; bid them go through fire and water, and they will

not disobey.” Commanders can achieve a great deal of discipline from their soldiers

if they are motivated to monitor and punish misbehavior.

2.5. CONTROLLING SOLDIERS 45

Food distribution is another fundamental aspect of soldier discipline. Soldiers

need food to survive; military orders and threats of punishment mean little to a

starving soldier. If a rebel group cannot provide food and essentials to their soldiers,

the group must allow soldiers to forage for themselves at the expense of following

orders. If food is supplied by the rebel leadership, commanders must distribute that

food to the troops, rather than diverting the shipments for personal profit. If no

food is supplied, the commander must take it upon himself to organize a system

for collecting large quantities of food from local villages and then distributing those

supplies. Money and resources are not required to feed soldiers, but a high degree of

organization is.

When commanders fail to create the proper environment for their troops, fighting

units become dysfunctional. In stressful situations, people seek support and accep-

tance among group members. If these social support networks do not form through

training, soldiers will take it upon themselves to create social bonds with their com-

rades. Dara Cohen (2010) describes how Sierra Leone’s RUF soldiers used gang rape

as a means for building social cohesion within fighting units. This behavior did not

benefit the rebel leadership: not only was gang rape a distraction from military ac-

tivities, the brutal behavior decreased fighting capacity as soldiers were crippled by

venereal disease. When control is weak within an armed group, sadistic behavior can

be used as method of socializing new recruits, and as a costly means of establishing

status among group members. In these circumstances, civilians pay a terrible cost.

Although rebel groups possess the tools to control their soldiers, these tools are in

the hands of rebel commanders. The challenge for rebel organizations, therefore, is to

create incentives that motivate commanders to train and monitor their soldiers, while

also convincing commanders to faithfully follow the orders of the rebel leadership.

The leader’s ability to motivate commanders depends on her level of financing, the

credibility of her promises, and the commander’s access to lootable resources. The

next section develops a formal model of the leader-commander relationship.


2.6 Basic Model

To begin, I consider the interaction between a rebel leader, L, and a high-level com-

mander, C, who are fighting to capture a state or homeland worth Π ∈ R+. The

rebel leader has certain policy goals but depends on the commander, who has access

to manpower and weapons, to implement those policies.

In the first stage of the game, the rebel leader sets a policy x ∈ [0, 1] and specifies a

monetary transfer t ∈ [0, T ]. The policy x specifies the extent to which the commander

must work towards group goals, rather than personal objectives. For example, a high

value of x might require the commander to set up training camps, organize food

distribution for the troops, and police civilian areas to prevent looting and lawlessness.

Low values of x imply that the commander has discretion over his manpower and

resources, and can deploy those resources for personal gain. The commander might

abandon the effort-intensive tasks of administering territory in favor of extracting

natural resources, profiting from lucrative trade routes, or looting civilians.

To have any chance of success, the rebel leader must be able to induce a minimal

level of cooperation from the commander, represented by xmin. Minimal cooperation

occurs when the commander cooperates with the leader’s overall military strategy

but does not take orders regarding the day-to-day tasks of managing the forces or

administering a territory. The key assumption of the model is that a low level of

control is associated with civilian abuse and a greater risk of factionalization.

Presented with (x, t), the commander either accepts or rejects the leader’s offer.

If the commander rejects, he uses his manpower and weapons to seize the group’s

territory but faces a risk of punishment by the leader. This defection results in

expected utility (1− e)V where V ∈ [0,Π] represents the value of lootable resources

and e ∈ [0, 1] represents the risk of punishment. Higher values of e correspond to

harsher, more certain punishment.

If the commander accepts the leader’s offer, the game enters a second stage. The

rebel group achieves its objectives according to a success function ρ(x) ∈ [0, 1]. This

function reflects the commander’s military skill as well as other factors such as the

2.6. BASIC MODEL 47

leader’s political expertise, the strength of the state, and geographic factors. What-

ever resources are not devoted to group ends can be used by the commander to

generate revenues for himself, generating (1− x)V in payoffs.

Because the rebel leader must achieve at least x = xmin to have any chance of

success, ρ(x) = 0 for all x ≤ xmin. For values x > xmin, the success function ρ(x)

is a smooth, strictly concave function, such that ρ(x) > 0, ρ′(x) > 0 and ρ′′(x) < 0

for all x ∈ (xmin, 1]. A discontinuity therefore exists for ρ′(x) at x = xmin where

ρ′(xmin) = 0 but ρ′(xmin + ε) might be significantly larger. Specifically, assume that

ρ′(xmin + ε) � VΠ

. To simplify the analysis in the basic model, let xmin = 0, so that

ρ(x) is continuous and strictly concave for all values of x ∈ (0, 1]. This assumption

will be relaxed in the next chapter, when the role of external patrons is considered.

If the rebels succeed, the leader proposes to keep a division of the spoils, α ∈ [0, 1],

and offers the remainder to the commander. The commander can accept the division

and receive a payout of (1− α)Π + (1− x)V + t. In this case, the leader receives her

share of the spoils, αΠ, plus any money not transferred to the commander in the first

stage, for a total payout of αΠ + T − t.If the commander rejects the offer, the commander and leader fight for control of

the government. The leader defeats the commander with probability p ∈ [0, 1], and

each player bears a cost of fighting, k > 0.9

The game proceeds as follows:

1. Leader offers a policy and transfer (x, t) to the commander.

2. Commander chooses dc1 ∈ {Reject,Accept}.

3. The group achieves its objectives with probability ρ(x).

4. Leader decides to keep a share of the spoils α ∈ [0, 1] and offers the rest to the

commander.

5. Commander chooses dc2 ∈ {Reject,Accept}.9In certain situations there may be an incentive for the group to split apart to increase its

bargaining leverage during peace talks. More factions, for example, could provide more seats at thebargaining table and a better total payout. In this case, k would be negative. This situation is notconsidered here.


Figure 2.1 depicts the game in extensive form.

2.6.1 Solving the Basic Model

Subgame perfect equilibrium is the relevant solution concept and the model can be

solved through backwards induction. The unique equilibrium of the model takes the

following form (proofs in the appendix):

Proposition 1. There exists some unique x∗ ≥ 0 that solves ρ′(x∗) = VΠ

. Let t∗ =

x∗V − eV − ρ(x∗)(1− α∗)Π and a∗ = p+ kΠ

.

• If T ≥ t∗ and ρ′(1) ≤ VΠ

, the leader offers the commander (x∗, t∗, a∗).

• If T ≥ t∗ and ρ′(1) > VΠ

, the leader offers the commander (1, t∗, a∗).

• If T < t∗, the leader offers (xl, T, a∗), such that xl = e+ T

V+ ρ(xl)(1− α∗) Π

V.

The commander accepts the leader’s equilibrium offer of (x, t) as well as the post-

war settlement, α. The level of control in the group, xl, increases with the amount of

financing available to the leader, the value of the commander’s future rewards, and the

threat of punishment, but only when the leader is cash-constrained such that T < t∗.

Lootable resources decrease control in the group, measured by x∗ or xl, regardless of

the leader’s access to financing.

When rebel leaders and commanders have perfect information, a bargain always

exists that is preferable to costly fighting. In this model, the rebel leader has the

power to make the offers and thus the leader captures all of the surplus produced

by the group. In equilibrium, the leader always makes the commander indifferent

between following orders and breaking from the group.

The leader’s access to funds shapes her strategic decision making. When the

leader has limited access to cash, such that T < t∗, the leader pays the commander

as much as possible—transferring the entire endowment, T , to the commander—

and then chooses x to make the commander indifferent between following orders and

breaking off from the group. To determine x, the leader must consider her ability

2.6. BASIC MODEL 49

LC

NL

C(x,t)

Accep

t

Reject

T (1-e)V

Win

[!(x)]

Lose

[1-!(x)]

T-t

(1-x)V+t

"Accep

t

Reject

p#-k+T-t

(1-p)#

-k+(1-x)V+t

"#+T-t

(1-")#

+(1-x)V+t

))

!●

●●

●

Fig

ure

2.1:

Bas

icm

odel

ofre

bel

orga

niz

atio

n


to punish the commander, as well as the value of the commander’s future rewards

and his access to lootable resources. Harsher punishments and more valuable future

rewards allow the leader to exert more control over the commander, even when funds

are tight. But valuable lootable resources increase the commander’s opportunity cost

for following orders. To keep the commander in the group, the leader must loosen

her grip and give the commander enough autonomy to profit from the resources in

his area.

The leader is in a stronger position when she has sufficient financing, such that

T ≥ t∗. In this situation, the leader uses cash to achieve the optimal level of con-

trol. The leader can use cash payments to compensate for a lack of credible promises.

Likewise, the leader can increase the rewards for following orders if she cannot rely on

harsh punishments to maintain discipline. Even in this scenario, however, lootable

resources undermine control in the group. Lootable resources determine the com-

mander’s marginal cost of compliance, whether T > t∗ or not. The leader’s optimal

level of control, x∗, balances the relative value of lootable resources with the value of

achieving the group’s objectives.

The model provides interesting insight into the competing effects of state wealth

and state capacity for control within rebel groups. Ceteris paribus, more valuable

states, indicated by greater values of Π, correspond to more disciplined rebel groups

because of the greater value of future rewards for the commander (proof in the ap-

pendix). This effect, however, depends on the assumption that the value of the state

is not correlated with state strength, captured in ρ(x).10 In reality, more valuable

states are also likely to have greater capacity to defend against an insurgency. Indeed,

per capita GDP is often used as a proxy for state capacity (e.g. Fearon and Laitin

2003). Stronger states decrease the rebels’ chance of success, which then decreases

the commander’s expected value of future rewards and increases his temptation to

profit during the conflict. As a result, the model makes no clear predictions regarding

10The relationship between Π and control in the rebel control also assumes that, as a state becomeswealthier, there is not a corresponding increase in the value of looting opportunities, V . Thisassumption can be justified on the grounds that wealth becomes harder to expropriate by force athigher levels of development, since capital becomes increasingly mobile. This relationship is exploredin a model proposed by James Fearon (2008).

2.6. BASIC MODEL 51

the relationship between state wealth, state capacity, and rebel discipline.

The main predictions of the basic model can be summarized in the following

hypotheses:

Hypothesis 1. The level of control in a rebel organization (weakly) increases

with the leader’s access to financing.

Hypothesis 2. The level of control in a rebel organization decreases with the

commander’s access to lootable resources.

Hypothesis 3. The level of control in a rebel organization (weakly) increases

with the value of the commander’s expected future rewards.

2.6.2 Commitment Problems and Cash Constraints

In this basic model, the commander expects to receive post-conflict rewards that

make him indifferent between laying down arms and embarking on a costly fight

for dominance. The rebel leader might promise greater rewards, but the commander

knows that these promises are not credible. Unable to rely on an external legal system,

the leader cannot write a binding contract at the start of the war that guarantees the

commander a greater share of the post-conflict wealth in exchange for higher levels of

cooperation. The lack of externally-enforced, binding contracts leads to the following

proposition:

Proposition 2. In cash-constrained groups such that T < t∗, the rebel leader’s utility

and the total welfare of the rebel group would increase from a binding contract that

limits the leader’s share of future rewards by specifying some α < α∗ at the start of

the conflict. When the leader has sufficient cash such that T ≥ t∗, a binding contract

has no impact on the leader’s utility or the group’s total welfare.

During civil war, a commander who receives little support from the rebel leader can

still be motivated by the thought of one day becoming Minister of Defense or receiving

some other valuable reward for his service. Knowing this, leaders are tempted to

promise extravagant rewards in the hope of convincing commanders to discipline their

troops and fight with determination, instead of looting locals or mining diamonds.

Such promises are not credible in this model, which assumes the commander is fully


aware of α∗, the leader’s equilibrium share of the spoils. The leader’s promises are

cheap talk and have no effect on the commander’s behavior.

Leaders with sufficient financing are not affected by the lack of externally-enforced

contracts. If the commander expects only a small reward in the future, the leader

compensates with a large cash transfer in the present. With enough cash, the leader

can achieve the optimal level of cooperation, regardless of the commander’s future

expectations. Leaders with limited cash, by contrast, compensate for less-valuable

future rewards by allowing commanders to use group resources for personal gain.

These group resources would produce greater surpluses if deployed towards capturing

the state; using these resources for personal, short-term gains creates inefficiencies in

the group. Cash-constrained leaders face commitment problems that decrease their

utility as well as the group’s total surplus.

2.7 Extending the Model

The model can be extended to consider leader-commander affinity, imperfect moni-

toring, private information, and multiple commanders.

2.7.1 Affinity

Loyalty and affinities matter during civil war. In the absence of externally enforced

agreements, alternative means of establishing trust and credibility—such as common

kinship, ethnicity, or religion—become more important (e.g. Bates 1983; Habyari-

mana et al 2009). In some groups, the rebel leader and commander share a relation-

ship that allows the leader to credibly promise to reward the commander above and

beyond the amount the commander would otherwise expect. For example, consider

a rebel group in which the leader and top commander are siblings. If the leader

appoints the commander—her brother—as Minister of Defense, the leader gains a

useful ally in government and also delivers a reward to her family. Knowing this, the

commander has less reason to question the leader’s promises. Affinity and social ties

between the leader and commander transform the division of post-conflict rewards

2.7. EXTENDING THE MODEL 53

from a zero-sum game into a more cooperative agreement.

Let φ(1−α) represent the affinity between the rebel leader and commander, where

φ(1−α) ∈ [0, 1] and φ(·) is an increasing, concave function. Higher values of φ(·) imply

that the leader has a greater incentive to increase the commander’s share of group

resources. When affinity is added to the model, the leader’s final payout becomes:

UL = [α + φ(1− α)]Π + T − t

The rest of the payouts remain unchanged.

The optimal choice of α is αopt such that:

φ′(1− αopt) = 1

From the previous section, we know that in the absence of affinity the leader will

propose to keep a share α∗ = p + kΠ

. Now, the leader will choose a share that solves

min{α∗, αopt}.If αopt < α∗, the leader makes the commander an offer that exceeds the comman-

der’s outside option, rather than simply matching it. In such cases, the commander

has a greater incentive to follow the leader’s orders during the war and work towards

group goals. Figure 2.2 depicts a high-affinity rebel group where the leader can credi-

bly commit to offering (1−αopt)Π, which exceeds the commander’s reservation utility.

By increasing the expectation of future rewards, the leader strengthens control over

the commander. The effect of affinity leads to the following hypothesis:

Hypothesis 4. The level of control in a rebel organization increases when the

leader and commander share personal bonds that establish trust.

2.7.2 Imperfect Monitoring

Although many rebel groups have access to radio equipment and other communica-

tions technology, the fog of war can prevent the leader from effectively monitoring

commander behavior. The leader might not be able to accurately observe whether the

commander faithfully attempted to follow orders or instead used group resources for


●

!’(1-"opt)=1

●

1-"*=1-(p+k/#)

!(1-")

1-"

Figure 2.2: Commander’s share of group resources in a high-affinity rebel group.


opportunistic ends. If the leader orders the commander to capture a town, for exam-

ple, the failure to take the town might have been caused by unexpected government

actions, or by the commander’s decision to loot rather than engage the enemy.

Suppose that, if the commander rejects the leader’s offer (x, t) and instead chooses

to loot, the leader can only observe the commander’s defection with probability σ ∈[0, 1].11 Thus with probability 1−σ the commander can choose x = 0 and still receive

the transfer t from the rebel leader, while avoiding punishment.

The standard response to imperfect information in a principal-agent relationship

is for the leader to condition the commander’s payout on some observable measure

related to the commander’s effort. The most natural measure in this context would be

whether the rebels win or not. But without enforceable contracts, the leader cannot

credibly commit to giving the commander additional payments once the war is over.

In rebel groups, imperfect monitoring decreases the effectiveness of cash transfers as

an incentive for commander cooperation. The introduction of imperfect monitoring

leads to the following proposition (proof in the appendix):

Proposition 3. If future rewards are not too valuable, such that (1− α)Π < x(t)V ,

then ∂2x∂σ∂t

≥ 0 and ∂2x∂σ∂V

≤ 0. If future rewards are sufficiently valuable, such that

(1− α)Π ≥ x(t)V , then ∂2x∂σ∂t

= 0 and ∂2x∂σ∂V

= 0.

The first partial derivative in the proposition, ∂2x/(∂σ∂t) ≥ 0, states that the

leader’s ability to induce cooperation through cash transfers depends on the leader’s

monitoring abilities. Cash transfers are more effective at inducing cooperation in

easy-to-monitor areas, such as the rebel headquarters, and less effective at induc-

ing cooperation in remote areas, where the commander knows that the probability

of detection is small. The second partial derivative, ∂2x/(∂σ∂V ) ≤ 0 states that

monitoring does less to strengthen leader control as the value of lootable resources

increases. In the resource areas, leaders accept lower levels of cooperation because

commanders face a higher opportunity cost for following orders. Monitoring makes

11Imperfect monitoring could take another form. Suppose that if the commander implements xthere is some probability ζ ∈ [0, 1] that the commander’s efforts fail to produce any group resources,causing the rebel leader to observe ρ(0) = 0. Punishment is assumed to be costless for the rebelleader so the leader would always punish the commander upon observing ρ(0) = 0.


little difference in these areas since leaders expect little control to begin with.

When commanders are motivated by future rewards, the leader does not have to

devote as much energy to monitoring. Under these circumstances, the commander

is motivated to cooperate to produce group resources that can divided at the end of

the war. As the final two partial derivatives of the proposition show, monitoring be-

comes unnecessary in the organization if future rewards are sufficiently valuable. The

commander will follow the leader’s orders even if the leader has no way of observing

defection. The value of lootable resources, however, plays a large role in determining

how valuable the future rewards must be.

Although simple to model, imperfect information has important implications for

the behavior of rebel commanders, summarized in the following hypotheses.

Hypothesis 5. The level of control in a rebel organization is greater in areas

that are easily monitored by the rebel leader.

Hypothesis 6. Monitoring has a greater effect on leader control among groups

in which the leader uses cash transfers to encourage cooperation,

and less effect among groups that lack leader financing or rely

instead on future rewards.

Taken together, Hypotheses 5 and 6 can be extended to analyze the spatial pat-

terns of rebel group predation. When leaders have access to cash for spot payments

but do not have sufficiently strong personal ties with the commander, group discipline

depends on effective monitoring by the rebel leader. Rebel members will be more dis-

ciplined in easy-to-monitor areas, such as the rebel headquarters, and less-disciplined

in remote or unstable areas.

The amount of civilian abuse in a given area depends on the leader’s ability to

monitor that area, as well as the number of soldiers stationed there. Soldiers near

the rebel headquarters may be less likely to prey on civilians due to monitoring and

the risk of punishment, but a large number of soldiers in the bases means that some

predation may nevertheless occur. By contrast, if few soldiers are posted in remote

areas, their predatory behavior may go unnoticed amidst the instability and violence

of the conflict.


For groups in which the leader relies on spot payments, the level of abusiveness

depends on two competing pressures: the leader’s monitoring abilities and the density

of group members. If monitoring is the dominant factor, then the level of abusiveness

is predicted to be lower near the rebel bases and higher in remote and unstable areas.

If the density of rebel members dominates, no clear predictions are obtained.

A different dynamic occurs when leaders cannot provide valuable spot payments

to their commanders. To compensate for the lack of resources, the leader allows

rebel members to loot locals in exchange for a minimal level of compliance. In this

situation, monitoring has little effect on behavior and the rebels prey on civilians

throughout the group’s territory. The areas hardest hit by predation will be near the

rebel bases, due to the high concentration of rebel soldiers, and other easily accessible

locations. The effects of density dominate the weak effects of monitoring, which leads

to the following hypothesis.

Hypothesis 7. When the leader does not share strong personal bonds with the

commander and has limited access to financial resources, the

level of predation committed by the group is highest near rebel

bases and in other easily accessible areas.

2.7.3 Private Information

The basic model assumes the leader and commander have perfect information and

thus a bargain always exists that is preferable to costly fighting. In reality, leaders

and commanders possess private information and have incentives to misrepresent

that information. For example, a commander may have an incentive to exagerate

the number of troops under his command to give the leader a false impression of his

strength, or the leader might form political alliances that increase her leverage over

the commander.

The model can be extended to consider private information between the leader and

commander. In the second stage, during peace talks, the leader offers the commander

a share of the spoils, α∗, based on the probability of defeating the commander in a

fight, p, and the costs of fighting, k. A bargain always exists if both sides agree on


the values of p and k, but bargaining failure might occur if the leader and commander

estimate these parameters differently (Fearon 1995).12

For a peace agreement to hold, the leader and commander must make subtle cal-

culations regarding their relative strengths and the costs of fighting. But the nature

of peace talks shifts the values of p and k based on changes in the battlefield and

the political environment. Neither the leader nor commander has full information

regarding these changes, as they are each involved in different aspects of the peace

process. The longer the peace talks continue, the more private information accu-

mulates and the greater the risk that the leader and commander will fail to reach a

mutually acceptable agreement. In the model, the leader would propose a division of

the spoils, α, based on her beliefs regarding p and k. If the leader’s proposed share

exceeds the commander’s estimate of α∗, the commander would reject the offer and

break from the group. Introducing private information into the model leads to the

following hypothesis:

Hypothesis 8. Peace talks increase the risk of group factionalization.

The civil war in Djibouti provides an example of how prolonged peace talks can

strain relations within rebel groups. Inter-communal violence escalated into civil war

in the early 1990s with the emergence of the FRUD rebels. Peace talks began in 1992

between the government and FRUD leader, Ahmed Dini, amidst growing government

repression and violence against civilians. But two years of sporadic negotiations failed

to produce results. Frustrated by the lack of progress and increasingly skeptical of the

leader’s excuses, FRUD’s top commander, Ougoureh Kifle Ahmed, renounced his alle-

giance to Dini and appointed a new leader, Ali Mohamed Daoud (MAR 2004). A new

round of talks led to a peace agreement in 1994, and commander Ougoureh received

a ministerial position in a coalition government. The scorned former leader, Ahmed

Dini, rejected the agreement and vowed to continue the struggle. Dini, however, un-

derestimated the commander’s popularity among the troops and proved capable of

12To model this explicitly, let p and k be represented by random variables. The leader has beliefsregarding the distribution of the random variables, but cannot observe the actual values when sheoffers of α. In this set-up, bargaining failure occurs in equilibrium with a probability determinedby the variance of the random variables and whether the leader’s beliefs are biased regarding theexpected values of the random variables.


only small-scale hit-and-run attacks. The violence in Djibouti subsided.

2.7.4 Multiple Commanders

The basic model assumes a rebel leader interacts with a single commander. Since

rebel groups often have several high-level commanders, control in the group might

be affected by interactions among the commanders. For example, commanders might

collude against the rebel leader, or vie for her approval. Commanders might loot

each other’s territory, or compete to occupy the most valuable areas. From the

rebel leader’s perspective, deploying commanders and distributing rewards become

strategic decisions. Formalizing these group effects is an interesting avenue of future

research. For now, I offer just a few observations on these dynamics.

Credible Promises

The attributes of individual commanders in the group affect the credibility of the

leader’s promises. Consider the following scenario. Suppose that the n commanders

of a rebel group expect to earn m positions in the post-conflict government and

that the leader decides how these valuable positions will be allocated among the

commanders. These positions are extremely valuable and every commander would

prefer a position in government to costly fighting. Each commander can only hold

a single position, but if m < n, some commanders will not receive a position in the

post-conflict government.

The commanders excluded from the post-conflict government must decide whether

to accept the leader’s offer of 0 rewards or engage in costly fighting. If commander

i can inflict cost ki on the rebel leader, the leader will allocate the positions in the

government to the m commanders who would impose the greatest costs by defection.

In this way, the remaining n−m commanders inflict the minimum damage.

If commanders are able to observe each others’ strengths, commanders know who

will receive rewards and who will not. The leader’s promises are entirely credible

to some commanders, and incredible to others. Understanding this, the rebel leader


might allocate spot payments strategically to increase her control over more trou-

blesome commanders. Suppose the leader allocates ti to each commander i subject

to the budget constraint∑ti ≤ T . If the leader allocates ti ≥ t∗ to commander i,

the leader can induce the optimal level of control, regardless of the credibility of her

promises. A utility-maximizing rebel leader attempts to choose ti so that ρ′(xi) is

the same for all commanders. As in the basic model, leaders with greater access to

financial resources can exert more effective control over the group.

Linking Control to Civilian Abuse

The theory assumes that a low level of control in a rebel organization will be associated

with higher levels of civilian abuse and predation. But even if a commander refuses to

cooperate with the rebel leader, he may have a personal incentive to provide order and

security in his territory. Rather than predatory chaos, a lack of control might instead

lead to a number of autonomous “stationary bandits” who each has an incentive to

provide order at a local level, even if their comrades are behaving opportunistically.

But as with status and ideology, the commander’s local incentives to create order

depend on the actions and incentives of other commanders in the rebel organiza-

tion. Rival commanders might loot the commander’s territory, removing valuable

goods and destroying the tax base. Even if the commander’s territory remains se-

cure, looting that occurs miles away can disrupt trade and local production, reducing

the commander’s revenue and the benefits of security. A lack of control within a

rebel organization undermines commanders’ individual incentives to create order and

increases the relative payoffs of looting for all members of the organization.

2.8 Discussion

Rebel groups use a variety of incentives to motivate and control their members. Be-

cause they operate outside of the state, rebel organizations cannot rely on contracts

or other means of enforcing agreements. Instead of regular salaries, rebel groups use

on-the-spot cash payments, the delivery of which is always somewhat uncertain. The

value of future rewards similarly depends on personal relationships and the balance

2.8. DISCUSSION 61

of power within the organization. But operating outside of the state also increases

the incentives available to the organization. Rebel leaders and commanders can al-

low their subordinates to loot from civilians in exchange for compliance. And rebel

members can be brutally punished or even summarily executed for failing to follow

orders.

The effectiveness of these incentives for motivating an individual depends on his

or her place in the rebel hierarchy. Soldiers spend their days surrounded by their

comrades. These tight social bonds, a credible threat of punishment, and access

to sufficient food are enough incentives to create disciplined soldiers. Commanders,

on the other hand, operate with a high degree of autonomy. They have command

over weapons and soldiers and are rarely surrounded by their peers. Punishment,

although possible, is more difficult with regards to commanders and is not enough to

motivate commanders to lead offensives or administer territory. Positive incentives

bind commanders to the leader and motivate them to fulfill their duties. Without

strong incentives, commanders are more inclined to use group resources for personal

ends, or break from the group and strike out on their own. The effectiveness of these

incentives depends on the leader’s monitoring abilities and other group characteristics.

A key assumption of the theory is that the leader’s ability to incentivize comman-

ders has downstream effects on the behavior of rebel soldiers. Controlling soldiers

requires few resources but tremendous effort on the part of commanders. When com-

manders are properly motivated, they provide military training, discipline, and food

to their soldiers. If commanders are not motivated to do these tasks, they allow

soldiers to forage for food and create their own, often brutal, methods for inducting

new members and establishing status in the group. The result, in most cases, is

abusiveness towards civilians.

In addition to the challenges they face within their organizations, rebel lead-

ers need to secure access to weapons and financing. The next chapter looks at

the motives of external patrons that supply these resources and how these motives

can strengthen—or undermine—the leader’s ability to incentivize commanders in the

group.


2.A Mathematical Appendix

2.A.1 Proof of Proposition 1

Proof. Beginning with the division of the post-conflict government in Stage 2, the

commander’s optimal strategy is to accept any division of the pie that is at least as

good as the expected value of fighting. The leader has perfect information regarding

the commander’s decision and, because fighting is costly, there is always a bargain

preferable to fighting for both the leader and commander. The leader’s optimal

strategy maximizes α subject to the commander’s incentive compatibility constraint:

α∗ = p+k

Π(2.1)

Stepping back to Stage 1, the commander’s optimal strategy is to accept the

leader’s policy (x, t) if it is at least as good as the opportunity for looting. Since a

rational rebel leader would not pay more than necessary, an optimal contract (x∗, t∗)

satisfies the following:

t∗ = x∗V − eV − ρ(x∗)(1− α∗)Π (2.2)

Because x∗ and t∗ are mutually dependent, a range of contracts might exist. The

leader chooses the contract that maximizes her utility by solving

maxx,t

T − t+ ρ(x)(α∗Π)

subject to Equation (2.2). If the leader has sufficient resources such that T ≥ t∗, the

leader can satisfy Equation (2.2) by setting t = t∗. The leader now solves:

maxx

ρ(x)Π− V x+ T + eV

In equilibrium, a leader with sufficient resources chooses x∗ such that:

ρ′(x∗) =V

Π(2.3)

2.A. MATHEMATICAL APPENDIX 63

If ρ′(1) > VΠ

, the leader chooses x∗ = 1. Recalling that xmin = 0, a unique solution

exists because ρ(x) is smooth and strictly concave over the interval x ∈ (0, 1]; ρ′(xmin+

ε) > VΠ

; and ρ′(x) is decreasing.

The alternative case involves a leader who faces a liquidity constraint such that

t∗ > T . In this situation the leader sets t = T and chooses xl to satisfy the comman-

der’s participation constraint:

(1− xl)V + T + ρ(xl)(1− α∗)Π = (1− e)V

with solution,

xl = e+T

V+ ρ(xl)(1− α∗)

Π

V(2.4)

The prove that an unique, interior solution exists when T < t∗, notice that Equa-

tion (2.4) is the intersection of two lines, f1(x) = x and f2(x) = ρ(x)(1−α∗) ΠV

+ TV

+e.

Since f2(x) > 0 for all x ∈ [0, 1] the leader can always convince the commander to

accept some xl > 0. Observe that f2(x) is increasing. To prove that the lines intersect

over the domain x ∈ (0, 1), it is sufficient to show that f2(1) < 1.

Consider the contrary, such that the two lines do not intersect and f2(1) ≥ 1. In

this case, the leader can set xl = 1 and induce full compliance from the commander.

This situation of full control occurs when ρ(1)(1−α∗)Π+T ≥ (1−e)V . The minimum

value of T required to induce full compliance is:

T = (1− e)V − ρ(1)(1− α∗)Π

This equation is identical to the expression for t∗ when x = 1, given by Equation

(2.2). This violates the assumption t∗ > T for a liquidity-constrained group. This

implies f2(1) < 0 and thus f2(x) ∈ (0, 1). The lines always intersect in the domain

x ∈ (0, 1).

To prove uniqueness when T < t∗ observe the following: f1(x) and f2(x) are both

increasing; f ′1(x) is constant; and f ′′2 (x) < 0 for all x ∈ (0, 1). The last observation

follows from the assumption of strict concavity for ρ(x). Once f2(x) crosses f1(x), the

slope of f2(x) will never be sufficient to cross the line again. As a result, a unique,


interior solution for xl exists on the domain x ∈ (0, 1).

The remainder of the proof concerns comparative statics. Proving that greater

access to financing for the leader increases control in the group when T < t∗ but not

when T ≥ t∗ requires two steps. First, because t∗ > T in the liquidity constrained

group, to prove xl < x∗ it is sufficient to show that ∂xl

∂T> 0 for all T ∈ [0, t∗).

Implicitly differentiating Equation (2.4) with respect to T yields:

∂xl∂T

=1

V − ρ′(xl)(1− α∗)Π

This expression implies that ∂xl

∂T> 0 so long as ρ′(xl) <

11−α∗

VΠ

.

The second step puts an upper limit on ρ′(xl). The optimal value of xl given by

Equation (2.4) is the intersection of two lines, a linear function f1(x) = x, and a

concave function f2(x) = ρ(x)(1 − α∗) ΠV

+ TV

+ e. These two lines intersect exactly

once on the domain x ∈ (0, 1) because the equilibrium is unique. At the point of

intersection the slope of f1(x) must be greater than the slope of f2(x). This implies:

ρ′(xl) <1

1− α∗V

Π(2.5)

Consequently ∂xl

∂T> 0 for all T ∈ [0, t∗) and thus xl < x∗.

Deriving the other comparative statics is more straightforward. The commander’s

share of future rewards does not appear in the equilibrium solution when T ≥ t∗ and

thus does not affect control in the group under this condition. To derive the effects

when T < t∗, implicitly differentiate Equation (2.4) with respect to (1− α∗):

∂xl∂(1− α∗)

=ρ(xl)Π

V − ρ′(xl)(1− α∗)Π

This expression is greater than 0 if ρ′(xl) <1

1−α∗VΠ

, which is true by Equation 2.5.

Thus control in the group, xl, increases with the commander’s share of future rewards,

(1− α∗).

Applying these steps with other variables reveals that the level of control is not


affected by the expectation of punishment, e, when T ≥ t∗, but increases with punish-

ment when T < t∗. Regardless of the level of financing, control in the group decreases

with the value of lootable resources, V , and increases with the value of government,

Π.

To derive the effect of group skill on control in the group, recall that the solution

for xl in Equation (2.4) is given by the intersection of two lines and always exists on

the domain x ∈ (0, 1). For any value of x on this domain, an increase in ρ(x) or ρ′(x)

shifts f2(x) upwards. Since both f1(x) and f2(x) are increasing and ρ(x) is strictly

concave, the intersection of the lines shifts to the right. Thus xl increases with an

increase in ρ(x).


Proof. In the cash-constrained group, the leader’s utility is UL = ρ(xl)αΠ. Implicitly

differentiating with respect to α:

∂UL∂α

= ρ(xl)Π + ρ′(xl)αΠ∂xl∂α

Implicitly differentiating the expression for xl in Equation (2.4) with respect to α

yields:∂xl∂α

=−ρ(xl)Π

V − ρ′(xl)(1− α)Π

Combining these expressions:

∂UL∂α

= ρ(xl)Π

(1− ρ′(xl)αΠ

V − ρ′(xl)(1− α)Π

)The leader’s utility increases by being able to commit to α < α∗ if ∂UL

∂α< 0. This

implies:

ρ′(xl) >V

Π

This condition holds for all xl < x∗, which is true by Proposition 1. The rebel

leader benefits from the ability to enter into binding agreements when T < t∗. The


commander, by contrast, receives the same payout regardless of the leader’s ability

to form binding contracts. Since the leader’s utility increases while the commander’s

utility remains constant, total welfare in the group increases if binding contracts are

possible.

When the leader has sufficient access to cash such that T ≥ t∗, the leader’s utility

is UL = ρ(x∗)αΠ. Implicitly differentiating with respect to α:

∂UL∂α

= ρ(x∗)Π

The right-hand side of this expression is always positive, which means the leader has

an incentive to maximize her share of the government, given by α, while still offering

a bargain that the commander is willing to accept. The maximum α that the leader

can achieve is α∗. When T ≥ t∗, the outcome does not depend on the ability to form

binding agreements.


With imperfect monitoring, the leader can induce cooperation from the commander

if:

σ(1− e)V + (1− σ)(V + t) ≤ (1− α∗)Π + (1− x(t;σ))V + t

Rearranging:

x(t;σ) ≤ (eV + t)σ + (1 + α)Π

V

The marginal effect of monitoring on control is found by differentiating:

∂x(t;σ)

∂σ=eV + t

V

The magnitude of the first derivative increases with t and decreases with V , thus∂2x∂σ∂V

≤ 0 and ∂2x∂σ∂t≥ 0. The level of control in the organization is more sensitive to

changes in monitoring at higher values of t and lower values of V .


Setting σ = 0 gives the conditions under which no monitoring is necessary:

(1− α)Π ≥ x(t)V

This implies ∂2x∂σ∂V

= 0 and ∂2x∂σ∂t

= 0. This condition is more likely to hold as the value

of future rewards increases.

Chapter 3

External Patrons

This chapter examines the motives of external actors that provide support to rebel

organizations and how these motivations shape which leaders are selected for support

and the amount of support they receive. Special emphasis is given to the motives

and influence of foreign governments, corporations, and diaspora communities. The

second half of the chapter formalizes the role of external patrons by investigating

the interactions between a patron and a rebel leader, based on the model developed

in the previous chapter. The main finding is that while some patrons benefit by

supporting trustworthy leaders and disciplined groups, other patrons, seeking to direct

the group’s activities, have an incentive to support lower quality leaders and withhold

resources that could otherwise strengthen control in the group.

3.1 The Motives of External Patrons

External actors provide support to rebel groups to achieve foreign policy objectives,

protect family members and ethnic kin, earn profits, or pursue ideological or hu-

manitarian causes. This section examines the motives of three common types of

patrons—foreign governments, corporations, and diaspora communities—and how

these motives affect their support for rebellion.

68

3.1. THE MOTIVES OF EXTERNAL PATRONS 69

3.1.1 State Sponsors

State support for insurgency has a long history. In the sixth century B.C.E, for

example, Athens and Eretria provided military assistance to the Ionians during their

rebellion against the Persians (Kohn 2007: 254). Today, state sponsorship remains

the most common type of external support for rebellion. Of the 73 groups covered in

the RAFD dataset, 45 groups (62%) received support from a foreign government at

some point during their history.1

Foreign governments sponsor rebel groups as a cost-effective strategy for accom-

plishing foreign policy objectives. Providing support to a rebel group is less expensive

than direct military intervention for the patron; and if successful, the rebels will gov-

ern the territory after the conflict, relieving the patron of the costs of occupation

(Salehyan 2008). State patrons use their rebel clients to increase their bargaining

leverage over a target government by weakening and destabilizing the conflict-affected

regime (Salehyan 2010).

In addition to saving costs, the rebel clients may have better local knowledge of the

terrain and population, increasing their military effectiveness relative to the patron,

and their status as locals can increase the legitimacy of the conflict (Salehyan 2010).

The ideological and political preferences of the rebels may also be compatible with

those of the patron. During the Cold War, for example the United States supported

anti-Communist insurgencies in places such as Nicaragua and Afghanistan, while the

Soviet Union backed Marxist guerrillas in Angola, Vietnam, and elsewhere. If rebels

capture the government, their preferred policies can create long-term benefits for the

state patron.

Sponsoring an insurgency, however, creates certain risks for states. First, pro-

viding support to rebels increases the risk of international war and violent reprisals

against the state patron (Bapat 2011; Gleditsch et al 2008; Schulz 2010). Second,

patrons risk losing control over the rebel group’s activities. The patron must be able

1Other studies find similar results. Byman et al.’s (2001) survey of insurgent movements findsthat 44 of the 74 rebel groups (59.5%) active during 1991-2000 received state support that wascrucial for their survival. According to the EACD dataset, over 44% of the 331 non-state armedgroups that were active during 1945-2003 (excluding coups) received military assistance from aforeign government (Cunningham et al 2009).

70 CHAPTER 3. EXTERNAL PATRONS

to credibly commit to ending violence if it hopes to gain leverage in negotiations.

If the state patron cannot control the rebel group, the conflict-affected state has no

incentive to offer concessions. (Bapat 2011; Salehyan 2010).

Faced with the risk of losing control over their rebel clients, state patrons can

employ three types of strategies to maintain their influence: selection, monitoring,

and sanctions (Salehyan 2010). State patrons may select rebels who share linguistic

and ethnic ties, or rebels who support their ideological and political goals. Selecting

such rebels can be tricky, however, as rebel leaders have an incentive to misrepresent

their preferences to attract external support. To screen for the most desirable rebel

leaders, patrons such as the United States, the Soviet Union, and Libya have at times

insisted on long periods of training and indoctrination for aspiring rebels, as a costly

signal of their commitment. Even after the patron selects a suitable rebel client, the

state may send military advisors or deploy soldiers within the rebel forces to monitor

the rebels’ behavior and gather information on the group’s plans.

When selection and monitoring mechanisms fail, state patrons can employ sanc-

tions against their rebel clients. The most straightforward sanction is ending their

support. By cutting off the flow of arms and money to the rebel leader and expelling

the rebels from their safe havens, the state patron increases the costs of rebellion and

decreases the rebels’ chances for success. This type of sanction is especially effective

when the rebel group cannot survive without the patron’s assistance, or at least when

the rebel leader cannot maintain a position of power in the group without external

support.

The extent of preference alignment between the state patron and potential clients

shapes the kind of rebel leader the state is likely to support and how much support

they are willing to provide. When states can find a compatible partner who shares a

strong attachment to the state’s ideological, political, or ethnic goals, states are more

inclined to support charismatic, trustworthy leaders who can mobilize a strong base

of support in the conflict areas. These patrons also have incentives to provide ample

support to strengthen the rebel group and help it achieve its goals.

When the state’s preferences do not align with those of the rebels, or the patron

is not able to screen potential clients and is unsure of their preferences, states face a


trade-off when providing support. In these cases, according to Idean Salehyan (2010:

506), state patrons “walk a fine line between empowering rebels enough to ensure

that they can impose costs on the target government but not so much that the rebels

do not fear abandonment nor be able to turn against the patron.” The best strategy

for these patrons is to support relatively weak leaders who lack personal bases of

support in the group. These leaders will be more dependent on the patron’s support

to maintain their position, and thus more willing to sacrifice group goals to satisfy

the patron’s orders. Further, to prevent the group from becoming too autonomous,

the patron may withhold resources that could be used to increase the group’s chances

of success and the leader’s ability to control her subordinates.

Rwanda and Uganda’s decision to support the AFDL in Zaire illustrates the dy-

namic that arises when patron preferences do not align with those of rebels. The

AFDL leader, Laurent-Desire Kabila, had experience leading rebellions in the 1960s,

but had no base of support in the group or among the local population. According

to Thomas Turner’s (2007) analysis, Uganda and Rwanda recruited Laurent-Desire

Kabila to lead the AFDL because they believed he was “weak enough to obey his

backers but strong enough to secure their common borders” (Turner 2007: 37). Lead-

ers with strong ties to their troops or the people might use this support to distance

themselves from the patron as the war progresses. Lower quality leaders, distrusted

by their troops and dependent on external support for their position, make better

partners for patrons who seek to induce costly actions from the rebels.

Although ideological, ethnic, or humanitarian objectives may play a role, the

desire to achieve a foreign policy objective, beyond the scope of the rebels’ goals,

is the driving force behind most state sponsors of rebellion. As a result, I predict

that rebel groups receiving state support will be characterized by more abusiveness

and lower quality leaders, on average, compared to groups that do not receive state

support.2

2Although not tested in the dissertation, this argument could also apply to paramilitary organiza-tions. Confronted by rebellion, some states supply weapons and encourage the formation of militias.The theory predicts that states have an incentive to maintain influence over the activities of theseparamilitary organizations, and thus will support relatively low-quality leaders who lack a base ofsupport among the soldiers and the population. The state provides weapons to these groups, butis unlikely to provide financial resources that could increase leader control. Supporting low-quality


3.1.2 Corporations

Corporations support rebellion to earn economic benefits or avoid some of the costs

associated with conflict, instability, and violence. Despite the media attention devoted

to conflict minerals and other economic activity during conflict, corporate support for

rebellion is relatively rare. Corporations provided high-level support to 12 of the 73

groups (16%) in the RAFD dataset. This statistic, however, ignores businessmen and

other economic actors who may exert influence over a rebel group during conflict.

Further, the frequency and importance of corporate sponsorship may be increasing,

due to the decline in state funding since the Cold War and the effects of economic

globalization (Ballentine and Nitzschke 2005).

Conflict can benefit corporations by decreasing regulation, taxation, and oversight,

but also threatens a company’s capital, staff, and corporate image. Corporations

gain by sponsoring rebel groups in areas that contain natural resources, smuggling

opportunities, or profitable markets. For example, the Taliban generated significant

revenue from partnerships with Pakistani trucking cartels and large-scale smugglers

during their fight to capture Afghanistan’s government in the 1990s (HRW 2001).

Unlike state patrons, corporations do not support rebels to achieve broader political

outcomes or increase their bargaining leverage in international negotiations.

What kind of rebel leaders are corporate sponsors most inclined to support? To

the extent that corporations have a desire to protect their reputation, they have an

incentive to support rebel groups that can provide security for the firm’s operations

and do not incite condemnation from the international community. Abusive rebel

groups, if connected to the corporation, can result in public scandals and interna-

tional sanctions on the company’s activities, reducing its profits. This incentive,

however, only applies to multinational corporations with large capital investments

and international accounts that can be frozen. Regional actors, by contrast, care

little about their international reputation. The trucking cartels that supported the

Taliban in Afghanistan, for example, faced little risk from international sanctions.

paramilitary leaders also mitigates the risk that the leaders will become popular and emerge as apolitical threat to the regime. The theory, therefore, predicts that paramilitary groups will tend tobe abusive towards civilians.


Corporations are more constrained in their choice of what kind of rebel leader

or group to support than their state counterparts. Unlike states, corporate spon-

sors cannot offer many of the critical resources that helps groups begin a rebellion.

Corporations do not control territory and are not legitimate end-users of military

equipment. As a result, corporations cannot offer safe haven or supply military hard-

ware to their rebel clients. Corporations can, however, provide financial resources to

their rebel clients and handle logistics associated with arms transfers. Rather than

recruiting rebel leaders and encouraging the formations of new groups, corporations

may be more inclined to decide whether or not to support existing groups.

By providing money to rebel leaders, corporations can strengthen control in the

rebel group. But because they are constrained in their choice of clients, corporations

may not always support high-quality leaders. In some cases, however, a corporation

can take actions to replace the rebel leader with someone more suited to its interests.

For example, Chapter 6 details a Lebanese timber company’s attempt to replace

the leader of Liberia’s MODEL rebels with another, higher-quality candidate. This

attempt failed, however, due to the intervention of MODEL’s primary supporter,

the government of Cote d’Ivoire. On average, I predict that rebel groups receiving

corporate support will be more disciplined than groups receiving support from state

sponsors, but will not necessarily have higher-quality leaders.

3.1.3 Diaspora Communities

Diaspora communities support rebellion to protect family members from insecurity

and government abuse as well as to shape political outcomes in their home countries.

The end of the Cold War increased the prominence of the diaspora as a means of

financing rebellion (Byman et al 2001). Numerous case studies have documented

the role of the diaspora in conflicts such as Sri Lanka, Northern Ireland, Kurdistan,

Somalia, and Palestine. Despite this attention, large-scale diaspora support for insur-

gency is relatively rare, and stronger financial regulations after the September 2001

terrorist attacks have further reduced the diaspora’s ability to fund insurgency. Of

the 73 groups in the dataset, only 10 groups (14%) received significant support from


the diaspora, although other groups may have indirectly benefited from the diaspora

through political pressure or remittances to family members in the conflict areas.3

Diasporas are more limited in the types of support they can provide than state or

corporate sponsors. Unlike states, diasporas cannot provide safe haven or direct access

to military supplies. And unlike corporations, diasporas do not manage international

operations that can provide logistical support. Nevertheless, diaspora groups can

provide political support and significant financial resources. Diaspora contributions

to Sri Lanka’s LTTE rebels during the late-1990s, for example, are conservatively

estimated at $18 million per year (Byman et al 2001: 50).

Diaspora support for insurgency, in contrast to support from states or corpora-

tions, is primarily motivated by ethnic concerns (Byman et al 2001). A diaspora

community may support a rebellion to win independence for an ethnic homeland

or protect family members from state-sanctioned violence. These communities have

incentives to support rebel leaders who are best able to create strong, effective in-

surgencies. The Hmong diaspora, for example, supported Vang Pao and his ULNLF

rebels to protect the Hmong people from abuses by the government of Laos (Lee

2005). Vang Pao was a hero of the Vietnam-era war in Laos and revered among the

Hmong for his military and political skills and commitment to his people (Washington

Post 2011). When diaspora communities are primarily concerned with the security of

an ethnic homeland and their families in the conflict areas, they are likely to support

high-quality leaders and encourage discipline among their rebel clients.

But diaspora communities do not always have an incentive to promote security

and can be coerced into financing less desirable groups. Unable to return home but

sheltered from the violence of war, members of the diaspora can promote extremist

agendas that are not in the rebels’ best interests. Rebels might be pressured to reject

a peace agreement or ethnically cleanse a territory to satisfy their diaspora patrons.

According to Paul Collier and Anke Hoeffler (2004), diasporas “preserve their own

hatreds” and significantly increase the risk of repeat conflict. Rebel groups can also

pressure members of the diaspora to provide funds. Sri Lanka’s LTTE rebels used

3Diaspora support may also be more common among smaller rebel groups that were not codedas controlling territory and thus do not appear in the dataset.


their sophisticated international networks to create social pressure on Tamil migrants

in Canada, Britain, and the United States. Groups might also threaten violence

against family members in the conflict areas to encourage contributions. In these

cases, diaspora groups have less ability to select high-quality leaders or influence

rebel behavior.

Although diaspora groups may, in some cases, encourage violence or succumb to

coercion, most diaspora communities become involved in rebellion to promote security

and improve the situation for their co-ethnics back home. As a result, I predict that

rebel groups receiving diaspora support will be characterized, on average, by higher

quality leaders and more discipline than groups receiving state support.

3.1.4 NGOs and Other Patrons

NGOs, civil society groups, and other actors can also become involved in supporting

rebellion. In particular, humanitarian aid organizations may feel pressure to provide

material support to rebel groups in exchange for access to the group’s territory. With-

out access, aid organizations cannot provide food, shelter, and medical assistance to

conflict-affected civilians; access also facilitates media attention and fundraising. Hu-

manitarian aid has a significant economic effect on conflict areas. The World Food

Programme (WFP), for example, spent some $11 million on food aid to Nicaragua

during 1989, which constituted over 1% of Nicaragua’s GDP.4 The humanitarian aid

industry has increased dramatically since the end of the Cold War. In the 1970s less

than $1 billion per year was spent worldwide on emergency aid to natural disasters

and conflicts; by the early 2000s, annual expenditures reached $10-15 billion (Fearon

2008b).

By providing aid to needy civilians, the aid industry is open to manipulation by

rebel groups. Rebels may have an incentive to exacerbate the humanitarian situation

in certain areas to attract more aid resources or use the presence of aid organizations

to bolster their political legitimacy (Lischer 2003). Rebel groups also gain valuable

food, medicine, and revenue by taxing and looting humanitarian aid. In Somalia, for

4Statistics from the World Bank. Available http://databank.worldbank.org/.


example, 5-10% of WFP’s 2009 aid deliveries—worth as much as $10-20 million—was

diverted to armed groups (UNSC 2010: 60). Even when aid reaches those in need,

rebel groups can tax civilians for a portion of their supplies. And by encouraging

civilians to relocate to refugee and displacement camps, humanitarian aid creates

captive audiences that rebel groups can exploit for taxation and recruitment.

Perhaps the most extreme example of misplaced aid occurred after the 1994 geno-

cide in Rwanda, when over 1 million refugees fled into neighboring areas of eastern

Zaire (now DR Congo). Hutu militants dominated the sprawling refugee camps and

used their control over aid distribution to increase their membership and revenues

(Adelman 2003). Rather than providing safety and relief for war-affected civilians,

the refugee camps became rebel bases, funded by the international community.

In some cases, aid organizations form partnerships with rebel groups. Rebel lead-

ers have an incentive to protect aid operations to provide support to civilians and

maintain access to food and medical supplies for their soldiers. Humanitarian aid

might also increase the incentives for discipline among rebel commanders and sol-

diers. During Ethiopia’s famine in the mid-1980s, humanitarian aid organizations

supplied an estimated $95 million to the TPLF rebels, who provided high levels of

security and other services to civilians during their struggle against the repressive

Derg regime (BBC 2010; CIA 1985).

NGOs, civil society groups, businessmen, and other actors can provide financial

and material support to rebel groups, but are unlikely to have the resources and

capacity to affect which rebel groups emerge. Humanitarian organizations generally

become involved in rebellion only after the conflict has begun. After all, the human-

itarian effects of conflict are precisely the reason for the humanitarian operations in

the first place. As a result, NGOs and other patrons are unlikely to significantly

affect the quality of the rebel leadership. By providing financial resources to the

existing leadership, however, these patrons may contribute to discipline among the

rebel forces.

3.2. THE IMPORTANCE OF EXTERNAL SUPPORT 77

3.2 The Importance of External Support

External support provides additional resources for rebellion, but also imposes costs

on the rebel group and its leader. External support can undermine the legitimacy

of the group among the population, and external patrons may require the group to

take actions that conflict with group goals. Whether or not a rebel leader accepts

external assistance depends on the patron’s goals and requirements, as well as the

leader’s ability to maintain an insurgency without the patron’s support.

Leaders who lack external support rely on their charisma and social networks to

attract committed followers who will follow orders even without monetary incentives

(Weinstein 2007). Rebel groups may survive for decades without external support,

but these insurgencies often remain small-scale and gradually fade into obscurity.

Scaling up a rebellion requires attracting recruits beyond the core membership and

supplying those recruits with weapons and other resources. The resources required

for large-scale rebellion—weapons, ammunition, and financing—generally come from

external partnerships.

The difficulty of fighting a war without external support is illustrated by the

challenge of acquiring arms and ammunition, the critical resources for rebel group

formation and survival. Although rebels have a variety of options for acquiring muni-

tions, large-scale shipments from an external patron provide substantial benefits that

are difficult to match through other means.

One option for acquiring arms and ammunition is to capture supplies by raiding

police stations or ambushing military convoys. Georgia’s civil war, for example, inten-

sified when former Soviet stockpiles fell into rebel hands (Demetriou 2002: 8-14). In

most cases outside of the former Soviet Union, however, rebel groups capture weapons

in small, unpredictable quantities. These weapons may be sufficient for occasional

ambushes, but offer little potential for rebel group expansion. And although captured

weapons may be reused over the course of the war, captured ammunition is quickly

spent. The complicated logistics required to carry out an offensive against the state

demand a reliable supply chain.

Some rebel groups manufacture their own weapons and ammunition. Such craft


production is perhaps most apparent in the conflicts in Iraq and Afghanistan, where

Improvised Explosive Devices (IEDs) account for the majority of U.S. combat deaths

(Wilson 2007). In Colombia, the FARC rebels are well-known for their use of chang-

ones, or homemade rifles, and the Philippines’ MILF rebels employed skilled gun-

smiths in multiple production facilities to produce hundreds of copies of the RPG-2

rocket-propelled grenade, the M-79 grenade launcher, .45 caliber pistols, and a variety

of anti-personnel mines (Davis 2003).

Relying on homemade supplies, however, is extremely difficult and costly for

rebel organizations. Even if the raw materials for homemade weapons are readily

available—metal pipes, for example—a large cadre of trained workers are required to

assemble the weapons. Further, homemade weapons are often unreliable, inaccurate,

and dangerous to use (Capie 2004: 7). Weapons production also depends on the sta-

ble control of territory by the rebel group. The MILF’s ability to produce weapons,

for example, was disrupted in 2000 after a government offensive (Davis 2003). The

challenges of producing homemade ammunition are even greater. Each soldier may

require hundreds of rounds of ammunition for a single attack and the raw materi-

als required to produce bullets are often as difficult to acquire as bullets themselves

(SAS 2006: 136-146). Whenever possible, rebel groups develop networks for procur-

ing modern, manufactured weapons and ammunition, rather than relying on craft

production (Capie 2004: 7).

Another alternative is to purchase weapons and ammunition from local markets.

For example, more than 6,600 gun dealers are located on the U.S.-Mexico border,

providing easy access to weapons for Mexican drug cartels (McKinley 2009). Local

arms markets, however, create problems for rebel groups due to unpredictable fluctu-

ations in supply and prices. In Somalia’s Bakara market, for example, prices for small

arms ammunition can vary as much as 50% from one month to the next (SAS 2006:

108; UNSC 2003b: 17). More importantly, relying on local markets is only feasible in

regions where arms are in abundant supply. If weapons need to be transported over

long distances, the risk of interception outweighs the potential profits of small scale

trade; only large shipments are economically feasible under these circumstances (SAS

2006: 111).

3.2. THE IMPORTANCE OF EXTERNAL SUPPORT 79

The most reliable option for procuring weapons and ammunition is through a part-

nership with a foreign government. Foreign governments have access to the military

supplies and logistics necessary for reliable, large-scale weapons shipments. These

states can often purchase weapons legally from the world’s top munitions producers

and thus avoid the black-market premium that rebel groups would otherwise have to

pay. By supporting the policy objectives of the foreign government, rebel groups may

even be able to obtain weapons at no monetary cost at all.

Foreign governments can provide a cost-effective source of arms for rebel groups,

but governments are not the only source for large-scale arms transfers. The explosion

in the availability of weapons and ammunition after the collapse of the Soviet Union

led to the emergence of international arms dealers (Byman et al 2001). The capability

of these “merchants of death” to supply weapons and logistics rivals that of any state.

Viktor Bout, the most notorious arms dealer, supplied weapons to numerous rebel

groups, governments, and international organizations—occasionally at the same time.

In Angola, both government and rebel forces relied on Bout’s shipments. In Congo

and Afghanistan, weapons for rebels competed for cargo space with supplies for the

United Nations (Economist 2009). When rebel groups cannot attract state support

for their activities, such large-scale smuggling networks may be their only means of

survival. In Sri Lanka, the LTTE rebels initially relied on munitions provided by the

Indian government. When this support ended in 1987, the group was forced to rely on

a variety of arms dealers in countries such as Croatia, South Africa, and Cambodia;

the group even acquired cargo ships to facilitate these arms transfers (Byman et al

2001: 117-22).

Foreign governments and arms dealers offer the best foundation for a large-scale

insurgency. These partnerships, however, come at a cost to the rebel group. The

group may have to sacrifice some of its goals or divert manpower to further the

policy objectives of its foreign supplier. The next section formalizes the patron-rebel

relationship.


3.3 A Model of External Support and Influence

External patrons supply weapons, resources, and other forms of support to rebel

groups and shape the quality and credibility of the leader by choosing which indi-

viduals to support. While some patrons benefit from disciplined rebel groups that

provide security in the conflict areas, other patrons benefit more from groups that

are willing to compromise group goals and take costly actions on the patron’s behalf.

External patrons play a role similar to venture capital firms. Aspiring rebel lead-

ers, like entrepreneurs, possess many of the raw materials required for a rebel group—

political and social grievances, access to commanders and recruits, military strategies,

and so on—but might lack the start-up capital necessary for group formation. With-

out support from external patrons, some rebel groups would never get off the ground.

A small investment in the right group at the right time overcomes these barriers to

entry and could yield high returns for the patron.

Going back to the basic model from the previous chapter, recall that a certain

amount of cooperation from the commander, denoted by xmin, is required for the

group to have any chance of success. For the analysis of external patrons, suppose

that xmin > 0. To form a group, the rebel leader must be able to offer some form of in-

centives to the commander, either through credible promises due to leader-commander

affinity or through cash transfers.

As a consequence of this assumption, rebel groups can be divided into two cat-

egories: groups that can meet the minimum threshold without support from an ex-

ternal patron (i.e. groups for which x(T = 0) > xmin), and groups that cannot

(x(T = 0) ≤ xmin). External patrons strengthen control in rebel groups by provid-

ing transfers of weapons and financial resources. Groups that otherwise might not

form become capable of threatening the state. But this assistance comes at a price:

external patrons benefit from rebel groups by achieving policy objectives, extracting

resources in rebel territory, or engaging in other activities with the group.

To formalize this situation, let the External Patron, P , choose a level of support

to provide to the rebel leader, T ∈ R+. Suppose that an external patron could

benefit in two ways by providing this support. First, the patron could benefit from

3.3. A MODEL OF EXTERNAL SUPPORT AND INFLUENCE 81

the effectiveness and power of the rebel group, ρ(x). More effective rebel groups are

better able to provide security for endangered civilians or economic activities, and are

more likely to capture the capital and become the new government. To the extent

that the patron benefits from the group’s effectiveness, we can say that the incentives

of the patron and leader are aligned.

Second, the patron could convince the rebel group to undertake some costly action,

A ∈ R+, on behalf of the patron. The action does not further the group’s goals, but

benefits the patron in some way. For example, the patron could demand that the rebel

group deploy its forces in border areas to protect the patron’s security, even though

those forces would be more productive for the group elsewhere. The patron could

also influence the timing of military offensives in the hopes of increasing its leverage

in international politics. Alternatively, patrons such as diaspora communities might

pressure the rebels to ethnically cleanse their territory, refuse offers of peace, or

undertake other actions that threaten the group’s local and international support.

To the extent that the patron seeks to induce costly actions from the group, we can

say that there exists a conflict of interest between the patron and the leader. Let

the costly action, A, be an exogenous parameter, not a strategic decision. Such a

situation occurs when the patron hopes to achieve a single, all-or-nothing policy goal.

A utility function for the patron captures the possibility of both aligned incentives

and conflicts of interest between the patron and the rebel leader. A possible functional

form for the utility of an accepted offer is:

UP (T ;A) = γ b ρ(x(T )) + (1− γ)A− T

where γ ∈ [0, 1] indicates the extent to which the patron’s incentives are aligned

with those of the rebel group, and b ∈ R+ is a scalar.5 Figure 3.1 depicts the model

with the addition of an external patron. The dotted lines represent the equilibrium

outcome of the basic model from the previous chapter, conditional on the level of

funding, T , provided by the patron.

To simplify the analysis, suppose that patrons fall into two extreme categories.

5If the leader rejects the patron’s offer, the utility would be the same except with A = 0 andT = 0.


PL

( x(T), t, !

)A

ccept

Reject

T" b

#(x(T

)) + (1

-")A-T

#(x(T

))!$ +

T - t - A

)!

●

●●

( x(0), 0

, ! )

L L))

" b #

(x(0))

#(x(0

))!$

Note:

The

dottedlines

representthe

basicm

odelof

leader-comm

anderinteraction

fromthe

previouschapter.

Figu

re3.1:

Am

odel

ofpatron

-leader

interaction

.


The first type of patron cares only about the effectiveness of the rebel group, such

that γ = 1. Since the patron’s incentives are perfectly aligned with those of the rebel

leader, patrons in this class can be described as “We’re-in-this-Together” patrons. The

second type of patron cares only about convincing the rebel group to take a costly

action, such that γ = 0. Since the patron wants to extract the maximum surplus

from the rebel group at minimal cost, this type can be described as “More-for-Less”

patrons. Each type is addressed in turn.

3.3.1 Type 1: The “We’re-in-this-Together” Patron

The “We’re-in-this-Together” patron provides a transfer to the rebel group and cares

only about the group’s effectiveness, ρ(x(T )). In this scenario, A = 0 and the rebel

leader bears no cost by accepting the patron’s offer. The external patron receives

utility UP (T ) = b ρ(x(T ))−T from supporting the group and the rebel leader receives

T . The unique equilibrium is summarized in the following proposition (proof in the

appendix):

Proposition 4. Let t∗ be the leader’s optimal cash transfer to the commander from

the previous chapter and let xl(T ) be the leader’s level of control when T ≤ t∗, also

defined in the previous chapter.

There exists some T1 ≥ 0 that solves ρ′(xl(T1)) = Vb+(1−α∗)Π

.

If xl(T1) > xmin, the “We’re-this-Together” patron’s optimal choice is T ∗1 =

min{T1, t∗}.

If xl(T1) ≤ xmin the patron offers T ∗1 = 0.

The leader always accepts the patron’s offer. The patron receives greater utility by

supporting groups with higher affinity and fewer lootable resources.

Because the patron’s incentives are aligned with those of the rebel leader, the

patron benefits from the same conditions that encourage control in the group. Im-

portantly, the groups most likely to be funded by such a patron are also the groups

that are most likely to form even in the absence of external support. Although ex-

ternal support increases control in the group, high-affinity groups can be disciplined

even without such support.


3.3.2 Type 2: The “More-for-Less” Patron

The “More-for-Less” patron provides a transfer to the rebel group and cares only

about its ability to convince the group to undertake an exogenously-determined costly

action, A ∈ R+. The external patron receives utility UP (T ;A) = A − T at the end

of the game if the group implements the action and nothing if the leader rejects the

patron’s offer. If the offer is accepted, the rebel leader receives T and bears cost A.6

The patron maximizes its utility by making the leader indifferent between accepting

the patron’s offer (A, T ) and forming a group without assistance, which leads to the

following proposition (proof in the appendix):

Proposition 5. Let t∗ be the leader’s optimal cash transfer to the commander from

the previous chapter and let xl(T ) be the leader’s level of control when T ≤ t∗, also

defined in the previous chapter.

There exists a unique T2 > 0 that solves ρ(xl(T2)) = ρ(xl(0)) + Aα∗Π

.

• If T2 < t∗ and T2 ≤ A the “More-for-Less” patron offers T ∗2 = T2.

• If T2 ≥ t∗ and t∗ ≤ A the patron offers T ∗2 = t∗.

• If min{T2, t∗} > A, the patron offers T ∗2 = 0.

The leader accepts the patron’s offer if T ∗2 ∈ [T2, t∗] and rejects the offer otherwise.

The “More-for-Less” patron maximizes its utility by supporting groups in which x(0) =

xmin.

As the rebels’ only source of financing, the “More-for-Less” patron can dictate

terms to the rebel group and extract whatever profit the group creates—up to the

point where the leader is better off forming a group without external assistance. If

the group could not form without assistance, the “More-for-Less” patron extracts all

surplus from the group, leaving the leader with zero utility. The patron, however,

6One might think that the cost of the action should be shared between the leader and commander.The leader, however, always makes the commander indifferent between joining the group and looting;the commander’s utility remains constant despite changes in group costs. Having the full cost borneby the rebel leader simplifies the analysis without substantively changing the results.


depends on the group to undertake a costly action, and thus benefits to some extent

from group strength and capabilities.

The patron’s competing incentives means that the patron benefits most by sup-

porting groups that could almost form without external assistance. The “More-

for-Less” patron’s payout increases with the leader’s credibility and group skill, but

decreases with the leader’s next best option. The patron maximizes its utility by

supporting the most qualified candidate from the pool of aspiring leaders who are

unable to launch a rebellion without support. In these groups, the leader depends

on the patron for her position and authority, and is willing to sacrifice the group’s

surplus to keep the patron on board. If the patron ceases its support, the leader will

not be able to induce x > xmin from the commander and the group will dissolve. The

leader’s promises are not quite credible enough for the group to survive on its own.

By supporting these groups, the patron achieves its objective for minimal cost.

The “More-for-Less” patron seeks out groups with somewhat trustworthy leaders

and skilled members—but not so trustworthy or skilled as to be able to motivate their

fighters and challenge the state without the patron’s support. This implies that, on

average, “More-for-less” patrons will be associated with groups that exhibit lower

levels of leader control and less leader-commander affinity than groups supported by

“We’re-in-this-Together” patrons. In the absence of external support, rebel groups

can form only if the leader’s promises are credible enough to achieve a minimal level of

cooperation, defined as xmin in the model. Since few groups emerge without support,

these groups are likely to be characterized by high levels of leader-commander affinity.

The influence of external patrons in rebellion leads to the following hypotheses:

Hypothesis 9. Patrons with preferences that conflict with those of the rebels

support groups with lower levels of leader control than patrons

with aligned preferences.

Hypothesis 10. Patrons with preferences that conflict with those of the rebels

support groups with less leader-commander affinity, on average,

than patrons with aligned preferences.

Hypothesis 11. Rebel groups that are able to form in the absence of external

support are characterized by high leader-commander affinity.


3.3.3 External Patrons and Commitment Problems

The model can be extended to account for the lack of externally enforced contracts

between patrons and rebel leaders. Suppose that, if the patron decides to support

the rebel leader, the patron provides T at the start of the game, but the rebel leader

decides whether or not to implement the costly action, A, at a later stage. If the

leader reneges on the agreement and refuses to implement A, the patron can take

actions to punish the leader, inflicting costs c ∈ R+ at the end of the game.

The war in Zaire/DR Congo provides an example of how patrons can punish

their former allies. The government of Rwanda installed Laurent Kabila as head

of the AFDL rebels and supported the group’s rapid advance across the country.

But after capturing the capital city, Kabila reneged on his promise to combat the

former genocidaires who were terrorizing the Rwandan government from bases in

Congo’s east. The Rwandan government reacted to Kabila’s duplicity by supporting

the formation of a new rebel group, RCD, to destabilize Kabila’s regime and pursue

its policy goals.

Under this new specification, the model requires that all agreements between the

patron and rebel leader be self-enforcing. Whether or not self-enforcement affects

the outcome of the game, however, depends on the type of patron supporting the

rebel group. The “We’re-in-this-Together” patron has incentives that are aligned

with those of the rebel leader. The patron provides a transfer, but the leader does

not have to implement any costly action in return. Since the leader has no incentive

to renege, the lack of externally enforced contracts has no effect on the patron-leader

relationship and the results of the game remain unchanged.

The situation is different when a “More-for-Less” patron is involved. This patron

supplies the minimum level of support necessary to convince the leader to implement

a costly action. These agreements create a risk of future opportunism, as the leader

will be tempted to renege on the agreement after receiving the patron’s transfer. The

leader will only implement the costly action if the costs of reneging exceed the costs of

the action. Understanding this, the “More-for-Less” patron will only support leaders

that can be punished severely. For example, the patron would favor leaders without

strong bases of support in their groups, making it easy to replace them if necessary.


Or the patron might support rebellions in weak states where other rebel groups can be

formed quickly if the first group does not meet the patron’s expectations. Introduc-

ing commitment problems into the model does not substantively change the results.

Commitment problems only intensify the incentives of a “More-for-Less” patron to

support a relatively weak rebel leader and withhold resources that could strengthen

leader control.

3.3.4 Multiple Patrons

The conflict economy creates various opportunities for rebel groups to form partner-

ships with foreign states, corporations, businessmen, arms dealers, or interest groups.

A rebel group might receive support from multiple patrons, or various patrons might

compete to earn a contract with a rebel group. The model is easily extended to

consider the effect of multiple patrons. Two scenarios are relevant. First, suppose at

least one of the patrons interested in supporting the rebel group is a “We’re-in-this-

Together” type with access to sufficient funds. Regardless of the characteristics of

the other patrons, the rebel group is guaranteed a source of revenue and support for

strong leadership. Multiple patrons strengthen control and discipline in the group.

Alternatively, suppose all of the patrons are “More-for-Less” types. None of the

patrons wants to supply ample resources or support a strong leader. But each patron

would benefit from convincing the rebel group to undertake a costly action. In this

situation, the rebel leader can leverage the patrons’ offers to increase the amount of

resources supplied to the group. To model this situation, let each patron make an

offer of Ti in exchange for Ai. The equilibrium outcome in this setting depends on

assumptions regarding how the bids from each patron are made, how many costly

actions the leader can implement, and the information available. Regardless of these

assumptions, the rebel leader will receive a transfer that is (weakly) greater than the

transfer received in the single patron case.7 Even when all patrons are “More-for-

Less” types, competition among patrons benefits the rebel leader and strengthens

control in the rebel group. This dynamic leads to a final hypothesis:

7See Andreu Mas-Colell et al (1995: Ch. 23) for a discussion of mechanism design and optimalstrategies for auctions.


Hypothesis 12. Rebel groups that receive support from multiple patrons are char-

acterized by higher levels of control than groups with only a single

patron.

The number and type of external actors willing to support rebellion depend on the

specific conflict environment and is largely exogenous to the rebel group itself. For

example, rather than competing with other patrons to support a single rebel group,

an aspiring patron could support the formation of a new rebel group. Alternatively,

a patron might hope to fund a rebel group but find no interest in rebellion among

the population or no leaders qualified to lead an insurgency. Conversely, a qualified,

charismatic rebel leader might have access to plenty of committed recruits, but be

unable to attract outside support for an insurgency. The “market for rebellion” that

matches patrons with rebel leaders could depend on numerous factors such as inter-

national relations, historical legacies, and natural resources. Many of these factors

are considered in the literature on civil war onset (e.g. Fearon and Laitin 2003), but

analyzing these variables in the context of external support for rebellion would be an

interesting avenue for future research.

3.4 Discussion

External patrons become involved in rebellion for their own reasons. Some patrons

hope to achieve policy goals, while other patrons hope to make a profit. Still other

patrons support rebellion to protect family members back home or advance an ide-

ological or ethnic agenda. When the preferences of patrons align with those of the

rebels, the theory predicts that patrons will use their influence to strengthen control

within the rebel organization, thereby increasing the leader’s ability to provide secu-

rity in rebel territories and prevent group factionalization. When preferences do not

align, the patrons have an incentive to support weaker leaders, providing the mini-

mum resources required to keep the leader in power to ensure that the leader stays

faithful to the patron’s wishes.

The theory focuses on two ways that patrons exert influence over the rebel group.

First, the patron decides how much cash to supply the rebel leaders, which the leader


can then use to provide spot-payments and create incentives for cooperation among

the top commanders. Second, the patron creates selection pressures that influence

which groups emerge from obscurity and which individuals are chosen to lead them.

Although it is probably rare for a patron to directly install a rebel leader, it has been

known to happen. Rwanda and Uganda chose Laurent-Desire Kabila to lead the

AFDL because they believed Kabila was strong enough to have a shot at capturing

the state, but weak enough to bow to the patrons’ wishes. In most cases, however,

the selection pressure is more subtle. Rebel groups often choose leaders with the goal

of maximizing the group’s access to external resources. Groups are willing to suffer

the pernicious effects of external influence because otherwise they would not have the

opportunity to rebel at all.

The theory’s focus on leader financing and selection pressure is perhaps too nar-

row, and arises largely from the requirements of clear modeling parameters and mea-

surable variables. External patrons likely influence rebels through a multitude of

subtle—and not so subtle—channels. For example, patrons influence groups through

the supply of weapons, the deployment of military advisors, and the promise of safe

haven, all of which are conditional on group performance. These forms of influence,

however, are not explicitly considered in the formal model or the regressions that

follow. A more nuanced look at external support and influence during Liberia’s Civil

War is presented in Part II of the dissertation.

3.A Mathematical Appendix


The patron solves:

maxT

bρ(x)− T

which leads to:

ρ′(x(T1))∂x

∂T1

=1

b(3.1)


In equilibrium the patron will always choose a transfer that is less than or equal to

t∗. To see this, consider the contrary, so that the patron transfers T1 > t∗. In this

situation, the rebel leader pockets all money beyond t∗, the optimal transfer to the

commander; any increase in T1 beyond t∗ does not increase the level of control in

the organization. Consequently, ∂x/∂T1 = 0 and the above expression cannot be

satisfied.

To solve for the patron’s equilibrium transfer, T ∗1 , consider a situation in which

T1 < t∗. According to Equation (2.4), this implies x(T1) = xl(T1) = e + T1

V+

ρ(xl(T1))(1− α∗) ΠV

. Implicitly differentiating with respect to T1:

∂xl∂T1

=1

V − ρ′(xl(T1))(1− α∗)Π

Substituting this expression into Equation (3.1) and rearranging yields:

ρ′(xl(T1)) =V

b+ (1− α∗)Π

There exists a unique T1 ≥ 0 that satisfies this expression because ρ′(xmin) is large

ρ′(·) is decreasing by assumption. Comparing this expression to Equation (2.3) reveals

that the patron will provide the optimal transfer T ∗1 = t∗ if b ≥ α∗Π; the patron will

provide a smaller transfer T ∗1 = T1 < t∗ otherwise. Since the patron receives no

benefit from supporting groups in which ρ(xl(T∗)) = 0, it is clear that xl(T

∗1 ) > xmin

whenever T ∗1 > 0.

The remainder of the proof concerns comparative statics. The effect of leader-

commander affinity on the patron’s utility is found by implicitly differentiating Equa-

tion (3.1) with respect to α:

∂xl∂α

=VΠ

ρ′′(xl(T ∗1 ))(b+ (1− α)Π)

Since ρ′′(x) < 0 for all x ∈ (xmin, 1), therefore∂xl(T

∗1 )

∂α< 0 and the patron’s utility

decreases with α. Since αopt is the minimum value of α, the patron has an incentive

to support groups in which the leader and commander share special affinities.


Repeating this process for V and Π reveals that the patron’s utility decreases

with the value of lootable resources and increases with the value of government. To

determine the effects of group skill on the patron’s utility, recall that higher group

skill translates into a larger value of ρ(xl(T )) for all xl(T ) > xmin. The increase in

ρ(xl(T )) leads to an increased marginal effect of T on xl, as shown by the expression

for ∂xl

∂T ∗1, above. The patron’s utility increases with group skill.


Proof. The leader’s reservation utility is what she could earn by forming a rebel group

with T = 0. The leader’s reservation utility is U0L = ρ (xl(0))α∗Π, where xl is defined

by Equation (2.4). If xl(0) ≤ xmin, the leader cannot form a rebel group without

external support and U0L = 0. With external support, the leader’s utility becomes

U1L = ρ (xl(T ))α∗Π + T − t− A.

The “More-for-Less” patron maximizes its utility by making the leader indifferent

between accepting the patron’s equilibrium transfer, T ∗2 , in exchange for implementing

A and her reservation utility. This implies:

T ∗2 = A+ t− [ρ(xl(T∗2 ))− ρ(xl(0))]α∗Π (3.2)

In equilibrium, T ∗2 is always less than or equal to t∗, the leader’s optimal transfer

to the commander. To see this, consider a counterfactual where the patron transfers

T2 > t∗. This implies that ∂t∗

∂T2= 0, since the leader never gives the commander more

than t∗. The excess resources are used for the leader’s personal profit, rather than

the production of group resources. The external patron could decrease T2 without

reducing the rebel group’s surplus or the leader’s willingness to implement A, thereby

increasing the patron’s utility. An optimal value of T ∗2 from the patron’s point of

view must be some T ≤ t∗.

For all T2 ≤ t∗, the rebel leader sets t = T2 in equilibrium. Equation (3.2)

simplifies to:

ρ (xl(T2)) = ρ (xl(0)) +A

α∗Π(3.3)


Since ρ(·) is increasing and strictly concave, there is a unique T2 ≥ 0 that satisfies

this expression. The patron never has an incentive to pay more than it receives from

the group, which leads to the final constraint: T ∗2 ≤ A.

The remainder of the proof concerns the claim that the patron maximizes its

utility by supporting a group in which x(0) = xmin. Consider a rebel group that

could form by itself, such that xl(0)) > xmin. In this case the maximum value of A

that can be extracted by the patron for a given value of T is:

A1 = [ρ(xl(T ))− ρ(xl(0))]α∗Π

Implicitly differentiating with respect to α∗:

∂A1

∂α∗=

[ρ′(xl(T ))

∂xl(T )

∂α∗− ρ′(xl(0))

∂xl(0)

∂α∗

]α∗Π + [ρ(xl(T ))− ρ(xl(0))] Π

Applying the proof for Proposition 1, it is straightforward to show that ∂xl(T )∂α∗

=∂xl(0)∂α∗

< 0 and ρ′(xl(T )) < ρ′(xl(0)). As a result,

∂A1

∂α∗> 0

When rebel groups can form on their own, a “More-for-Less” patron benefits by

supporting a rebel leader who gives fewer post-conflict rewards to her commanders

and thus has less credible promises. Among these groups, the patron maximizes its

utility by supporting groups for which xl(0) = xmin + ε.

Now consider a group that could not form without external assistance, such that

xl(0) ≤ xmin and ρ(xl(0)) = 0. In this case, the maximum value of A that could be

extracted by the patron for a given value of T is:

A2 = ρ(xl(T ))α∗Π

Implicitly differentiating with respect to α∗:

∂A2

∂α∗= ρ′(xl(T ))

∂xl(T )

∂α∗α∗Π + ρ(xl(T ))Π


Substituting the expression for ∂xl(T )∂α∗

in Equation (2.4) and rearranging yields:

∂A2

∂α∗=−ρ′(xl(T ))ρ(xl(T ))αΠ2

V − ρ′(xl(T ))(1− α∗)Π+ ρ(xl(T ))Π

This derivative is less than zero when:

V − ρ′(xl(T )))(1− α∗)Π < ρ′(xl(T ))α∗Π

This inequality is satisfied for all ρ′(xl(T )) ∈ (VΠ, 1

1−α∗VΠ

), which is always true by

Proposition 1. Thus,∂A2

∂α∗< 0

Among groups that could not form on their own, the patron benefits by supporting

leaders who can promise more rewards to their commanders and keep fewer post-

conflict spoils for themselves. Among groups that could not form, the patron benefits

most by supporting groups in which xl(0) = xmin.

To find the patron’s optimal choice, simply compare the values of A1 and A2 at

the points xl(0) = xmin + ε and xl(0) = xmin, respectively. It is clear that A2 > A1.

This result is generalizable so long as there are not increasing returns to T that offset

the effects of α. It is easily shown that ∂A1

∂T> 0 and ∂2A1

∂T 2 < 0, with similar results for

A2, and thus there are diminishing returns from T . The patron’s optimal choice is to

support a rebel group for which xl(0) = xmin.

Chapter 4

Rebel Abusiveness and

Factionalization

This chapter explores some of the theory’s predictions and causal mechanisms using

the Resources, Abuse, and Factionalization Dataset (RAFD), an original dataset of

all rebel groups that controlled territory during 1980-2003. The data do not allow for

an identification of causal effects, but can be used to assess how well the theory cor-

responds to the observed behavior of rebel groups compared to existing explanations.

The analysis shows that rebel leaders’ access to financing and credible promises are

strongly associated with lower civilian abuse and less risk of factionalization, while

lootable resources correspond to more abusiveness and a greater risk of factional-

ization. To account for possible endogeneity, I use pre-war resource endowments as

exogenous proxies for leader financing and lootable resources, and find similar results.

Peace talks are associated with a risk of factionalization that increases over time.

The results also show that the identities and motives of external patrons are sys-

tematically related to group behavior, as well as the leader’s credibility and access to

financing. Patrons with aligned preferences are associated with less abusive groups,

and also with increased leader financing, greater leader-commander affinity, and a

higher probability of post-conflict rewards for commanders. State patrons are, on

average, associated with more abusive groups and less leader financing. Rebel groups

94

4.1. DATA 95

receiving state support are 177% more likely to factionalize in their first year, al-

though this risk diminishes over time. Diaspora communities are associated with

less abusive rebels. Diaspora-supported rebels are 98% less likely to factionalize

in their first year than the baseline, although this effect also diminishes over time.

Commercially-motivated patrons are associated with less abusive groups and greater

leader financing, but, contrary to the theory’s predictions, appear to increase the risk

of group factionalization.

4.1 Data

To create the RAFD dataset, I begin with the universe of cases identified in the

Expanded Armed Conflict Dataset (EACD) v2.4 (Cunningham et al 2009). The

EACD dataset is composed of dyads that usually include a state and a rebel group.

I restricted the data to include only rebel groups that were coded as controlling

territory as some point during 1980-2003. This restriction is important because the

theory focuses on the delegation problem between rebel leaders and commanders,

which depends on the commander having autonomous access to weapons, manpower,

and lootable resources. This excludes most coups as well as small-scale insurgencies.

Relying on the EACD dataset has both limitations and benefits. The dataset

overlooks some important rebel groups and many splinter factions, but reduces the

risk of bias from selecting a new universe of cases. Most importantly, the EACD

dataset ensures compatibility with the larger PRIO dataset, which is commonly used

in studies of civil war.

For the set of rebel groups identified in the EACD dataset, I created entries for

each rebel-year, according to the EACD’s start and end dates for each group. For

example, some Burmese rebel groups formed in the 1940s and remain active to this

day. The dataset contains yearly records for these groups from the 1940s through

2008, when the data are censored. Some groups, therefore, may have 50-60 entries in

the dataset, even though the primary period of interest is 1980-2003. The result is a

dataset that includes 73 armed groups in 41 countries and 768 group years.

The dependent and explanatory variables were coded largely based on Keesing’s

96 CHAPTER 4. REBEL ABUSIVENESS AND FACTIONALIZATION

record of world events, the Uppsala UCDP conflict database, and Fearon and Laitin’s

(2007) conflict termination narratives. When available, I used other secondary sources

such as reports by Human Rights Watch, Amnesty International, and the United

Nations, as well as academic books and articles. For each group, I wrote narratives

that provide justification for the coding along with citations of sources.1

4.1.1 Dependent Variables

The dependent variables are civilian abuse and group factionalization. Civilian abuse

includes looting, rape, forced displacement, and physical violence against civilians.

CivAbuse is scaled from 0 to 1. Low civilian abuse (0 on the scale) corresponds to

situations where civilian casualties were confined to collateral damage from attacks

on military targets and possibly some selective violence against government agents.

Isolated reports of looting or civilian abuse during military operations are coded as

0.25 on the scale, while moderate civilian abuse (0.5 on the scale) corresponds to

sporadic reports of looting, rape, and/or killing of civilians; more common reports

are coded as 0.75 on the scale. High civilian abuse (1.0 on the scale) corresponds to

widespread looting, rape, forced displacement, and/or mass killing. Table 4.1 lists

the rebel groups in order of their average level of civilian abuse.

This scale of civilian abuse is best illustrated by several examples. In Djibouti,

for example, media reports chronicled civilian abuse by government forces but did

not mention any abuses by the FRUD rebels during the 1991-94 conflict. Secondary

sources emphasized the rebels’ good relations and considerable support among the

Afar people (Schraeder 1993). The CivAbuse variable for FRUD, therefore, was coded

as 0 for each year of the war. Other rebels, such as Sierra Leone’s RUF, were extremely

predatory and I coded CivAbuse as 1.0 for each year of their operations. Liberia’s

NPFL and INPFL rebels, the only Liberian groups included in the dataset, each

received a coding of 0.75.

Some rebel groups, such as Colombia’s ELN rebels, demonstrated considerable

variation in their treatment of civilians over the course of their lifespan. After forming

1These narratives are available from the author by request.

4.1. DATA 97

Avg. Civilian Abuse Rebel Group1 AFDL Renamo

Croat. B. & H. Rep. AbkhaziaFaction of SPLM Rep. S. OssetiaFAN RUFMJP Serb. B. & H.MPIGO USC FactionRCD Faction

0.99-0.50 FROLINAT ELFRCD MFDCLeb. Nat’l Mov. GAMLeb. Front Rep. ChechyaINPFL Rep. CroatiaNPFL Serb. KrajinaRep. Nag.-Karab. TalibanUIFSA

0.49-0.01 ELN UCKSPLM UWSAUFLA Faction UNITABCP CPN-M/UPFAut. W. Bosnia FARCBRA SSAD.R. Yemen LTTEFatah EPLFMLC KNUMPCI OLF

0 CPM POLISARIODniestr Rep. PUKEZLN Rep. DagestanFARS Rep. SloveniaFDR SNMFMLN TPLFFRUD UFLAGov’t Armenia ULNLFKDP/DPK UNRFKIO UPAMPA UPM/NRAMTA ZviadistsMuslim Bro.

Table 4.1: Average civilian abuse by rebel group, in order of descending abuse.


in 1966, the ELN remained fairly weak until the early 1980s. No reports of significant

abuse exist for this period and CivAbuse is coded as 0. The group began to raise

funds through kidnapping and extortion in the 1980s, and accelerated its efforts in the

mid-1990s with mass abductions (HRW 1998; BBC 2002). To reflect this change in

behavior, I code CivAbuse as 0.5 for the 1981-1995 period and as 1.0 for the 1996-2008

period.

Group factionalization is defined as an event in which the rebel leader loses control

over part of the group’s military force, which then declares itself to be an independent

group. Factionalization includes situations where a rebel group splits into two or more

competing, violent factions, as well as situations in which part of the rebel military

force decides to disarm or participate in a peace process against the orders of the rebel

leader. Factionalization does not include political divisions among rebel officials that

do not result in the emergence of an independent armed organization. The variable

Factionalize is coded as 1 if a factionalization event happened in a given year and 0

otherwise. I also coded the variable FacAttempt as 1 if rebel members attempted to

split from the leadership in a given year, whether or not the attempt resulted in the

emergence of a new faction. Table 4.2 lists the rebel groups, their lifespan in years

according to the EACD dataset, and the number of factionalization events. The last

column indicates each group’s average risk of factionalization, which is the number

of successful factionalization events divided by the group’s lifespan.

Burma’s Kachin Independence Organization (KIO) illustrates the factionalization

codings. The KIO formed in 1961 under the leadership of Brang Seng and suffered

a split in 1968 when one of it’s military wings, the “People’s Army 101 War Zone”,

declared its independence from the leadership (South 2008: 153). In 1975, three

of the group’s top officials were shot dead by group members in a dispute about

the distribution of group funds, but leader Brang Seng maintained control over the

military forces following this event. In 1991, however, the leader failed to prevent the

group’s Fourth Brigade from signing a truce with the government and forming the

independent Kachin Defense Army. In the dataset, I code Factionalize as 1 for 1968

and 1991 and FacAttempt as 1 for 1968, 1975, and 1991.

4.1. DATA 99

Group Lifespan Factionalize Attempts Failed Avg. RiskAFDL 2 0 0 0 0.00Aut. W. Bosnia 2 0 0 0 0.00BCP 43 1 4 3 0.02BRA 8 0 0 0 0.00CPM 3 0 0 0 0.00CPN-M/UPF 11 0 0 0 0.00Croat. B. & H. 2 0 0 0 0.00D.R. Yemen 1 0 0 0 0.00Dniestr Rep 1 0 0 0 0.00ELF 21 2 2 0 0.10ELN 43 1 1 0 0.02EPLF 20 0 0 0 0.00EZLN 1 0 0 0 0.00Faction SPLM 7 2 2 0 0.29FAN 12 1 1 0 0.08FARC 43 0 0 0 0.00FARS 1 0 0 0 0.00Fatah 8 0 1 1 0.00FDR 1 0 0 0 0.00FMLN 12 0 0 0 0.00FROLINAT 25 4 4 0 0.16FRUD 4 1 1 0 0.25GAM 10 0 1 1 0.00Gov’t Armenia 2 0 0 0 0.00INPFL 3 0 0 0 0.00KDP/DPK 31 1 1 0 0.03KIO 32 2 3 1 0.06KNU 60 5 5 0 0.08Leb. Front 8 2 2 0 0.25Leb. Nat’l Mov. 10 1 1 0 0.10LTTE 26 1 1 0 0.04MFDC 14 4 4 0 0.29MJP 2 0 0 0 0.00MLC 4 0 0 0 0.00MPA 1 0 0 0 0.00MPCI 1 0 0 0 0.00MPIGO 2 0 0 0 0.00MTA 21 2 2 0 0.10Muslim Bro. 4 0 0 0 0.00NPFL 7 2 2 0 0.29OLF 26 0 0 0 0.00POLISARIO 15 0 0 0 0.00PUK 18 0 0 0 0.00RCD 4 1 2 1 0.25RCD Faction 3 2 3 1 0.67Renamo 17 0 1 1 0.00Rep. Abkhazia 2 0 0 0 0.00Rep. Chechnya 12 1 1 0 0.08Rep. Croatia 1 0 0 0 0.00Rep. Dagestan 1 0 0 0 0.00Rep. Nag-Karab. 3 0 0 0 0.00Rep. Slovenia 1 0 0 0 0.00Rep. S. Ossetia 1 0 0 0 0.00RUF 10 0 0 0 0.00Serb. B. & H. 4 0 0 0 0.00Serb. Krajina 4 0 0 0 0.00SNM 10 0 0 0 0.00SPLM 22 3 6 3 0.14SSA 20 0 0 0 0.00Taliban 2 0 0 0 0.00TPLF 16 0 0 0 0.00UCK 2 0 0 0 0.00UFLA 3 1 1 0 0.33UFLA Faction 16 0 0 0 0.00UIFSA 6 1 1 0 0.17ULNLF 2 0 0 0 0.00UNITA 41 1 1 0 0.07UNRF 6 0 0 0 0.00UPA 3 0 0 0 0.00UPM/NRA 6 0 0 0 0.00USC Faction 6 1 1 0 0.17UWSA 1 0 0 0 0.00Zviadists 1 0 0 0 0.00

Note: Bold indicates the 21 most-abusive groups.

Table 4.2: Group lifespan (years) and factionalization.


4.1.2 Explanatory Variables for Commander Incentives

The theory’s primary causal mechanism relates to the leader’s ability to offer cash

incentives to commanders, the commanders’ access to lootable resources, and the

credibility of the leader’s promises. Precise estimates of rebel financing are not pos-

sible to obtain from secondary sources, and rebel groups often obscure their access

to resources. I was, however, able to estimate the order of magnitude of these re-

sources by using available secondary sources such as policy briefings and historical

studies.2 LeaderCash measures the leader’s access to financial resources and is coded

on a logarithmic scale. For example, financial resources of $10,000-99,999 in a given

year are coded as 4 on the scale, while resources in the range of $100,000-999,999

are coded as 5, and so on. As a baseline, I assume that all rebel leaders have access

to enough funds for their personal expenses, which I estimate to be in the range of

$10,000-$99,000; the baseline value of LeaderCash is 4.

In some cases, I coded LeaderCash by extrapolating pre- or post-war revenues

from key resources. For example, the leader of the breakaway Republic of Croatia,

Franjo Tudjman, controlled Croatia’s port during 1991. The port currently generates

an estimated 90 million euros in annual revenue for the Croatian government, but

this revenue was likely significantly lower during the war. Although no data exist for

this period, I estimate that Tudjman generated revenues equivalent to several million

dollars, and code LeaderCash as 6.

Better estimates exist for some rebel groups. A recent investigation by the BBC

determined that the leader of Ethiopia’s TPLF rebels, Meles Zenawi, generated an

estimated $95 million by taxing humanitarian organization during the famine of the

mid-1980s (BBC 2010). Reports also indicate smaller financial transfers to the TPLF

from other insurgent organizations during the early years of the war, but do not specify

the precise amounts. I estimate the annual revenue from these early transfers to be in

the range of $100,000-999,999. For the TPLF, I code LeaderCash as 5 for 1976-1983

when Zenawi relied mainly on small financial transfers, and as 7 for 1984-1991 when

a massive influx of humanitarian organizations generated increased revenue for the

2Narratives of each group include a bibliography of sources used and are available from the authoron request.

4.1. DATA 101

group.

Loot measures the maximum value of the lootable resources in the group’s territory

that could be extracted by a single commander and his troops, and is coded on the

same logarithmic scale. When rebel groups operate in very poor countries and have

access only to civilian assets, I estimate that looting would generate annual revenues

of $10,000-99,999 for a commander who devoted his time to predation. I code the Loot

variable as 4 in these cases. In wealthier countries, such as the former Yugoslavia,

looting civilian assets might generate hundreds of thousands of dollars in revenue. In

these cases, I code Loot as 5.

Lootable natural resources such as alluvial diamonds or illicit narcotics can dra-

matically increase commander payouts. Burma’s UWSA rebels, for example, operated

in areas of large-scale opium production. The group’s leader, Bao You, generated tens

of millions of dollars through drug production and large-scale exports (Chouvy 2010:

28). Although group commanders did not have access to these international markets,

they could probably have earned millions of dollars by looting the group’s stockpiles

of opium and heroin. For UWSA, I code Loot as 6. In Sierra Leone, RUF commanders

could easily sell rough diamonds to Lebanese intermediaries. With annual diamond

exports estimated at $25-125 million (UNSC 2000b: 17), I code Loot as 7 for the

RUF. Table 4.3 lists the rebel groups in the dataset according to their average access

to leader financing and lootable resources.

I code two different measures of leader credibility. The first measure, Affinity,

identifies any personal or ethnic bonds between the leader and commander. Personal

and familial bonds increase the credibility of the leader’s promises, even in the absence

of binding contracts. The LTTE’s top commanders, for example, shared family ties

with the group’s leader. But in most cases it is difficult to identify any pre-war

relationship between leader and commander. The Affinity variable, therefore, often

relies on ethnic background as a proxy for personal ties, which can be problematic.

For example, the top commander of Somalia’s USC-Mustihil faction, Osman Ali Ato,

was a member of the same Hawiye clan as the group’s leader, Mohammed Farah

Aideed. Nevertheless, considerable tension existed between Ato and Aideed, and

news media commented on their rivalry even before Ato split from the group in 1995.


Avg.Value LeaderCash Loot8 D.R. Yemen7 Aut. W. Bosnia MTA RUF

FARC NPFLFatah RCDFMLN TalibanGov’t Armenia TPLFKIO UCKLTTE UNITAMPCI UWSA

6 BCP AFDL RCDGAM FARC RCD FactionMLC Fatah RenamoRCD Faction KIO UIFSARep. Chechnya Leb. Front UNITARep. Croatia MTA UWSARep. Slovenia NPFL

5 Dniestr Rep. SSA Aut. W. Bosnia Rep. ChechnyaELN ULNLF BCP Rep. CroatiaEPLF UPM/NRA Croat. B. & H. Rep. SloveniaKDP/DPK USC Faction ELN Serb. B. & H.KNU FMLN Serb. KrajinaPOLISARIO GAM SSAPUK Gov’t Armenia TalibanRenamo KNU TPLFRep. Abkhazia MFDC UCKSNM MLC UFLA FactionSPLM MPCI USC Faction

4 AFDL MPIGO BRA MPABRA Muslim Bro. CPM MPIGOCPM OLF CPN-M/UPF Muslim Bro.CPN-M/UPF Rep. Dagestan D.R. Yemen OLFCroat. B. & H. Rep. Nag.-Karab. Dniestr Rep. POLISARIOELF Rep. S. Ossetia ELF PUKEZLN RUF EPLF Rep. AbkhaziaFaction SPLM Serb. B. & H. EZLN Rep. DagestanFAN Serb. Krajina Faction SPLM Rep. Nag.-Karab.FARS UFLA FAN Rep. S. OssetiaFDR UFLA Faction FARS SNMFROLINAT UIFSA FDR SPLMFRUD UNRF FROLINAT UFLAINPFL UPA FRUD ULNLFLeb. Front Zviadists INPFL UNRFLeb. Nat’l Mov. KDP/DPK UPAMFDC Leb. Nat’l Mov. UPM/NRAMJP LTTE ZviadistsMPA MJP

Note: Bold indicates the 21 most-abusive groups.

Table 4.3: Leader financing and lootable resources, by rebel group.

4.1. DATA 103

By focusing on observable proxies such as ethnicity, Affinity is an imperfect, but

still useful, measure of the personal ties and trust that exist between leader and

commander.

The second measure of credibility, Reward, indicates whether the top commander

received some post-conflict reward, conditional on the rebel group earning rewards

that could be distributed. In contrast to Affinity, this variable focuses on concrete

indicators of the leader-commander relationship. But this variable is also problematic

because it relies on post-war payouts to explain commander behavior during the war.

The variable assumes that commanders, as rational actors, have some expectation of

the rewards they will receive after the conflict. An additional shortcoming is that

Reward cannot measure the leader’s willingness to share post-conflict rewards among

rebel groups that were defeated militarily or, alternatively, remained active in 2008,

when the data are censored. These groups, therefore, must be excluded from models

involving the Reward variable.

Reward sometimes contradicts the Affinity indicator. For example, the leader of

Uganda’s UNRF rebels, Moses Ali, accepted a post-war position in government but

did not reward his top commander, Amin Onzi, even though the two shared ethnic

ties. Although Affinity would assume a connection between these individuals, Reward

implies that the leader’s promises were not credible. Both Affinity and Reward have

limitations but each provides some insight into the leader-commander relationship.

Among groups that earned post-conflict rewards, these variables exhibit a significant,

positive correlation of 0.70.

Finally, the theory predicts that peace talks strain the leader-commander relation-

ship, making factionalization more likely. To test this prediction, I code PeaceTalks

as 1 for every year in which peace talks occur.

4.1.3 Explanatory Variables for External Patrons

An external patron is any actor outside the rebel organization that provides support

central to the rebel group’s activities, including weapons, ammunition, food, financ-

ing, and safe haven. Diplomatic support, including statements on behalf of the rebel


group, do not count as external support in this analysis. The theory predicts that

patrons with aligned preferences will provide sufficient financial resources to trustwor-

thy leaders. These patrons, therefore, will be associated with more disciplined rebel

groups with a lower risk of factionalization. Patrons with conflicting preferences will

provide minimal resources to less credible leaders and will be associated with more

predatory, faction-prone groups. Rebel groups that attract support from multiple

patrons are predicted to be more disciplined than groups relying on a single patron.

The variable Patron is coded as 1 for every year in which the rebel leader receives

some support from an external patron. MultiPatron is coded as 1 if the group received

support from multiple patrons in a given year.

Because patrons do not provide objective statements about their motives, I have

to rely on various proxies to measure patron incentives. The most straightforward

proxy is the identity of the patron. StateSupport is coded as 1 if the rebel leader

received support from a foreign state in the given year. CorpSupport is coded as 1 if

the leader received support from a corporation; and DiaSupport is coded as 1 if the

leader received support from a diaspora community.

Foreign states support rebellion for geopolitical reasons such as increasing their

regional influence or destabilizing a neighbor (Byman et al 2001). Although other

motivations such as ideology or ethnic ties may be important for state sponsors, these

patrons are likely to require the rebel group to undertake costly actions to further their

political objectives. For the purpose of the analysis, state supporters are considered to

be “More-for-Less” patrons and are predicted to support groups with less-trustworthy

leaders, higher civilian abuse, and greater risk of factionalization. Corporate sponsors

become involved in rebellion to earn a profit and thus benefit from a strong rebel group

that can provide security. Diaspora communities often seek to protect family members

back home. For this analysis, corporations and diaspora communities are considered

to be “We’re-in-this-Together” patrons and are predicted to support groups with less

civilian abuse and lower risks of factionalization.

Relying on patron identities alone is somewhat problematic. A foreign state, for

example, might favor a strong rebel group for ideological or humanitarian reasons,

while a diaspora community may pressure rebels to ethnically cleanse their territory.

4.1. DATA 105

The variables also ignore civil society and aid organizations. As an alternative mea-

sure, I code several variables that relate to the patron’s motives, based on my analysis

of secondary sources. Ideology indicates whether the leader received support from a

patron motivated by ideology, most notably Marxism during the Cold War. Ethnic-

Ties indicates whether the leader received support from a patron with ethnic ties to

the population in the conflict area. Commerce indicates whether the leader received

support from a patron with commercial interests in the conflict area. Religion indi-

cates whether the leader received support from a patron with religious interests, such

as a jihadist group. Humanitarian indicates whether the leader received support from

a patron with humanitarian objectives, such as an aid organization.

Each of these variables indicates some degree of preference alignment between

the patron and the rebel group. The binary variable AlignedPrefs indicates whether

any of these variables equals 1 in a given year.3 The theory predicts each of these

variables to be negatively related to civilian abuse and the risk of factionalization.

The opposite would hold when the interests of patrons and rebels conflict. Conflicting

interests are difficult to code objectively and are left as a residual category, indicated

by a 0 value for AlignedPrefs.

Table 4.4 indicates whether each rebel group received support from foreign states,

diaspora communities, or corporations at any point,4 and also indicates the patron’s

likely motives for supporting the group. The line in the center of the table separates

groups that received only state support from groups that received support from non-

state actors or no external support. The 21 most abusive groups are bolded in the

table. There is a striking correlation between the most abusive groups and those that

received support from a state without aligned preferences, clustered at the top of the

table.

3Alternative codings could consider the sum of the aligned preference indicators or some propen-sity score. This binary variable, however, captures the essence of the argument.

4Patrons such as businessmen, civil society organizations, and aid organizations are not consideredto fall in any of these three identity categories.


Group State Diaspora CorporateName Support Support Support Ideology Ethnic Ties Commerce Religion HumanitarianAFDL XD.R. Yemen XELF XFaction SPLM XFAN XFROLINAT XINPFL XLeb. Front XLeb. Nat’l Mov. XMJP XRenamo XRep. Abkhazia XRep. S. Ossetia XUFLA Faction XBCP X XCroat. B. & H. X XFMLN X XKDP/DPK X XMFDC X XPUK X XRCD Faction X XRep. Nag-Karab. X XRUF X XSerb. B. & H. X XSerb. Krajina X XSPLM X XSSA X XUIFSA X XFARC X X XPOLISARIO X X XEPLF X X X XFatah X X X X XGAM X X X X XKIO X X X XKNU X X XLTTE X X XMLC X X XMPCI X X X XNPFL X X X XOLF X X XRCD X X XSNM X X XTaliban X X X XUNITA X X XUPM/NRA X X X XAut. W. Bosnia XELN XRep. Chechnya X X XUCK X XULNLF X XCPN-M/UPF XFRUD XMTA XTPLF XUSC Faction XUWSA XBRACPMDniestr RepEZLNFARSFDRGov’t ArmeniaMPAMPIGOMuslim Bro.Rep. CroatiaRep. DagestanRep. SloveniaUFLAUNRFUPAZviadists

Note: Bold indicates the 21 most-abusive groups. The horizontal line divides groups receiving onlystate support from groups receiving non-state support or no external support.

Table 4.4: External patron identity and motives, by rebel group.

4.1. DATA 107

4.1.4 Exogenous Proxy Variables

All of the explanatory variables suffer from a fundamental problem: they measure the

observed relationship between resources, external patrons, and rebel group behavior.

These relationships might depend on unobserved characteristics of the group and

therefore create an identification problem. Further, the variables may be biased due

to systematic measurement error. To overcome these problems to some extent, I use

exogenous proxies for two types of resources at the state level.

The first exogenous proxy, CapitalResources, is a binary variable that indicates

that a capital-intensive resource extraction industry existed in the state before the

start of the conflict. Natural resources that require capital-intensive means of ex-

traction include oil, Kimberlite diamonds, iron ore, timber, phosphates, and other

minerals. The key assumption is that these resources require investment and exper-

tise, and thus cannot be extracted by the rebel group without external assistance.

This variable does not include offshore oil, because a corporation would not have to

form a partnership with a rebel group to gain access to offshore reserves. The value of

these resources means that patrons are more likely to seek commercial partnerships

with rebel groups and provide financing to the rebel leader. The variable proxies for

LeaderCash and is predicted to correspond to more disciplined rebel groups.

The second exogenous proxy, LootResources, is a binary variable that indicates

whether a state possesses valuable natural resources that could be extracted without

large capital investments. Eligible resources include alluvial diamonds and gems,

as well as illegal narcotics. LootResources equals 1 if these resources were being

extracted at a significant scale anywhere in the country before the start of the conflict.

This variable proxies for Loot and, in contrast to CapitalResources, this variable is

predicted to correspond to higher levels of civilian abuse.

These variables are determined by geographic characteristics and are therefore

exogenous to rebel group characteristics. The effects of these variables on civilian

abuse can occur only through the group’s access to resources. These variables are

coded based on the CIA World Factbook, which lists natural resource endowments for

each country. Additional research determined whether the natural resources in each

country require capital-intensive extraction and whether the resources were exploited


before the start of the war. Binary variables are an extremely rough coding, but are

less prone to bias or coding errors than continuous variables.

4.1.5 Competing Theories and Control Variables

Other variables relate to competing theories of rebel group behavior. To test the-

ories of ethnicity, I code EthnicRebellion as 1 if the group was largely composed of

and claimed to represent a single ethnic group, defined as people either sharing a

recognized ethnicity or living in a well-defined homeland. To examine the effects

of separatist versus center-seeking conflicts, I code Separatist based on whether the

group appears in Marshall and Gurr’s (2005) list of separatist movements. The binary

variable Democracy indicates whether the government is democratic in a given year,

defined as having a combined score of at least 6 on the Polity IV scale (version 2009).

I also include a number of control variables in the dataset. To measure the effects

of government behavior, I code GovAbuse according to the same scale as CivAbuse.

I also code Peacekeepers to indicate whether peacekeepers were deployed in the con-

flict area. To identify regional effects I code dummy variables to indicate whether the

conflict took place in Africa, the Americas, Asia, the Middle East, or in the former

Socialist Bloc. Finally, I include measures of GDP per capita, GDPpc; the country’s

population, Pop; the number of refugees, Refugees ; estimated battle deaths, Bat-

tleDeaths ; and the amount of spending in US dollars by the World Food Programme,

WFP, as a measure of humanitarian aid. These data were collected from the World

Bank, UNHCR, and the PRIO dataset (Gleditsch et al 2002; Lacina and Gleditsch

2005). Table 4.5 provides descriptive statistics of the variables in the rebel group-year

dataset.

4.2 Civilian Abuse

The theory predicts that greater access to financing and credible promises by the

leader correspond to lower civilian abuse, while more valuable lootable resources cor-

respond to greater civilian abuse. Leader financing is acquired through partnerships

4.2. CIVILIAN ABUSE 109

Variable Min Max Mean St. DevCivAbuse 0 1 0.30 0.37Factionalize 0 1 0.06 0.23Loot 4 7 4.77 0.85LeaderCash 4 8 5.26 1.17Affinity 0 1 0.71 0.45NoReward 0 1 0.18 0.39PeaceTalks 0 1 0.25 0.43Patron 0 1 0.89 0.31MultiPatron 0 1 0.38 0.49StateSupport 0 1 0.64 0.48CorpSupport 0 1 0.10 0.30DiaSupport 0 1 0.15 0.36Ideology 0 1 0.15 0.36EthnicTies 0 1 0.25 0.43Commerce 0 1 0.30 0.46Religion 0 1 0.01 0.12Humanitarian 0 1 0.10 0.30AlignedPrefs 0 1 0.73 0.44EthnicRebellion 0 1 0.56 0.50Separatist 0 1 0.53 0.50Democracy 0 1 0.23 0.42GovAbuse 0 1 0.71 0.36Peacekeepers 0 1 0.08 0.28GDPpc 62 20,710 1,170 2,516Pop 507,200 1,140,000,000 52,720,000 156,296,465Refugees 0 3,810,000 247,500 496,163BattleDeaths 0 100,500 3,655 8,014WFP -630,000 248,800,000 9,534,000 22,778,291

Table 4.5: Descriptive statistics for the RAFD dataset.


with external patrons, whose motives shape the amount of financing provided. These

patrons also influence leader characteristics and affect the credibility of the leader’s

promises. Groups that lack any form of external support may be disciplined because

they form around tight bonds between the leader and group members. But groups

that receive high levels of external support are also likely to be disciplined. The

average effect of external support, therefore, is ambiguous.

This theory of commander incentives contrasts with existing theories. The the-

ory of member selection (Weinstein 2005, 2007) predicts that increased access to any

form of economic resources leads to increased civilian abuse as rebel groups attract

recruits who are motivated by financial gain. Theories of ethnic composition imply

that ethnically homogeneous groups will be more disciplined, but makes no predic-

tions regarding the effect of external support or resources. Theories of costly signaling

predict that external support will be associated with increased civilian abuse as rebels

demonstrate their commitment, and also that rebels in democratic states will commit

greater abuse. Table 4.6 compares the competing predictions for the signs of the coef-

ficients on the explanatory variables. My theory is supported against the alternatives

if I can demonstrate negative coefficients for LeaderCash, DiaSupport, CorpSupport,

AlignedPrefs, EthnicTies, and Commerce.

Few studies have systematically analyzed civilian abuse across rebel groups, so it

is worth beginning the analysis by exploring the data for general trends. Table 4.7

lists summary statistics for the various regions.5 It is clear that African rebel groups

dominate the dataset, comprising 42% of all rebel groups that controlled territory

between 1980 and 2003. Further, the African rebels are, on average, more abusive

than rebels in the other regions; African rebels also differ in their access to resources.

Consistent with the theoretical predictions, rebels in Africa have, on average, the most

valuable loot, the least amount of leader financing, and the lowest levels of affinity

between leader and commander.

Outside of Africa, the link between resources and rebel behavior is less clear,

5Regional statistics are derived by first computing the average value for each rebel group, and thenaveraging within a region. When computing means it is necessary to keep in mind that LeaderCashand Loot are coded on a logarithmic scale. For example, the average value of 4, 4, and 6 on thelogarithmic scale is 5.5, not 4.7 as would be implied by the standard calculation.


Commander Member Ethnic CostlyIncentives Selection Composition Signaling

LeaderCash - + 0 0Loot + + 0 0Affinity - - - 0NoReward + + 0 0Patron 0 + 0 +MultiPatron∗ - 0 0 0StateSupport + + 0 +DiaSupport - + 0 +CorpSupport - + 0 +AlignedPrefs - + 0 +EthnicTies - + 0 +Commerce - + 0 +EthnicRebellion - - - 0Democracy 0 0 0 +∗ Among groups receiving external support.

Table 4.6: Competing predictions of the effects of explanatory variables on civilianabuse.

Region Africa Americas Asia Mid. East Soc. Bloc AllNumber of Groups 31 4 16 7 15 73CivAbuse (Avg.) 0.47 0.16 0.21 0.31 0.45 0.38Factionalize (Avg.) 0.09 0.01 0.05 0.05 0.01 0.06LeaderCash (Avg.) 6.3 6.3 6.5 7.2 6.3 6.6Loot (Avg.) 5.7 5.5 5.5 5.4 4.8 5.5Affinity (Avg.) 0.50 0.90 0.81 0.57 0.53 0.60Separatist (%) 0.32 0.00 0.69 0.57 0.80 0.51Ethnic Rebellion (%) 0.54 0.25 0.56 0.43 0.87 0.58External Support (%) 0.84 0.75 0.81 0.86 0.53 0.77State Patron (%) 0.74 0.50 0.56 0.86 0.40 0.63Corporate Patron (%) 0.19 0.25 0.19 0.14 0.07 0.16Diaspora Patron (%) 0.13 0.00 0.19 0.14 0.13 0.14

Table 4.7: Regional descriptive statistics.


but still mostly consistent with the theory’s predictions. For example, rebels in the

Former Socialist Bloc are nearly as abusive as their African counterparts, and leaders

of these rebel groups have relatively little access to financing or affinity with their

commanders. On the other hand, these rebels have less access to valuable loot.

Rebel groups in the Americas committed the lowest levels of abuse on average. Their

access to leader financing and loot is similar to the Asian rebels, who committed

slightly more abuse, but the American rebels have higher affinity between leaders

and commanders. The Middle East is the only region that seems to challenge the

theory’s predictions. The rebel groups exhibit moderate levels of abusiveness, but

had access to large amounts of leader financing and not particularly valuable loot.

Their behavior could be explained, however, by relatively low levels of affinity.

The table does not reveal any clear relationships between rebel behavior and

separatist war aims or ethnic rebellion at the regional level. Nor does the table reveal

a clear link between external support and abusiveness. The majority of rebels in all

regions received external support, although the percentage is significantly lower in

the Former Socialist Bloc. For all regions, state patrons are the most common type

of external supporter of rebellion.

These regional statistics exhibit some changes over time. Figure 4.1 plots the level

of civilian abuse during the 1980-2003 period. The graph for all countries indicates

a general increase in rebel abusiveness that began in the early 1990s and persisted

through 2003. This abusiveness is consistent with media reports at the time that

proclaimed the emergence of “New Wars” involving rebel groups more akin to criminal

enterprises than ideologically-driven organizations (e.g. Kaplan 1994). This trend is

mirrored, to a greater or lesser degree, in the regional graphs. Abusiveness increased

dramatically during the 1990s in Africa, the Americas, and the Former Socialist Bloc,

and more modestly in Asia. The only region without a clear upward trend during this

period is the Middle East, where abusiveness declined in 1990 before rising briefly in

the mid-1990s before the temporary end of wars in the region.

It is important to keep in mind that these measures of abusiveness are coded

largely based on reports from news outlets and humanitarian organizations. As a re-

sult, the increased levels of abusiveness could be explained by greater media attention


to rebel atrocities beginning in the 1990s and persisting until today, possibly initiated

by the end of the Cold War or an increased demand in the West for conflict reporting.

These trends, however, could also be driven by substantive changes in rebel group

characteristics, such as a decline in their access to financing or increased access to

loot.

Figure 4.2 plots the value of leader financing (solid lines) and lootable resources

(dashed lines) over the same 1980-2003 period. These temporal graphs reveal a more

complicated relationship between resources and civilian abuse than suggested by the

summary statistics. Overall, the value of resources stayed fairly constant during the

period, but some regions experienced significant variation. In Africa, rebel groups

gained more access to lootable resources in the early 1990s, around the same time

as abusiveness increased.6 But this increased opportunity for looting was joined by

a similar increase in leader access to financing. Leader financing, however, dropped

off precipitously in the mid-1990s before recovering, to some extent towards the end

of the decade. The Middle East rebels lost their access to valuable loot in 1990 but

increased their access to leader financing. At the same time, these groups committed

less civilian abuse. Rebels in the Former Socialist Bloc lost much of their leader

financing during the mid-1990s, and exhibited a spike in abusiveness during the same

period. The trends in the Americas and Asia do not seem to correspond to the

theory’s predictions. Although abusiveness increased dramatically in the Americas

during the 1990s and modestly in Asia during the same period, rebels in both regions

had relatively constant access to resources.

These time graphs chart average rebel behavior and access to resources, which is

driven by two dynamics: (1) the entrance and exit of rebel groups from the dataset;

and (2) changes in behavior and access to resources among existing rebel groups.

These distinct processes raise the question of whether the main source of variation in

the data occurs between groups, or within them.

The relative importance of between-group versus within-group variation can be

assessed with an OLS model in which group dummy variables for the rebel groups are

6The increased value of lootable resources could be explained either by new groups emerging inresource-rich areas or by existing groups expanding their operations into resource-rich areas.


1980 1985 1990 1995 2000

0.0

0.4

0.8

All Countries

Year

Avg

. Civ

Abu

se

1980 1985 1990 1995 20000.0

0.4

0.8

Africa

Year

Avg

. Civ

Abu

se

1980 1985 1990 1995 2000

0.0

0.4

0.8

Americas

Year

Avg

. Civ

Abu

se

1980 1985 1990 1995 2000

0.0

0.4

0.8

Asia

Year

Avg

. Civ

Abu

se

1980 1985 1990 1995 2000

0.0

0.4

0.8

Middle East

Year

Avg

. Civ

Abu

se

1980 1985 1990 1995 2000

0.0

0.4

0.8

Former Socialist Bloc

Year

Avg

. Civ

Abu

se

Figure 4.1: Average civilian abuse over time, by region.


1980 1985 1990 1995 2000

45

67

8

All Countries

Year

Res

ourc

es (L

og)

1980 1985 1990 1995 2000

45

67

8

Africa

Year

Res

ourc

es (L

og)

1980 1985 1990 1995 2000

45

67

8

Americas

Year

Res

ourc

es (L

og)

1980 1985 1990 1995 2000

45

67

8Asia

Year

Res

ourc

es (L

og)

1980 1985 1990 1995 2000

45

67

8

Middle East

Year

Res

ourc

es (L

og)

1980 1985 1990 1995 2000

45

67

8


Year

Res

ourc

es (L

og)

Figure 4.2: Average value of leader financing (solid) and lootable resources (dashed)over time, by region.


the only explanatory variables. Indeed, when CivAbuse is regressed on group fixed

effects, an R2 value of 0.84 reveals that most variation is explained by differences

between groups rather than changes in behavior over time. Similarly, group fixed

effects explain 83% of the variation in LeaderCash and 90% of the variation in Affinity.

Fixed effects explain a smaller, but still substantial, percentage of the variation in

Loot (R2 = 0.70).

Changes in group behavior over time can be examined by regressing CivAbuse on

the interaction of ConflictYear and group dummy variables. This model estimates

the slope of civilian abuse within each group over time. Positive coefficients imply

that the group became more abusive over time, while negative coefficients imply less

abusiveness. Figure 4.3 plots the distribution of these slope coefficients for the rebel

groups. Most of the group coefficients are very close to 0, revealing little change in

behavior over time. Similar results are obtained for the groups’ access to resources.

The regional time trends are driven primarily by the formation of new rebel groups

or the resolution of conflicts, not changes in behavior among existing groups.

Changes over time in the prevalence and sources of external support are shown

in Figure 4.4. The graphs do not reveal any clear links with rebel behavior or ac-

cess to resources, although they do show some interesting variation. Overall, state

support for rebellion declined gradually from the late 1980s onwards, while support

from corporations and diaspora groups gradually rose during the mid-1990s. This is

consistent with the popular belief that state support for rebellion declined after the

end of the Cold War, although this trend is perhaps less dramatic than expected.

In the Former Socialist Bloc, state support was dominant during the early 1990s,

but was then largely replaced by support from diaspora communities. State patrons

were the only form of support for Middle Eastern rebels until the early 2000s, when

corporations and diaspora groups also became involved in supporting rebellion. The

other regions show fluctuations in the types of support, but no clear trends.


Slope Coefficient

Frequency

-0.10 -0.05 0.00 0.05

010

2030

40

Figure 4.3: Distribution of coefficients for within-group trends in civilian abuse overtime.


1980 1985 1990 1995 2000

0.0

0.4

0.8

All Countries

Year

% G

roup

s

1980 1985 1990 1995 2000

0.0

0.4

0.8

Africa

Year

% G

roup

s

1980 1985 1990 1995 2000

0.0

0.4

0.8

Americas

Year

% G

roup

s

1980 1985 1990 1995 2000

0.0

0.4

0.8

Asia

Year

% G

roup

s

1980 1985 1990 1995 2000

0.0

0.4

0.8

Middle East

Year

% G

roup

s

1980 1985 1990 1995 2000

0.0

0.4

0.8


Year

% G

roup

s

Figure 4.4: Percentage of groups receiving support from a foreign state (solid), cor-poration (dashed), and diaspora community (dotted) over time, by region.


4.2.1 Civilian Abuse and Commander Incentives

Table 4.8 lists the bivariate correlations among the key variables. Consistent with the

theory, and contrary to theories of member selection and costly signaling, LeaderCash

is negatively correlated to civilian abuse. Also consistent with the theory, Loot is pos-

itively correlated with abuse, while the commander’s expectation of future rewards

(measured by Affinity and Reward) is negatively associated with abusiveness. Con-

sistent with theories of ethnic composition, EthnicRebellion is associated with less

abusiveness. These bivariate relationships provide little indication of a link between

democracy and abusiveness, contrary to some theories of costly signaling (Stanton

2010, Pape 2003). The table indicates some correlations among the explanatory

variables, especially between Affinity and Reward, and between EthnicRebellion and

Separatist. To avoid problems associated with multicollinearity, these variables are

not included in the same regression models.

To analyze the data I use a basic OLS model with standard errors clustered by

rebel group and a variety of fixed effects. Fixed effects for each rebel group provide the

strongest inference, since they account for unobserved, time-invariant characteristics

of both the rebel groups and the governments they face. Unfortunately, the lack of

variation within groups on the key variables means that group fixed effects are likely

to “soak up” the effects of the explanatory variables. Consequently, I also present

models with fixed effects for region and country, in an effort to control for omitted

variables at these levels. To account for possible temporal trends I include year fixed

effects, although, in most cases, the inclusion of these fixed effects has no substantive

effect on the results.

The results of the regressions are presented in Table 4.9 and provide support for

the theory’s predictions. Models 1-3 present the theory’s explanatory variables on

the full dataset for each type of fixed effects. Models 4-6 use the alternative measure

of leader credibility, Reward, on the subset of rebel groups that received post-conflict

rewards. Model 7 presents the variables associated with competing theories as well

as control variables. These controls relate to government abusiveness (GovAbuse),

conflict intensity (BattleDeaths), conflict duration (ConflictYear), state wealth and

capacity (GDPpc), and country size (Pop). The effects of these control variables are


not analyzed in the existing literature, but each could change the rebel’s calculus re-

garding civilian abuse. For example, more abusive government forces could encourage

the rebels to respond in kind or, alternatively, could encourage discipline as a way

for the rebels to distinguish themselves from the abusive state. Larger populations

might make it easier for the rebels to commit abuse without consequences, but large

populations might also strengthen the rebels’ economic incentives to provide order

and collect taxes. Model 8 includes the theory’s variables as well as the alternative

explanations and all controls. Only regional fixed effects are used for these last two

models, as most of the variables are coded at the country level.

As predicted, the leader’s access to financial resources is associated with lower

levels of civilian abuse (Hypothesis 1). This relationship is large and significant

in most of the models, but disappears in Model 3 when group-level fixed effects are

included. Importantly, the effect of leader financing remains large and significant even

with the addition of competing explanations and the full menu of control variables.

An even stronger relationship is observed with the Loot variable. As predicted

(Hypothesis 2), the value of lootable resources in the rebels’ territory is significantly

related to higher levels of civilian abuse across all specifications, and this effect re-

mains substantial even with the inclusion of group fixed effects. The magnitude of

Loot, however, is reduced in Models 6 and 7, probably due to the relatively strong

correlation between Loot and Reward.

Leader-commander affinity is also associated with less abusiveness, as predicted

(Hypotheses 3 and 4), and the coefficient is large. In the full model (Model 8),

the coefficient for affinity is larger than any other coefficient except GovAbuse. Like

leader financing, however, the magnitude and significance of the Affinity coefficient

decreases dramatically with the inclusion of group-level fixed effects. The other mea-

sure of leader credibility, Reward, is less successful. When region-level fixed effects are

included (Model 4), Reward is strongly associated with less abusiveness, as predicted

by the theory. This effect, however, switches directions when country fixed effects are

included in Model 5 and disappears with the inclusion of group fixed effects.7 The

7The sign on the coefficient of Reward remains negative when time fixed effects are removed fromthe model but still loses significance with country and group fixed effects.


erratic estimates of Reward may be due to collinearity.

Theories of ethnic composition (Competing Hypothesis 4) also find support in the

regressions. EthnicRebellion has a large, negative coefficient in Model 7, revealing

that ethnically-motivated rebellions are generally less abusive towards civilians. This

relationship persists in Model 8, but the magnitude decreases by nearly half with the

addition of variables related to commander incentives, likely due to the strong correla-

tion between EthnicRebellion and Affinity. Theories of costly signaling that focus on

democracy (Competing Hypothesis 2) find mixed support. Democracy is associated

with more abusiveness in Model 7, as predicted by the theories, but the significance of

this relationship disappears with the inclusion of the commander incentive variables.

The theory of commander incentives does well against existing theories and the

control variables. Most importantly, the leader’s access to financing is persistently

related to lower levels of civilian abuse, which contradicts existing theories. Although

the level of government abuse is the single strongest predictor of rebel abusiveness,

the three variables proposed by the theory, collectively, have the largest explanatory

power. Indeed, the addition of the theory’s variables to Model 7 increases the R2

dramatically from 0.41 to 0.62.

These results, of course, are merely suggestive. The models can establish correla-

tion between group resources and behavior but do not allow for an identification of

causal effects. Fortunately, a focus on group resources is only one aspect of the theory.

Confidence in the theory is increased if I can show two additional relationships: (1)

the leader’s access to financing and her credibility are systematically related to the

motives of external patrons; and (2) different types of external patrons are associ-

ated with variation in rebel group behavior. These relationships are analyzed in the

following sections.

4.2.2 External Patrons and Leader Resources

A key causal mechanism in the theory is that external patrons shape rebel group be-

havior by deciding which rebel leaders to support and how much financing to provide


them (Hypothesis 10). Through selection pressure and monetary transfers, exter-

nal patrons shape the leader’s credibility and the incentives provided to high-level

commanders. In Table 4.10, the dependent variables are LeaderCash (Models 1-4),

Affinity (Models 5-7), and Reward (Models 8-10). The regressions include all rebel

groups that received external support. Since Affinity and Reward are binary vari-

ables, Models 5-10 are logit models. As before, all errors are clustered by rebel group.

Ideally a full battery of fixed effects would be used in these models, but the lack of

within group variation in Affinity, Reward, and other variables creates problems of

multicollinearity and over-determination when fixed effects are used. As a result, the

models may over-estimate the true relationships and the results should be treated

with caution.

The results for the LeaderCash regressions (Models 1-4) support the theory’s logic.

State supporters provide fewer cash resources to rebel leaders than corporations or

diaspora communities. This effect persists even when group fixed effects are used.

The theory assumes state patrons are often motivated by geo-political concerns and

thus have preferences that are not aligned with those of the rebel group. By contrast,

aligned preferences are associated with a large, positive increase in leader financ-

ing, shown in Model 3. Corporations and commercially-motivated sponsors provide

more financial resources to rebel leaders, as indicated by the positive coefficients on

CorpSupport and Commerce. Ethnically-motivated patrons also provide more cash

resources, indicated by the positive coefficient on EthnicTies. Multicollinearity pre-

vented DiaSupport from inclusion in Model 2.

The results for Affinity and Reward also support the theory’s predictions (Hy-

pothesis 10), although to a lesser extent. As expected, patrons are more likely to

support credible rebel leaders when their preferences are aligned with those of the

group, indicated by the positive coefficient on AlignedPrefs in Models 6 and 9. The

estimated effects are large: patrons with aligned preferences are associated with rebel

leaders who are up to 6 times more likely to share personal or ethnic ties with the

commander and up to 9 times more likely to share rewards. A large effect is also

observed among diaspora groups and ethnically-motivated patrons. Ethnic ties dra-

matically increase the probability of leader-commander affinity, but the relationship


is spurious since ethnically-motivated patrons are almost always associated with lead-

ers who share ethnic ties with commanders, which is coded as affinity in the dataset.

The persistent effect of ethnic motives on Reward, however, increases confidence in

the results: ethnically-motivated patrons are associated with leaders who are up to 9

times more likely to share post-conflict rewards with the commander.

Commercially-motivated patrons are associated with more cash for the rebel leader,

but their relationship with leader-commander affinity is ambiguous. Although corpo-

rations are associated with leaders who are up to 4.8 times more likely to share an

affinity with the commander, this relationship disappears when commercial incentives

or post-conflict rewards are considered. This is consistent with the theory’s conjec-

ture that corporations supply cash to rebel groups but have limited ability to affect

their composition or select the leader.

State patrons are associated with less cash for the rebel leader, but do not show a

clear relationship to leader-commander affinity or post-conflict rewards, contrary to

the theory’s predictions. This null relationship between state patronage and leader-

commander affinity could indicate some confounding factors that are not fully consid-

ered in the theory. First, ethnic ties may not be a good measure of leader credibility.

States might select less-trustworthy leaders from among a group of coethnics. Second,

the “market for rebellion” may be far from perfect. If the pool of aspiring rebel lead-

ers is limited, state patrons would have little ability to influence the characteristics

of the leaders they support.

Not supported in the data is the theory’s claim that groups that form in the

absence of external support will be characterized by higher quality leaders (Hypoth-

esis 11). Although a negative relationship between affinity and external support is

observed, the coefficient does not reach statistical significance (not shown).

Beyond these relationships, the regressions highlight the interesting effects of ideol-

ogy. While ideological patrons are not systematically related to either leader financing

or leader-commander affinity, they are strongly associated with a greater probabil-

ity of post-conflict rewards for the commander. Indeed, rebel leaders supported by

ideologically-motivated patrons are up to 22 times more likely to share rewards with

the commander. This result is likely an overestimate due to modeling issues, but


nonetheless suggests that ideology plays an integral role in overcoming the credible

commitment problem caused by the lack of externally enforced contracts. If leaders

and commanders share an ideology, the leader is more likely to make good on her

future promises.

4.2.3 External Patrons and Rebel Abusiveness

The final step in the analysis of civilian abuse is to show a systematic association be-

tween patron motives and the behavior of their rebel clients. Patrons with preferences

that conflict with those of the rebels support less-disciplined groups than patrons with

aligned preferences (Hypothesis 9). In Table 4.11, the dependent variable is once again

CivAbuse, but the main explanatory variables now relate to external patrons. All the

models use country fixed effects, although the results are substantively similar with

regional fixed effects. The lack of within-group variation on key variables precludes

the use of group fixed effects.

The models provide support for the theory compared to competing explanations.

The relationship between external support and rebel abusiveness is weakly positive,

but not statistically significant when country fixed effects (although the variable does

reach significance when only regional fixed effects are used). Among groups that

received external support, groups with multiple patrons committed relatively less

abuse, although this relationship is only significant with regional fixed effects (results

not shown).

State support for rebellion is positively correlated with civilian abuse by the

rebel clients, while support from diaspora communities is negatively correlated with

abuse. Corporate sponsors are not systematically associated with rebel behavior, but

commercially-motivated patrons are related to less abusive rebel groups. On average,

some degree of preference alignment between patrons and rebels is associated with

less abusiveness.

The negative coefficients on DiaSupport, AlignedPrefs, EthnicTies, and Commerce

support the theory’s predictions while contradicting existing theories of rebel behav-

ior. The analysis includes all rebel groups—even those that did not receive external


support—which increases the significance of the findings. Rebels that receive support

from patrons with aligned preferences are less abusive, on average, than the combined

pool of groups that received support from patrons with conflicting preferences and

groups that did not receive support at all.

4.2.4 Exogenous Proxies and Rebel Abusiveness

The explanatory variables do not allow for causal inference because they are coded

based on the observed behavior of rebel groups and external patrons. The variables

also suffer from significant measurement error due to the scarcity of information on

abusiveness and financial transactions during civil war. The proxy variables Capi-

talResources and LootResources overcome these problems, to some extent, for two

reasons. First, the variables are coded based on pre-war geographic endowments and

are thus exogenous to the rebel group. Second, the technology required to extract

each type of resource has implications for whether the money can be controlled by the

rebel leader, through partnerships with external actors, or looted by rebel comman-

ders and their soldiers. The theory predicts that CapitalResources will be negatively

correlated with civilian abuse, while LootResources will be positively correlated with

abusiveness.

Table 4.12 lists the results of a reduced-form OLS model with standard errors

clustered by country. Because the proxy variables are coded at the country level

and do not change over time, the regressions use average civilian abuse for each

rebel group as the dependent variable, yielding 73 total observations. The variables

are rough, but still correspond to the theory’s predictions (Hypotheses 1 and 2).

CapitalResources, which proxies for leader financing, is negatively related to civilian

abuse, although the coefficient becomes less significant (p = 0.16) when regional

fixed effects are introduced. Nevertheless, the negative relationship between capital-

intensive natural resources and civilian abuse challenges existing theories of rebel

behavior. LootResources is positively correlated with abuse, as expected, although

the coefficient only reaches significance in the second model. A more refined measure

of these variables may strengthen the results, but the exogenous variables nonetheless


add confidence to the previous results.8

4.3 Factionalization

I argue that rebel factionalization results from weak control in the organization, and

can be explained by the same incentives and external relations that lead to civilian

abuse. The leader’s access to financial resources and her credibility bind commanders

closer to the leader and reduce the risk of factionalization (Hypotheses 1, 3, and 4).

Lootable resources heighten the risk of factionalization by increasing the opportunity

costs of following the leader’s orders and providing independent access to revenues

for group commanders (Hypothesis 2). Peace talks heighten tensions in the group

because leaders must allocate post-conflict rewards based on imperfect information

to commanders who might be disappointed by their payout (Hypothesis 8). As be-

fore, the incentives in the rebel group are shaped by the motives of external patrons

(Hypothesis 9).

Understanding rebel factionalization has significant implications for civilian wel-

fare and international intervention. It is useful once again to begin by examining

regional and temporal trends. Factionalization is observed 43 times in the dataset

and affects 33% of the 73 groups. Figure 4.5 plots the average risk of factional-

ization by year for the various regions. Overall, the risk of factionalization seems

to increase during the mid-1990s, the same time period that saw a rise in civilian

abuse. This trend, however, is concentrated in certain regions. African rebels expe-

rienced dramatic increases in factionalization during the 1990s, after experiencing no

factionalization in the previous decade. Asian rebels were also more prone to faction-

alize during the 1990s, but rebels in the Americas and Middle East only experienced

factionalization in the 1980s. Strikingly, rebels in the Former Soviet Bloc did not

factionalize during 1980-2003.

To analyze the potential causes of group factionalization I use a Cox proportional

8These proxies can also be used as instruments in a two-stage least squares regression model.The signs of the coefficients are consistent with the theory in a two-stage model with a variety ofcontrols, but fail to reach significance.

4.3. FACTIONALIZATION 127

1980 1985 1990 1995 2000

0.0

0.2

0.4

All Countries

Year

1980 1985 1990 1995 2000

0.0

0.2

0.4

Africa

Year

1980 1985 1990 1995 2000

0.0

0.2

0.4

Americas

Year

1980 1985 1990 1995 2000

0.0

0.2

0.4Asia

Year

1980 1985 1990 1995 2000

0.0

0.2

0.4

Middle East

Year

1980 1985 1990 1995 2000

0.0

0.2

0.4


Year

Figure 4.5: Risk of factionalization over time, by region.


hazard model,

λi(t) = λ0(t)exi(t)β

where λi(t) is an unspecified, non-negative function of time (Therneau and Gramb-

sch 2000). Standard errors are clustered by rebel group (Lin and Wei 1989). The

proportionality assumption is tested for all models and, when necessary, interaction

terms are added to account for time-varying effects. I begin the analysis with a look

at commander incentives.

4.3.1 Factionalization and Commander Incentives

Table 4.13 displays the effects of group resources and peace talks on the risk of group

factionalization. The results support the theory’s predictions. Lootable resources

significantly increase the risk of factionalization across all specifications (Hypothesis

2). A unit increase in Loot, equivalent to a ten-fold increase in the value of lootable

resources, corresponds to 77-161% increase in the risk of group factionalization. The

variable Loot:Time is an interaction between the effects of lootable resources and time

and accounts for time-varying effects. This term is significant in Models 3 and 6 and

reveals a possible decreasing influence of lootable resources on group cohesion over

time. According to Model 3, lootable resources increase the risk of factionalization

by 161% during the group’s first year, but increase the risk of factionalization by only

43% in the fifth year.

The effects of leader financing also support the theory (Hypothesis 1), although the

variable loses significance when Reward is added to the model. A ten-fold increase in

leader finances reduces the risk of factionalization by 26-32%, and this effect remains

constant over time. Affinity also corresponds to the theory’s prediction (Hypotheses

3 and 4), decreasing the risk of factionalization by a constant 45-47% throughout the

group’s lifespan. Reward does not perform as well; the models do not show any sig-

nificant relationship between commander rewards and the risk of factionalization, but

this null result is probably due to the fact that Reward is coded as 0 for commanders

who break off from the group, biasing the results towards zero.

The models reveal an interesting relationship between peace talks and the risk of

4.3. FACTIONALIZATION 129

factionalization. According to Model 5, peace talks initially have no effect on group

cohesion, but contribute to a significantly greater risk of factionalization when they

occur later in a group’s lifespan. After the group has survived a year, peace talks

increase the risk of factionalization by 16%; after 5 years, the risk of factionalization

is 111% higher than baseline. This dynamic makes sense if peace talks are interpreted

as a bargaining process for distributing post-conflict rewards. It is easy to imagine

how the initial expectations of future rewards soon turn to frustration for rebel com-

manders as the war drags on. The commander may grow tired of the leader’s promises

and try to earn his own seat at the bargaining table by forming a new faction. Or the

leader, consumed by the negotiations and politics of the war, may become increas-

ingly reluctant to give up the best positions to his lieutenants in the field. Although

scholars have commented on the commitment problems that exist between armed

groups during peace processes, this model sheds light on the commitment problems

that exist within the groups during the negotiations. Peace talks can bring danger

and uncertainty for the rebel participants.

4.3.2 Factionalization and External Patrons

The theory predicts that the motives of external patrons will influence a group’s risk

of factionalization (Hypothesis 9). Patrons with social or commercial interests that

align with those of the rebels will make efforts to strengthen the leader’s control in

the group, thereby increasing the leader’s ability to prevent factionalization. Patrons

with preferences that do not align with those of the group will keep the leader weak,

contributing to a greater risk of factionalization.

Table 4.14 depicts the effects of patron identity and motives on the risk of group

factionalization. Some of the variables for patron motives had to be excluded to

prevent singularity. With the exception of Commerce, the results support the the-

ory’s predictions. Rebel groups that receive support from foreign governments face

a significantly higher risk of factionalization—up to 177% higher than the baseline

risk. This effect, however, quickly decreases over time, as indicated by the significant

coefficient on the interaction term, StateSupport:Time. Indeed, the increased risk of


factionalization associated with state support disappears entirely after 5-6 years.

This diminishing risk can be understood in the context of the theory. A state

patron might benefit from a rebel leader who lacks a strong base of support within

the rebel group. Early on, the rebel leader is dependent on the state patron, raising

suspicions among the commanders who may be inclined to break from the group. But

over time, the leader forms personal connections to the rebel commanders and can

establish some level of trust through repeated interactions. If the group survives the

first few years as a cohesive unit, the leader’s relationship with the state patron is no

longer a destabilizing influence on the group.

The effects of socially-motivated patrons also correspond to the theory’s predic-

tions. Rebel groups that rely on diaspora communities face a significantly lower risk

of factionalization, but this effect dissipates over time. In the first year of rebel-

lion, diaspora-supported groups are 98% less likely to factionalize than the baseline.

This effect persists for more than a decade in the group’s lifespan. After 10 years

of fighting, diaspora-supported groups are still about 70% less likely to factionalize

than the baseline group. The coefficients for EthnicTies show similar effects. When

rebel groups depend on patrons that are motivated by ethnic concerns, they are 97%

less likely to factionalize than the baseline in the first year and 34-55% less likely to

factionalize after 10 years. This effect, however, does not result from greater access

to leader financing, since the coefficients related to ethnic support do not substan-

tively change when controlling for LeaderCash. Rather, ethnic homogeneity within

the group and support from co-ethnics outside the conflict area may increase trust

between the leader and commander and improve monitoring and coordination among

group members.

The results for commercial interests do not support the theory’s predictions. Al-

though the coefficients for CorpSupport do not reach significance, the coefficients for

Commerce reveal a significant increase in the risk of factionalization when rebel groups

receive support from commercially-motivated patrons. Rebels that depend on these

patrons are 90-92% more likely to factionalize than the baseline. The reasons for this

relationship are not clear, especially since commercially-motivated patrons are linked

to greater leader financing, which in turn is linked to a lower risk of factionalization.

4.4. DISCUSSION 131

One plausible explanation comes to mind. The null result for CorpSupport indicates

that the increased risk of factionalization is not due to corporate sponsorship of re-

bellion, but rather to non-corporate economic actors in the conflict zone. Whereas

corporations have incentives to establish long-term relationships with rebel leaders,

non-corporate actors may be more inclined to strike short-term deals with any willing

partner. As a result, these patrons provide economic opportunities that encourage

rebel commanders to split from the group and seek their own profits.

4.4 Discussion

The theory of commander incentives finds consistent support across a variety of model

specifications, explaining both civilian abuse and the risk of factionalization. The

results are consistent with the theory’s central hypotheses regarding the effects of

resources and leader credibility (Hypotheses 1-4), the destabilizing effects of peace

agreements (Hypothesis 8), and the influence of external patrons (Hypotheses 9 and

10). When leaders have access to financing and their preferences are aligned with

external patrons, rebel groups are less abusive and face a lower risk of group faction-

alization. Lootable resources are associated with more abusive, faction-prone groups,

while peace talks increase the risk of factionalization over time.

The results do not support the claims that groups that form in the absence of

external support are characterized by higher quality leaders (Hypothesis 11) or that

multiple patrons are associated with more disciplined groups (Hypothesis 12). The

remaining hypotheses (Hypotheses 5-7) cannot be tested in a cross-national setting

and will be examined in the context of Liberia.

These findings pose a challenge to theories of costly signaling and member selec-

tion. Both theories are weakened by the lack of a strong relationship between civilian

abuse and external support (Competing Hypotheses 3 and 6). Some theories of costly

signaling are challenged by the lack of a robust relationship between abusiveness and

democracy (Competing Hypothesis 2). The theory of member selection is further chal-

lenged by the robust negative relationship between abusiveness and leader financing


and between abusiveness and support from patrons with aligned preferences (Com-

peting Hypothesis 6). Theories of ethnic composition do better, as the results show

a robust negative relationship between civilian abuse and ethnically-motivated rebels

(Competing Hypothesis 4). This relationship persists even after including variables

related to commander incentives, which indicate that ethnicity plays a larger role

in rebellion than simply contributing to leader credibility, as implied by the theory

proposed here.

The results may be challenged by an omitted variable that simultaneously deter-

mines group discipline, leader credibility, and leader access to financing. The use of

exogenous proxies for leader financing and lootable resources alleviates this concern

to some extent, but does not add much insight into how the theory’s causal mecha-

nisms work in reality. To trace the causal impact of resources, leader credibility, and

external patrons, I now turn to a detailed analysis of Liberia’s rebel organizations.

4.4. DISCUSSION 133

Civ

Abuse

Lea

der

Cas

h

Loot

Affi

nit

y

Rew

ard

Eth

nic

Reb

ellion

Sep

arat

ist

CivAbuseLeaderCash -0.28

Loot 0.19 0.47Affinity -0.50 0.17 -0.08Reward -0.57 0.25 -0.40 0.70

EthnicRebellion -0.38 -0.03 -0.19 0.51 0.48Separatist -0.43 0.02 -0.11 0.41 0.35 0.81

Democracy -0.04 0.15 0.10 0.22 0.18 -0.20 -0.19Note: Reward only applies to groups that received post-conflict rewards.

Table 4.8: Correlations among key variables.

134 CHAPTER 4. REBEL ABUSIVENESS AND FACTIONALIZATIONM

odel1

Model

2M

odel3

Model

4M

odel5

Model

6M

odel7

Model

8L

eaderCash

−0.08

∗∗∗−

0.08∗∗∗

−0.02

−0.18∗∗∗

−0.09

∗∗−

0.04∗∗∗

−0.11∗∗∗

(0.02)(0.02)

(0.03)(0.04)

(0.03)

(0.01)(0.02)

Loot

0.13∗∗∗

0.17∗∗∗

0.16∗∗∗

0.17∗∗∗

0.10∗∗

0.07∗∗∗

0.20∗∗∗

(0.04)(0.05)

(0.05)(0.06)

(0.04)

(0.02)(0.05)

Affi

nity−

0.37∗∗∗

−0.32∗∗∗

−0.11

−0.28∗∗∗

(0.08)(0.11)

(0.08)(0.06)

Rew

ard−

0.28∗∗

0.12∗

0.03

(0.12)(0.07)

(0.06)E

thnicRebellion

−0.48∗∗∗

−0.25∗∗∗

(0.11)(0.07)

Dem

ocracy0.21∗∗

0.09

(0.09)(0.08)

GovA

buse0.32∗∗∗

0.42∗∗∗

(0.11)(0.06)

BattleD

eaths/1,000−

0.04

−0.05

(0.06)(0.04)

ConflictY

ear0.01

0.01(0.01)

(0.00)G

DP

pc(L

og)0.10

0.16∗

(0.13)(0.09)

Pop

(Log)

0.02

0.06(0.07)

(0.05)(Intercept)

0.38∗∗

0.34

−0.62∗∗

0.500.61∗∗∗

0.11

−0.85

−1.66∗∗∗

(0.18)(0.28)

(0.28)(0.32)

(0.17)

(0.10)(0.73)

(0.50)U

nitF

ER

egionC

ountryG

roupR

egionC

ountryG

roupR

egionR

egionY

earF

EY

esY

esY

esY

esY

esY

esY

esY

esN

768768

768336

336336

440440

R2

0.530.75

0.890.63

0.940.96

0.480.67

adj.R

20.48

0.720.87

0.560.92

0.950.41

0.62R

esid.sd

0.270.20

0.140.28

0.120.09

0.300.24

Clustered

errorsin

parentheses∗∗∗

p<

0.01;∗∗p<

0.05;∗p<

0.10

Tab

le4.9:

OL

Sm

odels

ofciv

ilianab

use

with

variables

forcom

man

der

incen

tives.

4.4. DISCUSSION 135

Lea

der

Ca

shA

ffin

ity

Rew

ard

Mod

el1

Mod

el2

Mod

el3

Mod

el4

Mod

el5

Model

6M

od

el7

Mod

el8

Mod

el9

Mod

el10

Sta

teS

up

port

−0.5

2∗∗

−0.4

7∗∗

−0.2

9−

0.2

4(0.2

4)

(0.2

1)

(0.7

3)

(1.0

2)

Corp

Su

pp

ort

1.0

3∗∗

∗0.4

81.5

7∗

0.7

5(0.3

4)

(0.3

8)

(0.9

0)

(0.8

9)

Dia

Su

pp

ort

0.6

3∗

4.2

3∗∗

∗2.0

9∗∗

(0.3

6)

(1.1

9)

(1.0

3)

Align

edP

refs

1.1

7∗∗

∗1.8

9∗∗

2.2

1∗∗

(0.4

0)

(0.7

6)

(0.9

0)

Ideo

logy

0.4

00.5

43.1

3∗∗

(0.3

9)

(1.0

6)

(1.4

0)

Eth

nic

Tie

s0.7

9∗∗

3.1

2∗∗

∗2.2

6∗∗

∗

(0.3

2)

(0.8

6)

(0.8

1)

Com

mer

ce0.7

8∗∗

∗0.7

4−

0.7

6(0.3

0)

(0.7

6)

(0.8

0)

Rel

igio

n1.4

8∗∗

∗

(0.3

7)

Hu

man

itari

an

0.7

6∗∗

(0.3

0)

(Inte

rcep

t)5.9

3∗∗

∗5.0

5∗∗

∗3.3

9∗∗

∗4.8

5∗∗

∗0.6

3−

0.4

3−

0.0

14

0.2

0−

1.2

3−

0.1

3(0.5

1)

(0.2

0)

(0.3

9)

(0.4

9)

(0.7

7)

(0.5

4)

(0.4

8)

(0.9

3)

(0.7

7)

(0.5

9)

Un

itF

ER

egio

nG

rou

pG

rou

pR

egio

nN

on

eN

on

eN

one

Non

eN

on

eN

on

eY

ear

FE

Yes

Yes

Yes

Yes

No

No

No

No

No

No

N686

686

686

686

686

686

686

313

313

313

R2

0.4

40.8

80.8

90.4

3ad

j.R

20.3

80.8

60.8

70.3

7R

esid

.sd

0.9

20.4

40.4

20.9

2A

IC741.5

6750.2

4720.7

6400.8

0361.1

7351.7

5B

IC814.0

6786.4

9793.2

5460.7

4391.1

4411.6

5lo

gL

−354.7

8−

367.1

2−

344.3

8−

184.4

0−

172.5

9−

159.8

5

Clu

ster

eder

rors

inpa

rent

hese

s∗∗∗p<

0.01

;∗∗p<

0.05

;∗p<

0.10

Tab

le4.

10:

Model

sof

lead

erre

sourc

esam

ong

reb

els

rece

ivin

gex

tern

alsu

pp

ort,

wit

hva

riab

les

for

pat

ron

iden

tity

and

mot

ives

.


Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7Patron 0.12

(0.11)StateSupport 0.11∗ 0.12∗∗

(0.06) (0.06)CorpSupport 0.05 0.17

(0.13) (0.12)DiaSupport −0.41∗∗∗ −0.26∗∗

(0.10) (0.10)AlignedPrefs −0.17∗∗ −0.14∗

(0.08) (0.08)Ideology −0.01 −0.05

(0.08) (0.09)EthnicTies −0.18∗∗ −0.09

(0.10) (0.07)Commerce −0.15∗∗ −0.21∗∗

(0.08) (0.10)Religion 0.01 0.18

(0.12) (0.13)Humanitarian −0.15 −0.10

(0.10) (0.10)LeaderCash −0.10∗∗∗ −0.06∗∗ −0.05

(0.03) (0.03) (0.04)Loot 0.16∗∗∗ 0.17∗∗∗ 0.21∗∗∗

(0.04) (0.05) (0.06)Affinity −0.30∗∗∗ −0.29∗∗∗ −0.24∗∗

(0.09) (0.10) (0.11)(Intercept) 0.64∗∗∗ 0.57∗∗∗ 0.95∗∗∗ 0.74∗∗∗ 0.26 0.37 0.08

(0.15) (0.14) (0.14) (0.12) (0.20) (0.27) (0.29)Unit FE Country Country Country Country Country Country CountryYear FE Yes Yes Yes Yes Yes Yes YesN 768 768 768 768 768 768 768R2 0.67 0.70 0.69 0.70 0.78 0.77 0.78adj. R2 0.62 0.66 0.64 0.65 0.74 0.73 0.74Resid. sd 0.23 0.22 0.22 0.22 0.19 0.19 0.19

Clustered errors in parentheses∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.10

Table 4.11: Models of civilian abuse with variables for patron identity and motives.

4.4. DISCUSSION 137

Model 1 Model 2CapitalResources −0.19∗ −0.14

(0.11) (0.10)LootResources 0.17 0.28∗∗∗

(0.14) (0.11)(Intercept) 0.43∗∗∗ 0.39∗∗

(0.08) (0.17)Fixed Effects None RegionN 73 73R2 0.06 0.17adj. R2 0.04 0.10Resid. sd 0.38 0.37Clustered errors in parentheses.∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.10

Table 4.12: Reduced-form model of civilian abuse with exogenous proxies for leaderfinancing and lootable resources.


Model

1M

odel2

Model

3M

odel4

Model

5M

odel6

LeaderC

ash−

0.38∗∗

−0.30

∗−

0.48−

0.32∗

−0.44

(0.17)(0.18)

(0.33)

(0.18)

(0.32)

Loot

0.76∗∗∗

0.65∗∗

1.11∗∗∗

0.59∗∗

1.05∗∗∗

(0.28)(0.26)

(0.30)

(0.26)

(0.31)

Loot:T

ime

−0.03

−0.02

−0.15

∗∗−

0.02−

0.15∗∗

(0.03)(0.03)

(0.06)

(0.03)

(0.07)

Affi

nity−

0.60∗

−0.63

∗

(0.35)(0.35)

Rew

ard0.24

0.15

(0.74)(0.77)

PeaceT

alks0.21

−0.15

0.31

(0.40)(0.45)

(0.56)

PeaceT

alks:Tim

e0.12∗∗

0.15∗∗

0.10(0.05)

(0.06)

(0.11)

N768

768336

768768

336W

ald12

on3

df14

on4

df30

on4

df11

on2

df19

on6

df37

on6

dfp

=0.01

p=

0.01p

=0.00

p=

0.00p

=0.00

p=

0.00R

20.02

0.02

0.03

0.01

0.03

0.03

(Max

0.33)(M

ax0.33)

(Max

0.34)(M

ax0.33)

(Max

0.33)(M

ax0.34)

Clustered

errorsin

parentheses∗∗∗

p<

0.01;∗∗p<

0.05;∗p<

0.10

Tab

le4.13:

Cox

prop

ortional

hazard

model

ofgrou

pfaction

alizationw

ithvariab

lesfor

comm

ander

incen

tives.

4.4. DISCUSSION 139

Model 1 Model 2 Model 3 Model 4StateSupport 1.20∗∗∗ 1.06∗∗

(0.44) (0.48)StateSupport:Time −0.18∗∗∗ −0.18∗∗

(0.06) (0.07)CorpSupport 0.31 0.71

(0.42) (0.55)DiaSupport −4.51∗∗∗ −4.32∗∗∗

(1.46) (1.45)DiaSupport:Time 0.30∗∗∗ 0.32∗∗∗

(0.09) (0.09)LeaderCash −0.47∗∗ −0.45∗∗

(0.23) (0.23)Loot 0.64∗∗ 0.60∗∗

(0.26) (0.24)Loot:Time −0.04 −0.05∗

(0.04) (0.03)Commerce 0.64∗∗ 0.65∗∗

(0.28) (0.33)EthnicTies −3.69∗∗∗ −3.72∗∗∗

(1.01) (1.06)EthnicTies:Time 0.29∗∗∗ 0.33∗∗∗

(0.11) (0.09)N 768 768 768 768Wald 21 on 5 df 51 on 8 df 23 on 3 df 38 on 6 df

p = 0.00 p = 0.00 p = 0.00 p = 0.00R2 0.03 0.04 0.03 0.04

(Max 0.33) (Max 0.33) (Max 0.33) (Max 0.33)Clustered errors in parentheses∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.10

Table 4.14: Cox proportional hazard models of group factionalization with variablesfor patron identity and motives.

Part II

Liberia’s Civil War

140

141

Don’t play with my pepper bush.

NPFL leader Charles Taylor warning his troops against

harassing civilians (C26).

Rebels need to eat.

Market seller, Kolahun (B17).

142

Monrovia

Kola (Kola Town)

Kenema

Buedu

Sulima

Nzérékoré

Guiglo

Taï

Guéckédou

Danané

Tabou

KailahunPendem

bu

Toulépleu

Irié

Bendaja

Mendekom

a

Kodeke

Belle Yella

Brewerville

Grand C

ess

Nana Kru

Nyaake

Sagleipie

Sasstown

Sehnkwehn

Tapeta

Timbo

Palala

Bong Town

Suakoko

Tawake

Pelokehn

Galio

Tobli

Klay

Kongo

Yekepa

Yela

Gahnpa

(Ganta)

Careysburg

EdinaTrade Tow

n

Kolahun

Marshall

Harbel

Belefuanai

ZigidaVahun

Gelahun

Yella

Zorzor

Butlo

Plibo

Drubo

(Dubw

e)

Kahnple

KpeaplePoabliTow

abli(Tow

ai Town)

Zienzu

Hartford

Guata

Gboyi

Nem

eke

Bo

Yibuke(Kaobli)

Tatuke

Wiesua

Kpein

Babu

Duabo

Tiehnpo

YakakahnShabli

Kopo

Zekera

Gbange

Debli

Ghapo

Bokoa

Gonglee

Juazohn

Tubm

anburg

Kakata

Robertsport

Buchanan

Harper

River C

ess

Tchien

(Zw

edru)

Voinjam

a

Gbarnga

Greenville

Barclayville

Saniquellie

Fish T

own

Bopolu

C. P

almas

Wologizi Range

Mt.

Nuon-F

aW

onegizi Range

Nimba Range

Mt.

Wuteve

Putu Range

Mt. G

letohn

LakeP

iso

Sassandra

St. John

Nuon

Mani

Lawa

Loffa

Makona

Moa

Magowi

St. Paul

Gbeya

Mano

ViaNianda

Morro

Gwen Creek

Cava

lly

Cavally

Cavalla

Cess (Cestos)

Sino

Dugbe

Dubo

Sehnkwe hn

Grand Cess

AT

LA

NT

ICO

CE

AN

SinoeB

ay

Loffa

TimboD

ube (Duobe)

BO

MI

MA

RG

IBI

LO

FA

GB

AR

POL

U

RIV

ER

GE

ES

INO

E

GR

AN

DC

AP

EM

OU

NT

BO

NG

RIV

ER

CE

SS

NIM

BA

MO

NT

SE

RR

AD

O

GR

AN

DB

ASSA

GR

AN

D

KR

UM

ARYLAND

GR

AN

D G

ED

EH

CÔ

TE

D'IV

OIR

E

GU

INE

A

SIE

RR

A L

EO

NE

Map N

o. 3775 Rev. 6 U

NIT

ED

NA

TIO

NS

January 2004D

epartment of P

eacekeeping Operatons

Cartographic S

ection

LIB

ER

IA

The boundaries and nam

es shown and the designations used

on this map do not im

ply official endorsement or acceptance

by the United N

ations.

LIB

ER

IAInternational boundaryC

ounty boundary

Railroad

National capital

County capital

Airport

Town, village

Road

0 80 km

20

010

50 mi

4060

2030

40

12°11°

10°9°

8°7°

8°7°6°5°

8°7°6°5°

12°11°

10°9°

8°7°

Chapter 5

An Overview of Liberia

This chapter provides an overview of Liberia’s political development and a description

of the major events and actors in the civil war. Although often portrayed as a savage

war of either tribal animosity or simple greed, Liberia’s war was rooted in legitimate

social and political grievances. Rebellion was initially seen as a necessary antidote to

a corrupt, entrenched regime and a legacy of oppression. The war, however, quickly

took on a life of its own as a multitude of armed groups competed for influence,

forcing civilians to either choose sides or flee for safety. A timeline of Liberia’s war

can be found in the chapter’s appendix.

5.1 A Settler Republic, 1822-1980

In early 1822 a ship containing a few dozen settlers and a handful of officials from

the American Colonization Society (ACS) landed on a small island in the Mesurado

river, just inland from what was then known as the Grain Coast of West Africa.

These settlers—mostly freed slaves from the United States—had reached the end of

an arduous two-year journey that claimed the lives of more than a dozen of their

fellow travelers. After tense negotiations with the native peoples, the ACS officials

secured some land in exchange for a few boxes of clothing and metal goods. Liberia

was born.

The official goals of the American Colonization Society included humanitarian,

143

144 CHAPTER 5. AN OVERVIEW OF LIBERIA

evangelical, and political concerns.1 Emphasizing the humanitarian mission, Rev-

erend Robert Finley, the driving force behind the ACS in America, stated “Our fa-

thers brought them here, and we are bound, if possible, to repair the injuries inflicted

by our fathers” (in Pham 2004: 6). American settlers in Africa, the ACS argued,

would also “civilize” and “Christianize” the native population, and possibly open up

new markets for American businesses.

But these concerns did not obscure the primary goal of the organization: removing

the growing population of free blacks that posed a political and social threat to the

United States, a country that still practiced slavery. As the Times of London reported,

“The chief recommendation of the project consists, of course, not in what it will bring,

but in what it will carry off. It will aid the United States in getting rid of their free

negroes” (Times of London 1850). Not surprisingly, the majority of free blacks and

abolitionists viewed the ACS’s plan as a means of reinforcing the institution of slavery

(Sawyer 1992: 35).

Near the mouth of the Mesurado river, the settlers built rudimentary houses and

named their town Monrovia, in honor of U.S. president James Monroe. Their numbers

gradually increased as more ships arrived from the United States, and as they were

joined by Africans rescued from intercepted slave ships. Over the next two decades,

settlers established new towns along the coast, most notably at Cape Palmas (now

Harper) and Bassa Cove (now Buchanan). Eventually some 19,000 Americans were

transported to Liberia (Dunn and Holsoe 1985: 17). The living conditions in these

fledging societies remained dire, with rampant disease and frequent battles with the

local people (Levitt 2005). Some 50% of the early settlers died of tropical diseases; a

few settlers even voluntarily returned to slavery in exchange for passage back to the

United States (Shick 1971; Pham 2004: 12-13).

The native population of Liberia was relatively sparse when the settlers arrived.

Heavy rainfall, poor soils, dense forest, and treacherous coastal waters limited human

settlement. Nevertheless, a culturally-diverse mix of 16 ethnic groups inhabited the

region. Most of the local population practiced traditional hunting, farming, or fishing,

1Amos Sawyer (1992) provides a detailed discussion of the various motives for colonization amongthe ACS and other organizations, as well as the social context surrounding free blacks in earlyAmerica.

5.1. A SETTLER REPUBLIC, 1822-1980 145

while a smaller subset specialized in trade. The Vai ethnic group along the northern

coast traded with Europeans, exchanging valuables such as gold, spices, and slaves

for salt and European merchandise. The Kru people focused on the coastal trade,

traveling by canoe to exchange kola nut and other goods, while the long-distance

inland trade was dominated by ethnic Mandingoes, whose Muslim social networks

connected them to markets throughout West Africa (Konneh 1996).

European colonial expansion in the mid-19th century created pressure on the

United States to define its relationship with its citizen-settlers in Africa. Specifi-

cally, Great Britain did not recognize the right of the ACS to collect customs duties

on ships traveling through Liberian waters. Faced with no source of revenue and an

indecisive United States, the settlers declared their status as an independent country

on 26 July 1847. The settlers adopted a constitution modeled on that of the United

States and drafted at Harvard University. The government included a bicameral leg-

islature, an independent judiciary, and an executive branch. At a time when few

people in the world could vote, the settlers cast their ballots for their first elected

president: Joseph Jenkins Roberts, a native of Norfolk, Virginia.

Upon taking office, Roberts embarked on a whirlwind tour to gain recognition

for the infant republic. Great Britain obliged, and dispatched a naval ship to give a

21-gun salute to the new flag in Monrovia’s harbor; other European nations quickly

followed suit (Times of London 1892). The United States, however, refused to rec-

ognize the new republic that it helped create. The reasons were clear: the United

States could not tolerate a black ambassador. As Willard Salisbury, Sr., a Democratic

member of the House of Representatives, put it: “How fine it will look...to welcome

here at the White House an African, full-blooded, all guilded and belaced... If that

is agreeable to the taste and feelings of the people of this country, it is not to mine!”

(in Pham 2004: 22). Not until 1862, amidst the turmoil of the Civil War, did the

United States recognize Liberia’s independence.

Despite the disappointing response from the United States, the Americo-Liberians,

as they became known, adopted many of the customs of their homeland. They at-

tempted to recreate America in Liberia, building Baptist churches, wearing coat-tails

and top hats to formal events, and constructing plantation-style homes. Economic


and living conditions improved substantially during this period, as the republic ex-

ercised its new power to tax trade and invested in infrastructure (McPherson [1891]

2009). Fraternal orders such as the Freemasons became popular among the Liberian

elite and the lively sounds of piano music filled the streets from downtown cafes.

Although progressive for their time, the Americo-Liberians did not extend their

new-found liberty to Liberia’s native people. Even the government’s motto, “The

Love of Liberty Brought Us Here,” hints at the second-class status of the indigenous

population. The locals were not granted citizenship and were largely ignored by

the settler population. This “benign neglect” ended in the late 19th century when

Britain and France, as neighboring powers, required that the Liberian government

secure its territory as a condition of sovereignty. In response, the Americo-Liberian

regime constructed a system of indirect rule, whereby local chiefs collected taxes and

provided labor for public works projects. These chiefs were supported by a nascent

army, the Liberian Frontier Force (ICG 2009: 2).

Indirect rule quickly descended into brutality, as the native population endured

forced labor and onerous, arbitrary taxation in exchange for no government services.

Foreign travelers such as Graham Greene (1936) commented on the extreme cruelty

of the system.2 At least five large-scale revolts erupted in the first 50 years of in-

dependence and these were crushed with brutal force by the Liberian Frontier Force

and other settler militias (Levitt 2005). In 1931 the League of Nations reprimanded

Liberia for its use of forced labor and slavery (Times of London 1931). The report

detailed widespread abuses, most notably the practice of rounding up young men

by force and sending them to the brutal plantations of Fernando Po (now part of

Equatorial Guinea).

The Americo-Liberian regime continued to expand its presence in the hinterlands

and integrate the territory into the central government. A tumultuous multiparty sys-

tem eventually consolidated under the True Whig Party and Liberia began attracting

foreign investment. In 1926 Firestone secured a 99-year lease on 1 million acres to

form the world’s largest rubber plantation. The company employed 20,000 Liberians

2For a sympathetic description of the Liberian government’s policy of indirect rule, see R.C.F.Maugham ([1920] 1969).

5.2. THE MILITARY YEARS, 1980-1989 147

and constructed new roads, port facilities, and railway lines. In 1942 Liberia granted

the United States permission to build military bases in the country (FAS 1985).

Economic and political progress accelerated in 1944 with the election of President

William V.S. Tubman. Tubman implemented an “Open Door Policy” that attracted

foreign investment and allowed Liberia to begin exploiting its abundant natural re-

sources. Mining companies extracted iron ore from various deposits and a multitude

of Lebanese-owned shops appeared on Monrovia’s streets. Tubman also enacted a

“Unification Policy” with the aim of improving the political and social status of the

Liberian majority. Suffrage was finally extended to the indigenous population and

Tubman toured the interior, meeting with chiefs and listening to local grievances

(FAS 1985). Tubman earned widespread support among many segments of Liberian

society and ruled until his death in 1971, at which point he was succeeded by William

Tolbert.

Tolbert continued to expand economic investment and political participation, but

long-running problems such as nepotism, corruption, and an intolerance of dissent

plagued Tolbert’s administration. The economic downturn of the 1970s rattled the

stability of the True Whig government. Opposition groups multiplied, despite gov-

ernment repression. And, lacking the charisma of his predecessor, Tolbert struggled

to attract support for austerity measures. When the government proposed in 1979

to increase the price of rice, widespread riots erupted, triggering a police crackdown

that left approximately 40 dead and more than 500 injured (FAS 1985: 66; Pham

2004: 76). The repression failed to end the protests and Tolbert ultimately caved to

the rioters’ demands by artificially lowering the price of rice. The “rice riots” sub-

sided three weeks later after 700 Guinean soldiers arrived in Monrovia to support the

battered government, but the situation remained volatile (Pham 2004: 76).

5.2 The Military Years, 1980-1989

Early in the morning of 12 April 1980 a group of 17 enlisted men entered the Executive

Mansion and killed President Tolbert in his bed. The soldiers were led by Master-

Sergeant Samuel Kanyon Doe, a 28-year-old, semi-literate junior officer from Tuzon,


a small village near the Cote d’Ivoire border. Doe announced the coup on the radio

and celebrations filled the streets in many quarters of Monrovia. More than a century

of Americo-Liberian domination had come to an end.

The People’s Redemption Council (PRC), as the junta became known, quickly

took steps to consolidate its power. Ten days after the coup, on 22 April 1980, thirteen

leading members of Tolbert’s government were tied to wooden stakes on the beach

in Monrovia and publicly executed in front of a large crowd of spectators and the

international media. Over the following months, potential rivals within the military

and government were hunted down and executed. The PRC also implemented pop-

ulist measures, such as increasing the salaries of soldiers and civil servants. Between

1980 and 1983, despite severe public finance problems, the number of government

employees increased from 18,000 to 56,000 (Sawyer 1992: 376, fn. 80).

The coup’s gratuitous violence unnerved the United States, but Doe made clear

that he would remain staunchly pro-U.S. in exchange for ample flows of aid. For

Doe, U.S. support held the key to consolidating his personal power within the PRC

and the Liberian government. Although the original PRC contained members from a

variety of ethnic groups, Doe progressively filled the top positions with family friends

and other loyal confidants. These appointments created the impression of ethnic

favoritism, as the top ranks became increasingly filled with people who shared Doe’s

Krahn ethnicity.3

Doe also entered into economic partnerships with Mandingo traders and busi-

nessmen. Traditionally viewed as “outsiders” by many Liberians, the Mandingos

dominated the regional trade routes and controlled much of the small-scale business

in Liberia. By appointing Mandingos to positions in rural government and grant-

ing them preferential business opportunities, Doe solidified support from a powerful

commercial class and also shared in their profits (Ellis 1999: 61; Phan 2004: 84).

3Many observers claim that Doe favored ethnic Krahns, but this analysis is too simplistic. In1981 Doe ordered the execution of two powerful members of the PRC, Thomas Weh Syen andNelson Toe, both of whom were Krahn. These executions alienated the population of Syen andToe’s home region, leading to a sub-ethnic divide. While Doe maintained the support of Krahnsfrom the Gborbo and Konobo clans, he lost the support of Tchien Krahns from Gbarzon, Webbo,and Gbeapo districts of Grand Gedeh County. In Doe’s regime, personal loyalty counted more thanethnicity (See Ellis 1999: 56; Also C44).

5.2. THE MILITARY YEARS, 1980-1989 149

The preponderance of Krahn and Mandingo in Doe’s regime, however, increased re-

sentment from the majority of Liberians who felt excluded by the first “indigenous”

government.

United States assistance to Liberia increased dramatically during the Doe regime.

From a pre-coup level of $20 million, annual U.S. economic aid tripled by 1984, with

$35 million in budgetary assistance, $13 million in development aid, and $15 million

in rice subsidies. Nearly 200 Peace Corps volunteers implemented a variety of projects

throughout Liberia (FAS 1985: 244). Military aid also expanded during this period.

Rising from a pre-coup level of $2.9 million, military aid increased to $6.3 million

in 1981 and $12.6 million in 1982 (FAS 1985: 279). By early 1984, some 3,000

soldiers in the Armed Forces of Liberia (AFL) had received training from United

States personnel and dozens of officers traveled to the United States for advanced

instruction (FAS 1985: 280).4

Despite this involvement, Doe and his ministers complained that the United States

was “short changing” Liberia and not providing enough assistance to the government.5

For its part, the United States was often exasperated by Doe’s erratic behavior and

profligate spending. Confidential correspondence from the U.S. embassy in Mon-

rovia at the time describes Doe as “extremely emotional, quick to anger and to take

offense.”6 Doe was also reluctant to transition to civilian rule, a key condition of

continued U.S. aid. Only through sustained pressure did the United States convince

Doe to hold elections in 1985. The elections were marred by irregularities.

The irrationality and cruelty of the Doe regime alienated much of the Liberian

population, and frequent purges within the government and military created powerful

enemies among the Liberian diaspora. After the flawed 1985 elections, a group of

dissident soldiers slipped into Liberia from neighboring Sierra Leone. The group

was led by Thomas Quiwonkpa, the former Commanding General of the AFL who

4The Liberian Frontier Force was reorganized in 1959 and became the Armed Forces of Liberia.5Declassified correspondence between the U.S. embassy in Monrovia and the U.S. Secretary of

State, 8 July 1987 (Ref. MONROV 07122).6Declassified correspondence between the U.S. embassy in Monrovia and the U.S. Secretary of

State, 6 June 1988 (Ref. MONROV 05689).


was forced into exile in early 1984 after being accused by Doe of plotting a coup.7

Quiwonkpa’s forces seized control of the national radio station and moved on the

Executive Mansion, but were soon overrun by Doe’s personal security forces.

Quiwonkpa was captured, castrated, and beaten to death, and the victorious sol-

diers paraded his mutilated corpse through the streets. Quiwonkpa and his supporters

were mostly ethnic Gio and Mano from Nimba County. Their coup attempt prompted

Doe’s security forces to carry out a brutal pogrom against Gio and Mano civilians that

left an estimated 1,500 dead (Ellis 1999: 60). The AFL was purged of Gio members

and hundreds of former AFL soldiers fled to refugee camps in Cote d’Ivoire, where

they were joined by thousands of civilians who had escaped the violence in Nimba

County.

After crushing the raid, Doe faced no serious opposition within Liberia. But his

brutality had created many enemies abroad. During one of his frequent purges, Doe

executed Adolphus Tolbert, the son of the late president and husband to the adopted

daughter of Cote d’Ivoire’s leader, Felix Houphouet-Boigny. Houphouet-Boigny per-

sonally intervened to extract a promise from Doe to spare Adolphus Tolbert’s life,

but Doe later reneged on this promise (Ellis 1999: 53; Pham 2004: 81). By the time

Doe was sworn in as a civilian president in January 1986, he faced the spectre of

thousands of Liberian refugees, many with military training, and a host of hostile

neighboring governments.

Eventually the United States lost its appetite for supporting Doe. American aid

fell from $53.6 million in 1986 to $19.5 million in 1989 (Pham 2004: 90). The decline

in aid was compounded by Doe’s economic policies. Liberia’s per capita GDP declined

from $728 in 1980 to only $400 in 1989.8 By the late 1980s Samuel Doe had few friends

at home or abroad. His regime was on its last legs.

7Several of Quiwonkpa’s military supporters staged a raid in late 1983 into Nimba County, Qui-wonkpa’s home area, with the hope of igniting a large-scale rebellion against the Doe regime. Thisincursion failed to gather steam and resulted in considerable violence as Doe’s security apparatusretaliated against ethnic Gio civilians, in what became known as the “Nimba Raid.” After theevents, Quiwonkpa fled Liberia and sought asylum in the United States (See Ellis 1999: 58; Pham2004: 84).

8Statistics are from the World Bank in constant 2000 US$. Data available http://databank.worldbank.org/.

5.3. THE WAR BEGINS, 1989-1992 151

5.3 The War Begins, 1989-1992

On Christmas Eve, 1989, some 100 rebels crossed into Liberia from Cote d’Ivoire and

captured the border town of Butuo, in Nimba County.9 These “special forces” were

led by Charles Taylor, a former official in the Doe regime with extensive international

connections. Taylor formed the group by recruiting former AFL soldiers, mostly

ethnic Gio and Mano, who had fled Doe’s purges in 1984-5 and settled in refugee

camps in Cote d’Ivoire. Through Taylor’s contacts, the soldiers received 18 months

of military training in Libya, logistical support from Burkina Faso, and weapons

courtesy of Cote d’Ivoire. Within days of the invasion, Taylor announced on BBC

World Service the arrival of the National Patriotic Front of Liberia (NPFL).

The NPFL quickly gained momentum. The group established a training center

in Butuo and attracted hundreds of Nimba County residents who had suffered under

the Doe regime. Town after town fell to the NPFL as Doe’s soldiers withdrew from

hostile territory. By late January the NPFL was pushing beyond Nimba County

and advancing towards Monrovia and the coast. In May, two of Liberia’s largest

cities, Gbarnga and Buchanan, fell into rebel hands. Figure 5.1 depicts NPFL’s rapid

advance across Liberia.

The NPFL’s military successes obscured certain problems within the group. Weapons

shipments were delayed because the NPFL lacked money to pay for transporta-

tion, and some commanders wondered if Taylor could fulfill his lofty promises (C39,

TRC159). After two months, Prince Yormie Johnson, who had led the NPFL’s first

crossing into Liberia, broke from the group with about 35 loyal soldiers and formed

the Independent NPFL (INPFL).10 Heavy fighting between the AFL and NPFL pro-

vided cover for the small splinter faction, allowing it to travel fast and light through

the bush, picking up recruits along the way (C17, TRC0D1).

9Before the invasion, the NPFL dispatched several dozen soldiers to Guinea and Monrovia. Thegoal was three simultaneous attacks on Liberia, but the plan failed after the arrest of 22 NPFLmembers in Monrovia and logistical problems for the group sent to Guinea (C24; Ellis 1999: 76).

10In his testimony to the Truth and Reconciliation Commission, Prince Johnson claims he splitwith the NPFL immediately after crossing into Liberia, in December 1989, because Charles Taylorwas a “liar” and did not fulfill his promises (TRC0D1). This timing, however, contradicts indepen-dent, first-hand accounts from NPFL and INPFL commanders who claim Prince Johnson split fromthe group after approximately 2 months (C17, C24).


[Monrovia

Tubmanburg

Voinjama

Gbarnga

Buchanan

April 1990

Dec. 1989

May 1990

June 1990

June 1990

June 1990

June 1990

Figure 5.1: NPFL’s advance across Liberia.

5.3. THE WAR BEGINS, 1989-1992 153

By the end of June, 1990, Prince Johnson’s troops had captured the western

suburbs of Monrovia while the NPFL rapidly closed in from the north and east. Six

months after the invasion began, Charles Taylor’s NPFL controlled more than 80%

of Liberia, including Liberia’s second-largest port, Buchanan, and all of Liberia’s iron

ore and timber resources. Taylor proclaimed himself President of “Greater Liberia”

and formed a civilian administration, the National Patriotic Reconstruction Assembly

Government (NPRAG).

The NPRAG established courts, police, and immigration offices, with the goal of

increasing protection for civilians (C08). Shops and markets reopened in Gbarnga

and other major towns, lending an air of normalcy to Taylor’s Greater Liberia. But

the NPRAG did not stand in the way of a virtual ethnic cleansing of ethnic Krahn

and Mandingo civilians by NPFL soldiers in the group’s territory. The Krahn and

Mandingo were seen as collaborators with the Doe regime and were resented by many

Liberians for the privileges and wealth that accrued to (some) Krahn and Mandingo

under Doe’s government.

Upon hearing the first reports of a rebel incursion, Doe failed to take the threat

seriously. After all, the AFL numbered over 6,000 trained soldiers while the NPFL

could muster barely 100 troops. Doe was taken aback when the first AFL deployment

was beaten back with heavy losses by the NPFL. The AFL rapidly imploded in the

face of the rebels, whose strength grew with each day. By June, desertions in the AFL

dropped the numbers down to 2,000 troops (Pham 2004: 99). In desperation, the

AFL began providing weapons to any civilians motivated to fight against the NPFL.

These volunteers, mostly Mandingo and Krahn targeted by the NPFL, became known

as the “1990 AFL” for their lack of training and discipline.

President Doe begged the United States for assistance, but his pleas fell on deaf

ears.11 The United States withdrew military observers from Liberia in January 1990

after reports of civilian abuse by the AFL. And Iraq’s invasion of Kuwait on 2 August

11In a letter leaked to the Washington Post from Doe to U.S. President George H.W. Bush on 4June 1990, Doe wrote, “If I have failed in regards at times, I ask for the forgiveness of my people...We in Liberia have always considered ourselves ‘stepchildren’ of the United States. We implore youto come help your stepchildren who are in danger of losing their lives and their freedom” (in Pham2004: 100).


1990 eliminated any possibility of U.S. intervention. The only option left for Doe was

to appeal to the regional powers for assistance. The Economic Community of West

African States (ECOWAS), under Nigeria’s leadership, agreed to send troops on the

condition that Doe step down in favor of a transitional government; Doe had no choice

but to agree. A respected intellectual, Amos Sawyer, was chosen as interim president,

and on 24 August 1990 a flotilla containing 3,000 peacekeepers from Nigeria and other

West African states arrived in Monrovia’s port.12

The ECOWAS Military Observer Group (ECOMOG) faced immediate challenges.

Charles Taylor, already the self-proclaimed president of Greater Liberia, resented the

peacekeepers’ plans to install a transitional government and declared war on the

foreign troops. ECOMOG forces were poorly staffed and under-equipped to face

the NPFL threat. To compensate, ECOMOG entered into informal alliances with

the AFL and INPFL. The situation in Monrovia was dire: the NPFL had cut off

the city’s electricity and water supply and civilians struggled to gain access to food.

ECOMOG faced constant mortar attacks, but the peacekeepers, in conjunction with

INPFL and AFL forces, managed to keep the NPFL outside of central Monrovia.

During a lull in the fighting on 9 September 1990, President Doe, who had been

holed up in the Executive Mansion for weeks, ventured to Monrovia’s port for an

unscheduled meeting with the ECOMOG commander, Ghanaian General Arnold

Quianoo. While at the ECOMOG base, Doe was captured by Prince Johnson and

taken back to INPFL headquarters.13 Shaky, handheld video footage chronicles the

subsequent events: a visibly drunk Prince Johnson orders his men to torture Doe,

who lies sobbing and bleeding in a corner of the room.14 The men cut off his ears

before finally shooting him to death. The triumphant INPFL soldiers then loaded

12The Banjul Agreement, 7 August 1990, authorized the creation of a transitional governmentand the deployment of the ECOWAS Military Observer Group (ECOMOG).

13The motivation for Doe’s fateful trip remains shrouded in mystery. According to some theories,Doe was lured to the ECOMOG base by some conspiracy involving either the U.S. or Nigeriangovernments—or both. More likely, Doe was offended that ECOMOG had failed to pay him acourtesy call upon arrival and took it upon himself to confront General Quainoo. Many accounts ofDoe’s capture exist; for good overviews of the evidence, as well as the grisly details of Doe’s torture,see Ellis (1999: 3-14) and Pham (2004: 106-9). Prince Johnson shares his side of the story duringhis testimony to the Truth and Reconciliation Commission (TRC0D1).

14Bootleg videos of this event were a common sight on Monrovia’s streets until about 2005.

5.4. A FAILED STATE, 1992-1997 155

Doe’s body into a wheelbarrow and paraded his corpse through the streets.

Doe’s death triggered a change in leadership in ECOMOG. Nigerian Major Gen-

eral Joshua Dogonywaro took over as head of the peacekeeping mission and, giving

up all pretense of impartiality, launched a major offensive against the NPFL rebels,

resulting in a short-lived ceasefire in November 1990. The calm allowed ECOMOG

to install Amos Sawyer as the head of the Interim Government of National Unity

(IGNU) in Monrovia, but ECOMOG was unable to maintain its momentum against

the NPFL. The ceasefire also weakened ECOMOG’s troubled ally, the AFL, by pro-

viding a tempting opportunity for AFL soldiers to quit their posts and join a swelling

number of Liberian refugees in Sierra Leone.15

A military stalemate existed through 1991, with ECOMOG and INPFL forces

largely confined to Monrovia, and the NPFL controlling most of the hinterland. All

factions—including the ECOMOG peacekeepers—were responsible for looting, rape,

and violence against the civilian population. Aid organizations struggled to meet

the demands of hundreds of thousands of desperate civilians. Meanwhile the NPFL,

through its control of key resources and port facilities, earned tens of millions of

dollars through partnerships with foreign iron and timber corporations.

On 15 October 1992, the NPFL launched a savage attack on Monrovia that became

known as “Operation Octopus.” NPFL soldiers went on a looting rampage in the city,

while ECOMOG forces struggled to push back the offensive. Thousands of civilians

were killed in the fighting, mostly due to ECOMOG’s indiscriminate shelling and

use of napalm (Ellis 1999: 99). The NPFL failed to capture the city, but the group

succeeded in breaking the back of Prince Johnson’s INFPL. His forces in disarray,

Prince Johnson fled the country for exile in Nigeria.

5.4 A Failed State, 1992-1997

Liberia’s civil war entered a new phase in early 1992 when a new rebel group, the

United Liberian Movement for Democracy (ULIMO) captured strategic towns along

15C02, C06, C09, C12, C13, C28. Some AFL soldiers who wanted to leave Liberia receivedassistance from ECOMOG, which provided transportation by ship from Monrovia to Freetown (C12).


the Sierra Leone-Liberia border. ULIMO had its roots in the Liberian refugee camps

in Sierra Leone, where hundreds of AFL soldiers sought refuge after the NPFL in-

vasion. The refugees’ situation deteriorated even further in March 1991, when the

Revolutionary United Front (RUF), with support from the NPFL, attacked the bor-

der areas of Sierra Leone. The RUF conducted a brutal campaign that targeted Sierra

Leone civilians as well as Liberians in the sprawling refugee camps.

ULIMO’s history began when a group of Krahn ex-AFL soldiers, faced with the

RUF threat, formed a self-defense group in the refugee camps under the leadership of

Albert Karpeh, Liberia’s ambassador to Sierra Leone and a retired AFL General.16

The group received weapons from the Sierra Leone military in exchange for supporting

the government’s attempt to push back the RUF rebels. The self-defense group

attempted to enter Liberia on several occasions, but the outnumbered fighters were

repeatedly routed by the NPFL.

To increase the group’s strength, Karpeh agreed to join forces with Alhaji G.V.

Kromah, a Mandingo intellectual who led an organization of Mandingo exiles in

Guinea.17 The group members lacked military training, but provided much needed

manpower for Karpeh’s self-defense forces. The newly joined forces adopted the

name ULIMO, with Karpeh as leader and Alhaji Kromah as the group’s spokesman.

Kromah, however, was dissatisfied with his position and installed himself as leader

after orchestrating Karpeh’s assassination.

With Alhaji Kromah at the helm, ULIMO swept across northwestern Liberia in

1992, as the NPFL clashed with ECOMOG forces near Monrovia. The strategic towns

of Tubmanburg and Klay fell to the rebels and, in August 1992, ULIMO reached the

Po River Bridge, just 25km from Monrovia. The Sierra Leone government stopped

supplying weapons to ULIMO after it crossed into Liberia, but the group soon formed

a partnership with ECOMOG to fight the NPFL in exchange for ammunition (C09,

C13, C44; Ellis 1999: 95). By the time the NPFL rebels launched Operation Octopus

and defeated Prince Johnson’s INPFL, ULIMO had opened up a new front in the

war. Figure 5.2 depicts ULIMO’s advance against the NPFL.

16The self-defense group was named the Liberian United Defense Force (LUDF).17Kromah’s organization was named the Movement for the Redemption of Liberian Muslims

(MRM).

5.4. A FAILED STATE, 1992-1997 157

[Monrovia

Tubmanburg

Voinjama

Gbarnga

Buchanan

Aug. 1992

NPFL

ULIMO

Figure 5.2: ULIMO’s advance against the NPFL.


In concert with ULIMO’s advance, ECOMOG stepped up the pressure on NPFL.

The peacekeepers implemented a blockade of NPFL areas, preventing trade and even

humanitarian aid from entering Greater Liberia and carried out a controversial bomb-

ing strategy that inflicted damage on civilian and humanitarian targets (Ellis 1999:

100; Scott 1995: 11). This dual offensive, while devastating to the civilian population,

brought the NPFL back to the bargaining table. On 25 July 1993, representatives

from the various factions signed the Cotonou Agreement. The agreement sought to

end the fighting by giving the warring factions a stake in government. A new tran-

sitional government was created, headed by an executive council of representatives

from the AFL, ULIMO, and NPFL.

For all its promise, the Cotonou Agreement failed to stem the violence. In Oc-

tober 1993 a new faction emerged, the Liberian Peace Council (LPC). Formed with

ECOMOG assistance and composed mostly of former ULIMO and AFL soldiers, the

LPC opened yet another front against the NPFL, this time in Liberia’s remote south-

eastern region. Mostly a Krahn organization, the LPC based itself in Greenville and

fanned outward towards Grand Gedeh, Maryland, and Grand Bassa Counties, leav-

ing a trail of civilian abuse in its wake. The LPC proved especially painful for the

NPFL, as valuable timber areas slipped beyond NPFL control and the LPC’s reign of

terror made it impossible for foreign companies to continue their operations. When

the port of Buchanan fell to the LPC in 1994, the NPFL’s access to revenue and

weapons, central to its survival, began to dwindle.

The Cotonou Agreement also created problems for ULIMO. Disagreements over

appointments in the transitional government led to a split in March 1994 along

ethnic lines. Fighting broke out in ULIMO’s headquarters at Tubmanburg after

Mandingo forces loyal to Alhaji Kromah attempted to disarm their Krahn comrades

(C02, C04, C06).18 Throughout ULIMO territory, Krahn and Mandingo soldiers de-

serted their commanders in favor of ethnically homogeneous units (C02, C04, C06,

C09). The Mandingo forces loyal to Alhaji Kromah became known as ULIMO-K,

18Mandingo commanders present in Tubmanburg at the time insist that the Krahn soldiers startedthe shooting, although it is generally agreed that the Krahn fighters withdrew from Tubmanburgafter some of their men were captured and disarmed by Mandingo forces (C18, C43, C44).

5.4. A FAILED STATE, 1992-1997 159

while the Krahn forces reconfigured themselves as ULIMO-J under Roosevelt John-

son, ULIMO’s chairman of military affairs. After a few days of confusion, fierce

fighting erupted between the two groups in the areas around Tubmanburg and Lofa

Bridge.19

Repeated attempts at peace agreements seemed only to intensify the fighting. The

Akosombo Agreement in September 1994 supplemented the Cotonou Agreement but

served mainly to trigger a split in the NPFL. The Akosombo Agreement allocated

high-level positions in the transitional government to three of NPFL’s top officials.

When Charles Taylor reneged on the agreement, these officials decided to hold onto

their new offices and break from the NPFL (C08, C26). With the military backing of

NPFL commander Nixon Gaye, the rogue officials formed the Central Revolutionary

Council (NPFL-CRC).

The conflict intensified through early 1995 as the NPFL faced an unstable coalition

of hostile forces, all of whom received assistance from the ECOMOG peacekeepers.

Fighting raged over 80% of Liberia’s territory and the civilian population suffered

extreme hardship and abuse (UNSC 1995a). Humanitarian aid rarely extended be-

yond Monrovia and the number of Liberian refugees swelled to nearly 800,000.20 The

NPFL remained the dominant military power, but became increasingly vulnerable

after losing ground in the southeast and briefly losing control of its headquarters at

Gbarnga. Figure 5.3 depicts territorial control in 1995.

In August 1995 the factions returned to the negotiating table to sign the Abuja

Agreement, which established yet another transitional government and set a timetable

for disarmament. Huge crowds greeted Charles Taylor as he entered Monrovia for the

first time since the start of the war to assume his position on the six-member Council

of State. Unlike previous agreements, the Abuja ceasefire led to a decrease in fighting

among the warring factions, as their leaders settled into the intrigues of Monrovia

politics. Minor skirmishes continued, but the next major violence did not occur until

three months later, in November 1995, when ULIMO-J attacked ECOMOG forces

19ULIMO also faced difficulties in parts of Lofa County due to the Lofa Defense Force (LDF), asmall militia that allegedly received support from the NPFL (Ellis 1999: 102). The LDF was nevera significant military power and soon joined forces with ULIMO in an anti-NPFL coalition.

20According to UNHCR estimates.


[Monrovia

Greenville

Tubmanburg

Voinjama

Gbarnga

Buchanan

NPFL

LPC

ULIMO-J

ULIMO-K

Figure 5.3: Territorial control in Liberia, 1995.

5.5. RETURN TO WAR, 1997-2005 161

near Tubmanburg.

The attack on the peacekeepers served to isolate ULIMO-J and led to a leadership

struggle within the organization. Charles Taylor capitalized on the group’s weakness

by forming an alliance with his former enemy, ULIMO-K leader Alhaji Kromah.

On 6 April 1996, NPFL and ULIMO-K forces attempted to arrest ULIMO-J leader

Roosevelt Johnson at his headquarters in Monrovia. Krahn fighters from ULIMO-J,

LPC, and AFL came to Johnson’s aid, and Monrovia descended into some of the

worst violence of the war. Fighting in the city center carried on for weeks, as gunmen

preyed upon civilians and half of Monrovia’s swollen population of 1.3 million fled

their homes (UNSC 1996b).

When the fighting subsided, Charles Taylor and Alhaji Kromah had cemented

their dominance. The groups returned to the negotiating table and in August 1996

reaffirmed the Abuja Agreement. Disarmament began in November and, with an eye

towards the upcoming elections, Charles Taylor and Alhaji Kromah ordered their

soldiers to cooperate; ULIMO-J and LPC reluctantly followed suit. By June 1997,

more than 21,000 soldiers had disarmed, surrendering more than 10,000 weapons and

1.2 million pieces of ammunition (UNSC 1997). The increasing security paved the

way for elections on 19 July 1997, when some 85% of more than 750,000 voters cast

their ballots. The elections were heavily scrutinized by international observers, who

judged the process to be transparent, free, and fair. When the votes were counted, a

landslide 75% of voters had chosen Charles Taylor to be their president (Pham 2004:

133). After seven years of war, Charles Taylor achieved his goal of becoming Liberia’s

head of state.

5.5 Return to War, 1997-2005

Upon taking office, President Charles Taylor rewarded NPFL commanders with po-

sitions in his new government but also extended offers to his former rivals. Alhaji


Kromah and Roosevelt Johnson were both offered positions in Taylor’s administra-

tion21 and fighters from the various factions were incorporated into the state security

apparatus (C12, C18). Within a few months, Taylor solidified his control over the

government. Although Roosevelt Johnson used his government position to maintain

some base of support, all the valuable positions in government were controlled by

Taylor’s most trusted allies. And Taylor’s new legitimacy as a democratically-elected

head of state gave him control over new sources of revenue and a political advantage

over any potential troublemakers.

Once secure in his power, Taylor—like Doe before him—began purging his gov-

ernment of potential rivals. Mandingo civil servants felt increasingly targeted by

Taylor, and many left Liberia to seek better opportunities in Guinea (C46). In early

1998 Taylor forced some 2,500 Krahn soldiers in the AFL into early retirement (C32;

Also Pham 2004: 177). This move crippled the organization, allowing Taylor’s newly

formed personal security force, the Anti-Terror Unit (ATU), to become the primary

military organization in the country.

The ATU’s first operation was to invade Roosevelt Johnson’s headquarters in

Monrovia, with the intention of “disarming” him (C18). Johnson and his supporters

fought back, and on 18 September 1998 Monrovia once again erupted in fighting.

The weakened Krahn forces were no match for Taylor’s well-trained and fiercely loyal

ATU fighters. Roosevelt Johnson fled to exile in Nigeria and, over the coming months,

hundreds of Krahn soldiers, civil servants, and civilians left the country after being

“hunted” by Taylor’s forces (C02, C03, C09, C25, C31, C41, C48). Barely a year

after the end of a brutal civil war, refugee camps in Guinea and Cote d’Ivoire began

to swell with new arrivals.

In April 1999, small groups of former ULIMO soldiers began carrying out un-

coordinated “hit-and-run” attacks into Liberia from the refugee camps in Guinea.

Exiled politicians tried to unite these forces into a coherent military organization,

but struggled to gain outside support for an insurgency. Sierra Leone seemed like a

21Alhaji Kromah, who placed 3rd in the 1997 elections was appointed chairman of a ReconciliationCommission, but declined and soon left Liberia; Roosevelt Johnson became Minister of Rural De-velopment. Other ULIMO officials also received positions. For example, Raleigh Seekie, ULIMO’sinitial chairman, was appointed Auditor General in Taylor’s government (C04).

5.5. RETURN TO WAR, 1997-2005 163

natural ally, as attacks from the Taylor-backed RUF rebels continued to threaten the

government, and small groups of Liberians were fighting alongside the government

forces. But Sierra Leone refused to provide a base for Liberian dissidents and briefly

arrested some of the aspiring group leaders. Cote d’Ivoire and Guinea also declined

to provide assistance to avoid stirring up trouble with their Liberian neighbor (C09,

C29). Without external support, the exiles had no hope of overthrowing Taylor and

returning to their homes.

The Liberian dissidents received a break in 2000 after a small group of Guinean

rebels, supported by Charles Taylor, briefly seized the Guinean town of Guekedou.

After defeating the rebels, Guinean President Lansana Conte became more sympa-

thetic to the exiled Liberians’ goals and Guinean military commanders began provid-

ing limited weapons and ammunition to the Liberian dissidents (C46). The Liberian

politicians competed with each other to lead the organization, but were eventually

forced to accept Sekou Conneh—a former used car salesman with no political or mil-

itary experience—as their chairman. Conneh was married to the spiritual advisor

of the Guinean president, and President Conte had insisted on Conneh as leader if

the group hoped to receive any military assistance. The exiled politicians set aside

their misgivings and formed the Liberians United for Reconciliation and Democracy

(LURD).

In July 2000, LURD invaded Liberia from Guinea and briefly captured Voinjama,

the capital of Lofa County. Taylor’s forces pushed back the rebels and then retaliated

by crossing into Guinea and destroying several towns. Outraged by this attack—and

worried about the loyalty of his own security forces—President Conte dramatically

increased his military support to LURD (ICG 2002a: 4).22 Conte’s goal was to

keep Taylor tied-up militarily in Lofa County, thereby damaging Taylor’s regime and

preventing future incursions into Guinea (UNSC 2001: 30). For the next 18 months,

LURD and Taylor’s forces rode a see-saw of military victory and retreat, with LURD

capturing large swaths of Lofa County, only to pull back a few months later when

confronted by a government offensive.

22Guinea also benefitted from increased military support from the United States during this period,which increased its supply of weapons and made supporting LURD less costly for Conte’s regime(ICG 2002a).


The tide turned in early 2002 when LURD swept across northern Liberia and

captured important towns such as Tubmanburg and Klay. Although Taylor’s forces

regained some territory, LURD became firmly entrenched in Lofa County and proved

itself to be an existential threat to Taylor’s regime. Taylor dramatically expanded

the size of his ATU forces and formed an additional militia by recalling former NPFL

commanders into military service (C19, C21, C26; Also ICG 2002a). By May 2002,

intense fighting and extreme levels of civilian abuse resulted in more than 100,000

internally displaced people and some 60,000 refugees (ICG 2002b: 1). But even

with an expanded security apparatus, Taylor could not counter the rebels’ onslaught,

largely due to a United Nations arms embargo (C07, C21, C26). Figure 5.4 depicts

LURD’s advance across Liberia.

An invasion by a new rebel group in March 2003 compounded the threat to Tay-

lor’s regime. From their base in the “Peace Town” refugee camp, the Movement

for Democracy in Liberia (MODEL) crossed into Liberia from Cote d’Ivoire and cap-

tured Zwedru, the capital of Grand Gedeh County.23 MODEL was composed of Krahn

former ULIMO fighters, under the leadership of Thomas Yaya Nimely, a trained psy-

chologist with United States citizenship. Yaya Nimely had spent the past three years

trying to gain support for an insurgency against Taylor, and finally succeeded after

civil war erupted in Cote d’Ivoire in September 2002.24 Tens of thousands of Liberian

refugees, many with military experience, were trapped on Cote d’Ivoire’s frontline as

Taylor-backed rebels captured the northern half of the country.

In a repeat of the strategies used by Sierra Leone and Guinea, Cote d’Ivoire

began supplying weapons to the Liberian refugees in exchange for military support.

These Liberian forces fought alongside the Ivorian military in fierce battles to retake

key towns such as Blolekin and Toulepleu.25 Within a few months, the group had

received enough military supplies to begin its assault on Liberia (C23, C25, C29,

C31). The group swept across southeastern Liberia, capturing the country’s second

23”Peace Town”, officially named Nicla refugee camp, was located on the outskirts of Guiglo, Coted’Ivoire.

24For a time, Yaya Nimely was part of the LURD leadership, but left the group after Sekou Connehwas appointed leader (C09, C29).

25The group called itself Liberians Mobilized in Action (LIMA).

5.5. RETURN TO WAR, 1997-2005 165

[Monrovia

Voinjama

Zwedru

Foya

TubmanburgGbarnga

Buchanan

March 2003

Feb. 2002

Figure 5.4: LURD’s advance across Liberia.


largest city, Buchanan, in July 2003 and reaching the outskirts of Monrovia by early

August. Figure 5.5 depicts MODEL’s advance. At the same time, LURD penetrated

Monrovia from the north, in a series of devastating attacks that became known as

“World War I”, “World War II”, and “World War III”.

Under intense international pressure and facing a critical shortage of arms, Charles

Taylor resigned as president on 11 August 2003 and accepted exile in Nigeria. An

ECOWAS peacekeeping force (ECOMIL) landed in Monrovia and celebrations filled

the streets as rival rebel soldiers dropped their weapons and embraced on the Vai

Town bridge, which just hours before had marked the front line of fierce fighting.

Two months later, a robust United Nations peacekeeping mission (UNMIL) deployed

and grew to include over 15,000 military personnel.

A transitional government took control of the country in October 2003. Led by

a civilian, Gyude Bryant, the government’s ministries and other key positions were

divided among the warring factions. The transitional government’s two-year tenure

was characterized by spectacular corruption that, in the words of some Liberians

at the time, made Charles Taylor look like a “responsible” president. Nevertheless,

the United Nations successfully disarmed more than 100,000 soldiers, receiving some

28,000 weapons and 6.5 million pieces of ammunition (UNSC 2005b: 5).26 U.N.-

supervised elections were held in 2005 and Liberians elected Ellen Johnson Sirleaf, a

Harvard-trained economist with experience in international organizations, to the top

position. On 16 January 2006, amidst a groundswell of optimism, the new govern-

ment took office and, with substantial international support, commenced rebuilding

a country devastated by fourteen years of war.

5.6 Discussion

Liberia’s civil war does not lend itself to easy explanations that focus on ethnic

animosity or greed. Explanations that focus on tribal war cannot explain the ethnic

26The number of officially disarmed soldiers is highly inflated, as the disarmament program offeredan irresistible opportunity for factions to distribute patronage to their family and friends. From ananalysis of the command structures of the various factions, the number of full-time soldiers wasprobably around 10,000 during 2000-2003 and almost certainly did not exceed 20,000.

5.6. DISCUSSION 167

[Monrovia

Voinjama

Zwedru

Foya

TubmanburgGbarnga

Buchanan

LURD

March 2003

July 2003

Aug. 2003

April 2003

May 2003

May 2003

Figure 5.5: MODEL’s advance across Liberia.


diversity of NPFL, ULIMO, and LURD and overlook the legitimate political and

social grievances of the people. Claims that the war was simply a battle for control

over Liberia’s natural resources cannot explain the variation in behavior among the

armed groups. Liberia’s rebels faced complex organizational challenges, and their

behavior reflects these challenges. The next chapter goes inside Liberia’s rebel groups

to show how partnerships with external patrons and the groups’ access to resources

shaped the behavior of group members.

5.A Timeline of Liberia’s War

Date Event

24 Dec. 1989 Charles Taylor’s NPFL invades Liberia from Cote d’Ivoire

Feb. 1990 Prince Y. Johnson splits from Taylor’s group and forms INPFL

May 1990 NPFL captured Buchanan, Liberia’s second-largest port

June 1990 INPFL and NPFL forces reach Monrovia

Aug. 1990 ECOWAS peacekeepers deploy in Monrovia

Sept. 1990 President Doe killed by Prince Y. Johnson’s forces

Nov. 1990 IGNU government under Amos Sawyer installed

March 1991 RUF rebels invade Sierra Leone, prompting formation of LUDF/ULIMO

Early 1992 ULIMO invades Liberia from Sierra Leone

Aug. 1992 ULIMO reaches outskirts of Monrovia

Oct. 1992 NPFL launches “Operation Octopus” and defeats the INPFL

July 1993 Cotonou Agreement is signed

Oct. 1993 LPC emerges in the southeast

March 1994 ULIMO splits into ULIMO-J and ULIMO-K

Sept. 1994 NPFL political officials split from Taylor and form NPFL-CRC

Aug. 1995 Abuja Agreement is signed

6 April 1996 Attempted arrest of ULIMO-J leader sparks weeks of fighting in Monrovia

Nov. 1996 Disarmament begins

July 1997 Charles Taylor elected president

18 Sept. 1998 Crackdown on former ULIMO-J members sparks unrest and displacement

5.A. TIMELINE OF LIBERIA’S WAR 169

July 2000 LURD invades Liberia from Guinea and briefly capture Voinjama

Early 2002 LURD advances across northwestern Liberia

March 2003 MODEL invades Liberia from Cote d’Ivoire and captures Zwedru

May 2003 LURD begins the first in a series of attacks on Monrovia

July 2003 MODEL captures Buchanan and advances on Monrovia

11 Aug. 2003 Charles Taylor resigns as president and takes exile in Nigeria

Aug. 2003 ECOMIL peacekeepers deploy in Monrovia

Oct. 2003 Transitional government led by Gyude Bryant takes office

Oct.-Nov. 2005 Ellen Johnson Sirleaf elected president

Chapter 6

Inside Liberia’s Rebel

Organizations

Liberia’s civil war involved four main rebel groups: NPFL, ULIMO, LURD, and

MODEL. This chapter discusses the formation and operation of these groups, as

well as key challenges faced by the groups. The narratives support the theory’s three

causal mechanisms: (1) external patrons shape the characteristics of rebel leaders and

the financial resources available to them; (2) the leaders’ ability to control their com-

manders depends on spot payments, credible promises, and the presence of lootable

resources; and (3) commanders are capable of controlling their soldiers, although they

may choose not to.

6.1 NPFL: Open for Business

NPFL leader Charles Taylor came to power through his connections with foreign

governments and corporations and his ties to skilled commanders. External patrons

provided Taylor with financial resources that he used to create incentives and enforce

discipline among the group’s commanders, at least in stable and accessible areas. On

two occasions Taylor’s ability to offer these incentives faltered and the NPFL split

into competing factions.

170

6.1. NPFL: OPEN FOR BUSINESS 171

6.1.1 Formation

The roots of the NPFL date back to the 1985 coup attempt against Samuel Doe by

General Thomas Quiwonkpa. Members of the government and military who shared

Quiwonkpa’s Gio ethnicity were targeted in the aftermath, and hundreds of Gio sol-

diers left Liberia for refugee camps near Danane, Cote d’Ivoire. Almost immediately,

the soldiers began organizing to return home and overthrow Samuel Doe’s regime.

The soldiers met weekly and appointed a former General, Nicholas Podier, as their

leader because he had international experience and the soldiers hoped to convince a

foreign government to supply weapons and training (TRC0D1; Huband 1998: 48-50).

Podier’s attempts to secure external support failed to bear fruit, and the soldiers

began looking for other, more promising leaders.1 The soldiers were contacted by

Alfred Mehn, a well-respected government official who had fled Doe’s purges. Mehn

informed the soldiers about another former official, Charles Taylor, and explained that

Taylor could arrange training for them in Libya, as well as support from Burkina Faso

and Cote d’Ivoire. Taylor also had relationships with leading Ivorian, French, and

Lebanese business people who had commercial interests in Liberia (Ellis: 1999: 89).

Most of the soldiers had never heard of Taylor and were skeptical about his

promises, especially since Taylor was imprisoned in Ghana at the time. Influen-

tial soldiers in the group, however, shared family ties with Taylor’s wife, Agnes, and

convinced others to rally behind Taylor’s leadership (C39, TRC0D1). Some refused

to join Taylor’s forces and rallied instead behind Boima Fahnbulleh, a prominent

political activist and former government minister. Fahnbulleh, however, struggled

to find external support, and his soldiers were absorbed into Taylor’s forces after

Fahnbulleh’s personal finances were exhausted (TRC0D1).2

True to his word, Taylor arranged passage to Libya for 156 men for intensive

military training.3 Although the training was difficult, NPFL soldiers were well fed

1Later in 1987 Podier was captured in Liberia and executed by Doe’s forces (Huband 1998: 51).2Mark Huband (1998) tells a slightly different version of this story. Fahnbulleh and his partner Joe

Wylie sent 13 troops to Libya, but Charles Taylor undermined the soldiers’ confidence in Fahnbulleh,causing most of the group to return to Ghana. Only 3 members of Fahnbulleh’s group decided toremain with Taylor.

3The figure of 156 initial recruits comes from a document entitled “Original Roster of the SpecialForces Commandos of the National Patriotic Front of Liberia,” provided by a senior NPFL official

172 CHAPTER 6. INSIDE LIBERIA’S REBEL ORGANIZATIONS

and received a salary of $150-200 per month from the Libyan government (C39). Still,

some members of the “Special Forces” remained uncertain about Taylor’s leadership

even after arriving in Libya. One of the group’s commanders, Cooper Miller, took

advantage of Taylor’s frequent absences from the camp and tried to rally support for

a change in leadership. Miller and his deputy, Augustus Wright, had recruited many

in the initial group and claimed that Taylor wasn’t necessary for securing external

patronage (Huband 1998: 52). Instead, the group should choose their leader through

elections. Other members reported this dissent to Taylor, who immediately had Miller

and Wright arrested and expelled from the group (C39, TRC0D1). Taylor’s authority

was not questioned again.

6.1.2 Resources and Commander Incentives

After 18 months of training, the “Special Forces” returned to Cote d’Ivoire to begin

their invasion of Liberia. Taylor had assured the group that plenty of weapons and

ammunition would be available for the offensive and promised to reward each com-

mander after the war with $75,000 and a house. In addition, Taylor pledged that the

highest positions in his post-war government would be reserved for the Special Forces

(C24, C39). Despite these assurances, the NPFL began its mission with few resources.

A weapons shipment from the Ivorian military was delayed for several weeks because

the NPFL lacked money to hire trucks to transport the munitions (TRC0FF). The

group’s first attack on a military outpost across the border in Butuo relied on just a

few shotguns.

The commander who led the initial attack, Prince Y. Johnson, saw this lack of

resources as a sign that Taylor could not deliver on his lofty promises (TRC0D1).

Johnson also feared that Taylor’s escape from a U.S. prison a few years earlier would

damage the NPFL’s international image and prevent the United States from ever

supporting the group (C39). A few weeks into the war, rather than returning to

present in Libya at the time (C39). This figure differs slightly from other published figures. Ellis(1999: 75) writes that 167 recruits were trained in Libya, while Pham (2004: 93) cites 168 re-cruits. Both of these estimates likely derive from Mark Huband’s (1998) descriptions of the NPFL’sexperiences in Libya and his estimate of 168 Liberian recruits.


base as ordered, Johnson decided to split from the group. His forces continued their

offensive through Butuo, captured weapons from a military base, and proceeded down

the road to Karnplay. Traveling light and fast, Prince Johnson and his “Independent”

NPFL forces (INPFL) planned to push forward all the way to Monrovia and beat the

NPFL to its goal of capturing the government (C17).

Some of Prince Johnson’s men deserted the rogue commander and reported John-

son’s plans to Taylor (TRC0D1). But with few weapons and limited manpower,

Taylor could not prevent Johnson from breaking off from the group or catch up with

his forces (C24). As Johnson blazed a trail across Liberia, Taylor had to regroup

his forces. He constructed a training base to expand the group’s membership and

resolved the logistical problems. By early March, supplies from the Ivorian military

were complemented by the delivery of some 250 automatic weapons and hundreds

of hand grenades and RPGs from Burkina Faso (Huband 1998: 62). These weapons

shipments, however, failed to keep pace with the group’s expansion. In desperate

need of hard currency to purchase additional weapons from arms dealers, the group

looted machinery from the abandoned Bong Mining Corporation, a German-owned

operation with connections to the Doe regime (Reno 1999: 95).

The NPFL’s access to revenue increased dramatically in May 1990 with the cap-

ture of Buchanan, Liberia’s main port for timber and iron exports. Although many

companies had suspended their activities at the outbreak of war, they were eager to

resume operations. Charles Taylor reaffirmed his commitment to his foreign partners

and stated that the NPFL would provide “maximum protection to expatriate logging

companies” (in Reno 1999: 95).4 Foreign corporations found in Taylor a capable

business partner. The bureaucracy of the Doe era was eliminated, as Charles Taylor

personally negotiated contracts and granted operating licenses in exchange for direct

payments to his international bank accounts (Ellis 1999: 90).

After reaching the outskirts of Monrovia in June 1990, Charles Taylor reorganized

his forces from a single Battle Group into four Divisions: Army, Marines, Navy, and

Executive Mansion Guard (EMG). Each division was headed by a chief of staff who

4This quote was originally published in 1992 in the Patriot, the NPFL’s official news organ.Taylor’s relations with these foreign firms, however, began years earlier.


reported directly to Taylor and had a well-defined zone of operation. The Army and

Marines Divisions were responsible for the frontline surrounding Monrovia, and were

based near Kakata and in the Firestone plantation. The Navy Division, based in

Buchanan, controlled most of the coastline as well as the timber areas. The EMG

Division was based at the NPFL’s headquarters in Gbarnga and controlled Bong and

Nimba Counties. While the Army and Marines Divisions were primarily responsible

for fighting, the Navy and EMG Divisions provided security to timber and iron-ore

companies and ensured that Buchanan port remained open for business. Figure 6.1

depicts the NPFL’s command structure.

By late 1990, resources were being extracted at large-scale from Taylor’s Greater

Liberia. The Liberian Minerals Company (LIMINCO) reportedly paid Charles Taylor

$10 million per month for the opportunity to extract and export iron ore from NPFL

territory (Reno 1999: 100).5 Another mining operation in Nimba County netted

Taylor about $80,000 per month (Ellis 1999: 164; Reno 1999: 100). During 1992

alone, logging firms removed an estimated 200,000 cubic meters of hardwood, valued

at $20 million, from NPFL territory (Reno 1999: 97). Even smaller logging outfits

in less-stable areas such as Lofa County could deal directly with Taylor and enjoy

NPFL protection in exchange for taxes similar to those paid under the Doe regime

(C45).

The economy of Taylor’s Greater Liberia also included imports of consumer goods

and agricultural exports. Lebanese merchants reopened their shops in Gbarnga af-

ter purchasing import licenses from Taylor and paid monthly taxes to the NPFL

(B43). Taylor also reached an agreement with the K&K corporation to import rice

for NPFL soldiers.6 In the agricultural sector, Charles Taylor signed a memoran-

dum of understanding with Firestone in 1992 that allowed the corporation to resume

rubber operations. Firestone provided communications equipment, a staging ground

for NPFL forces, and $2 million per year in exchange for security and access to the

5This figure, which is also cited by Liberia’s Truth and Reconciliation Commission (TRC 2009:Vol. III, Title III, par. 83) and Pham (2004: 122), derives from a report by the Economist IntelligenceUnit in 1992. The reliability of this figure is questionable, but the estimate remains the best available.

6Based on an NPFL document that describes the distribution of 3,475 bags of rice among NPFL’svarious divisions; Also C39.


Ch

arle

s T

aylo

r

Arm

y D

ivis

ion

Mar

ines

Div

isio

nN

avy

Div

isio

nE

MG

Div

isio

n

Join

t C

hie

f of

Sta

ff

(rota

tin

g)

Joh

nso

nT

. Lea

mah

(1990-5

)

Ro

lan

d D

uo

(1995-6

)

Joh

n T

eah

(1990-6

)N

ixo

n G

aye

(1990-4

)

Mel

vin

So

gb

and

i (1

994-6

)

Cas

siu

s Ja

cob

s (1

990-3

)

Pri

nce

C. J

oh

nso

n (

1993-6

)

3-4

Bat

tali

on

s4 B

atta

lion

s2-3

Bat

tali

on

s4-5

Bat

tali

on

s

Not

e:A

rrow

sin

dica

tech

ain

ofco

mm

and

and

influ

ence

.D

otte

dar

row

sin

dica

tew

eake

rin

fluen

ce.

Fig

ure

6.1:

NP

FL

com

man

dst

ruct

ure

.


Buchanan port (Ellis 1999: 100; Reno 1999: 167; TRC 2009: Vol. II, p. 226). Har-

vesting rubber, however, proved difficult during the war and Firestone was never able

to resume full operations. In 1993 Firestone officially exported only 12,000 tons of

rubber from Liberia, just 11% of pre-war levels (Ellis 1999: 167). But outside of these

official channels, a black market trade in rubber thrived. NPFL soldiers and civilians

looted rubber from Firestone’s plantation for sale to ECOMOG forces, earning $300

per truckload (C09; Ellis 1999: 167).

From late 1990 through 1993, Charles Taylor generated revenues estimated at $75-

250 million per year.7 Taylor used some of this revenue to purchase weapons from

arms dealers in the former Soviet Union and relied on his commercial partnerships

for deliveries (Ellis 1999: 90). For example, the ITI logging company maintained an

airstrip on its concession in River Cess that the NPFL used for importing military

supplies (Global Witness 2003a: 21). Through his links with external actors, Taylor

created a formidable military and economic machine.

Charles Taylor also used his financial resources to provide cash payments to his

top commanders as reward for following orders (C15, C24, C26, C38). Payments of

$10-30,000 in cash were not unusual, although commanders were expected to share

some of the money with their soldiers (C19, C39). These rewards encouraged com-

manders to compete for Taylor’s approval and strengthened Taylor’s ability to punish

commanders who did not perform to expectations. Taylor was especially harsh with

commanders who failed on the battlefield. High-level commanders including Cassius

Jacobs, Putu Major, and James Kpehe were executed on Taylor’s orders because they

lost territory and weapons to rival armed groups (C24, C38).

In addition to performing on the battlefield, Taylor expected his commanders to

maintain order and provide security for civilians. Taylor often gave stern speeches

to his commanders and warned against harassing civilians by saying, “Don’t play

with my pepper bush,” where “pepper bush” is a Liberian expression for something

important and personal (C26). Civilians held the key not only to NPFL’s food supply,

7The lower estimate of $75 million per year comes from congressional testimony from William H.Twaddell, Acting Assistant Secretary of State for African Affairs (USDS 1996) and is reproduced inEllis (1999: 90-1). Scholar Will Reno (1998: 99) surveys the available secondary sources to producethe high estimate of $200-250 million per year.


but also to Taylor’s legitimacy. Foreign firms could justify their business with Taylor

only if Taylor maintained the appearance of a responsible leader (Reno 1998: 102).

A large civilian population in NPFL territory also gave Taylor more leverage during

peace negotiations. To strengthen civilian security and his own legitimacy, Taylor

created the NPRAG government to settle civilian disputes and provide security in

major towns under NPFL control. The NPRAG established a court system and

civilian police force, and its various ministries were staffed by respected members

of the elite, such as the U.S.-trained lawyer Lavala Supuwood. When NPFL forces

refused to obey the NPRAG’s authority in stable areas, Taylor personally intervened

to maintain order (C08).

NPFL commanders who failed to provide security for civilians were removed from

stable areas and either punished or transferred to the frontline, where looting was

less damaging to the group (B43, C08, C24, C39). For example, in mid-1990 Charles

Taylor heard on the BBC that NPFL soldiers had looted vehicles and shops in Harper,

a major town in Liberia’s southeast. Taylor dispatched his Inspector General, Moses

Blah, to Harper to arrest the local NPFL commander, Brooks Gonkarnue, as well

as his deputy, who was known as “Gio Devil.” After apprehending the commanders,

Blah returned the looted items to civilians and remained in town to restore order

(C39).

In addition to spot payments, Charles Taylor offered post-conflict rewards to group

members. When Taylor was elected president in 1997, he provided valuable positions

to all four of his top commanders. The chief of staff of the Marines Division, Melvin

Sogbandi, received a cabinet position in Taylor’s government (C15). The chief of

staff of the Navy Division, Roland Duo, became head of security for the OTC logging

company and gained access to lucrative commercial opportunities (TRC0FB). The

chief of staff of the EMG Division, Prince C. Johnson (not to be confused with INPFL

leader Prince Y. Johnson), was appointed Commanding General of Liberia’s Armed

Forces, while the chief of staff of the Army Division, John Teah, was appointed Colonel

(C24). Lower-level commanders also received positions in the civil service and armed

forces(C26, C27).


6.1.3 Factionalization

Aside from Prince Y. Johnson’s split in the first weeks of the war, the NPFL suffered

another split in 1994 during peace talks in Akosombo, Ghana. The talks were intended

to strengthen the 1993 Cotonou Agreement, which was viewed by the international

community as the best roadmap for peace in Liberia. As part of the agreement,

ministerial positions in the transitional government were allocated to several NPFL

political officials, including the NPRAG’s Minister of Defense Tom Woewiyu, Minister

of Justice Lavala Supuwood, and Minister of Internal Affairs Samuel Dokie. Having

run the NPFL’s civilian administration, these individuals were now ready to accept

their post-conflict rewards in an internationally-recognized government. Charles Tay-

lor, however, was not satisfied with the agreement and seemed intent on resuming the

conflict (C08).

In the hopes of maintaining their positions in the transitional government, Woewiyu,

Supuwood, and Dokie split from the NPFL and formed the Central Revolutionary

Council (NPFL-CRC). To provide force to their split, they recruited one of Taylor’s

top commanders, Nixon Gaye, chief of staff of the Marines Division, and several hun-

dred NPFL fighters. The group fought alongside ECOMOG peacekeepers and various

ULIMO factions against Taylor. The anti-Taylor coalition sparked intense fighting

throughout Liberia but ultimately proved no match for Taylor’s NPFL forces. In

1995, Nixon Gaye, top commander of the NPFL-CRC and former ally of Taylor, was

executed on Taylor’s orders (C19, C24).

6.1.4 Controlling Soldiers

Immediately after entering Liberia, the NPFL constructed a training camp at Butuo,

just across the border from Cote d’Ivoire, and quickly opened two other training

camps at Tiaplay and Gborplay (C24). The official training program lasted three

months. Experienced commanders such as John Teah led the training programs in

the first months, and training exercises were often overseen by Charles Taylor himself.

Recruits spent the first three weeks practicing military maneuvers with sticks before

receiving rifles. Beyond physical exercise and basic drills, recruits were also instructed


in guerrilla tactics and the “Eleven General Principles of Leadership”, developed by

the U.S. Marine Corps.8 Competent recruits received positions as squad leaders after

the completion of training, while the most promising recruits were placed in command

of platoons (C24).

Figures 6.2 and 6.3 depict the NPFL training base in Gborplay in March 1990.

The first image shows NPFL recruits undertaking a military drill, supervised by

Charles Taylor (center foreground, in military fatigues). The second image shows

NPFL recruits lined up outside of a weapons depot, ready to receive rifles after three

weeks of training with sticks.

The demands of war soon forced the NPFL to compromise its three month training

program. A recruit who joined the NPFL in January 1990 received 2 months of train-

ing (C17). A recruit who joined in June 1990 received one month of training (C38).

By September 1990, it was common for training to last only three weeks (B43). Nev-

ertheless, the basic training program remained the same: recruits spent the first few

weeks practicing drills with sticks before receiving weapons, and any failure to follow

orders resulted in beatings or a few days in jail (B43). Despite the declining training

standards, particularly promising recruits were singled out for advanced training and,

in some cases, were sent to Libya to receive specialized instruction (C26).

The NPFL’s recruitment practices also changed over the course of the war. Ini-

tially the NPFL was flooded with recruits in Nimba County who had been tormented

under the Doe regime and viewed the NPFL as a means of protection and retribution.

As the war progressed, the NPFL began abducting some recruits, but mostly relied

on attracting individuals who sought protection from the violence and insecurity of

the conflict areas. In a randomized survey, James Pugel (2007) finds that 18% of the

NPFL sample claim they were abducted by the group, while 35% claim they joined

the group to protect their family; another 20% of respondents said they joined the

group because they were scared.

8The eleven principles are as follows: (1) Know yourself and seek self-improvement; (2) Betechnically and tactically proficient; (3) Seek responsibility and take responsibility for your actions;(4) Make sound and timely decisions; (5) Set the example; (6) Know your soldiers and look out fortheir well being; (7) Keep your subordinates informed; (8) Develop a sense of responsibility in yoursubordinates; (9) Ensure the task is understood, supervised, and accomplished; (10) Build the team;(11) Employ your unit in accordance with its capabilities.


Source: C24, with permission.

Figure 6.2: NPFL recruits undergo military training, supervised by Charles Taylor(center, in military fatigues). Gborplay, March 1990.


Figure 6.3: NPFL recruits wait to receive weapons after three weeks of training withsticks. Gborplay, March 1990.


Commanders were responsible for deploying and monitoring their troops. The

most important element of maintaining discipline was the provision of food. On

occasion, most notably during 1990-91, the NPFL leadership supplied rice to com-

manders from a central depot in Buchanan port, and the commanders distributed this

rice to the soldiers (C39). More commonly, commanders were responsible for organiz-

ing their own system of food distribution, either by collecting food from village chiefs

or by assigning soldiers to collect 50% of food passing through NPFL “gates” in each

town. This system of food distribution was open to abuse. Centralized rice shipments

from the NPFL leadership were occasionally diverted by high-level commanders and

sold in local markets, rather than distributed to troops (C39). And soldiers stationed

at the gates sometimes looted, raped, or killed civilians passing through.

Perhaps the worst abuses were committed against Krahn and Mandingo civilians.

In the early period of the war, the NPFL targeted Krahn and Mandingos to solidify

popular support in Nimba County. Krahn and Mandingos were viewed as collabora-

tors with the Doe regime, and targeting these civilians freed up land and goods that

could be redistributed to NPFL supporters. When the NPFL captured Barkedou, a

bustling Mandingo town in Lofa County, in July 1990, soldiers slaughtered an esti-

mated 387 civilians and beheaded the local imam; NPFL forces also gunned down

dozens of civilians in Mandingo villages such as Selega and Kuruka (TRC 2009: Vol.

II, pp. 219, 248). Even after the NPFL solidified control over most of Liberia, the

Krahn population of Grand Gedeh County continued to suffer high levels of violence

and insecurity (C24). NPFL forces abused and preyed on Krahn and Mandingos with

little risk of punishment from their superiors, but these pogroms were not systemat-

ically coordinated by the NPFL leadership.

Also abusive were the NPFL soldiers positioned on the frontline. These soldiers,

who were constantly on the move, looted food and other valuables as they engaged

with enemy forces and the ECOMOG peacekeepers. Areas of intense fighting did

not allow for standardized tax collection and larger operations were viewed as a

means of rewarding fighters in lieu of a salary. NPFL’s invasion of Monrovia in

1996, for example, was commonly referred to as “Operation Pay Yourself.” These

frontline soldiers were responsible for much of the extravagant brutality of the war


that was depicted in the international media. Intoxicated soldiers paraded the streets

of Monrovia during major battles wearing wedding dresses and halloween costumes,

looting from civilians and spraying the streets with bullets. At roadblocks on the

frontline, some NPFL members developed reputations for killing people in macabre

ways. Common stories include NPFL members that sliced open the bellies of pregnant

women, taking bets on the gender of the unborn child. Other soldiers flaunted their

reputation for brutality by decorating their roadblocks with human skulls.

By contrast, when NPFL forces were based in towns for extended periods, the

soldiers developed relationships with the local communities and generally had good

relations with civilians (B26). In these cases, civilians could report soldier misconduct

to commanders, who would then discipline the offending persons.

6.1.5 Discussion

Charles Taylor became the NPFL’s leader through his ability to attract external

financing from a variety of sources. Taylor sat at the center of a complex network of

international arms dealers, corporations, and businessmen and generated hundreds of

millions of dollars for his campaign against the Liberian government. To complement

these commercial endeavors, Taylor built up a force of perhaps 12,000 NPFL fighters

and a civilian administration that installed police and courts in NPFL areas.

The NPFL’s experience corresponds well to some of the theory’s key assumptions

and predictions. Charles Taylor used his substantial resources from partnerships with

multiple external patrons to provide cash incentives to his commanders, resulting in

increased effort and cooperation (Hypotheses 1, 9, 12). Taylor also encouraged coop-

eration among commanders through his charisma and promises of future rewards—

promises that were credible to some NPFL members because of family ties and his

wife’s ethnicity (Hypotheses 3, 4, 10). Indeed, all of Taylor’s top commanders were

rewarded after Taylor was elected President in 1997.

Not all NPFL members, however, believed in Taylor’s leadership qualities or the

credibility of his promises. In the first weeks of the war, Prince Y. Johnson split from

the group because a promised arms shipment failed to arrive and Taylor’s checkered


past seemed destined to thwart the group’s success. Later in the war, top NPFL

officials split from the group because their ministerial positions in the transitional

government matched any post-conflict reward that Taylor could promise. And despite

the incentives offered by Taylor, NPFL members became distracted by looting in areas

that contained rubber (Hypothesis 2) and were extremely predatory in remote and

unstable areas (Hypothesis 5). The NPFL also suffered a split during peace talks

(Hypothesis 8).

The NPFL’s behavior is only partially explained by competing theories. The

NPFL abuses that occurred in unstable areas correspond to theories of territorial

competition, but these theories cannot explain the abuse in remote areas and in the

vicinity of lootable resources (Competing Hypothesis 1). These abuses also corre-

spond to theories of costly signaling, ethnic composition, and member selection, but

these theories cannot explain the relative discipline of NPFL members in the group’s

headquarters at Gbarnga and in other large towns (Competing Hypotheses 3, 4, 6).

At least two explanations exist for the brutal behavior of some NPFL soldiers.

First, some soldiers could be inherently sadistic, and the NPFL consciously deployed

these soldiers to the frontline areas. Second, the brutality was a response to the stress

and incentives of the war in areas where commanders provided little discipline and

oversight. Interviews with commanders, soldiers, and civilians provide evidence for

both of these explanations. The persuasiveness of the first explanation, however, is

undermined by the changed behavior after the war of many individuals responsible

for these brutal acts. If sadism were inherent in their personalities, one would expect

these individuals to continue the behavior after the war. But in fact, many of these

individuals have returned to their families and communities, and such sadistic acts

are no longer common in Liberia.9

Charles Taylor created a formidable economic and military machine. His success,

however, ultimately encouraged the formation of rival groups, such as the United

9Violent crime, including rape and murder, remain a problem in Liberia but these crimes are notaccompanied by mutilation and there is no evidence to suggest a systematic link to soldiers whocommitted sadistic acts during the war. For a discussion of post-war trends in violence, see SAS(2011). Ritual killings and other forms of “witchcraft” also occur in Liberia, but these practices arelinked to a traditional belief system that predates the war.


Liberation Movement for Democracy (ULIMO).

6.2 ULIMO and its Offspring: Self-Destruction

ULIMO leader Alhaji Kromah came to power by assassinating the group’s initial

leader, who was trusted by the group’s commanders. ULIMO members accepted

Kromah because he offered the best chance for attracting support, but Kromah could

offer few real incentives for cooperation. Commanders engaged in diamond mining at

the expense of military activities and failed to restrain the predatory behavior of their

troops. Many ULIMO members defected from the group to join the LPC, and later

factionalized when Kromah tried to exclude them from positions in a transitional

government.

6.2.1 Formation

ULIMO has its roots in the collapse of the Armed Forces of Liberia (AFL) in the

face of the NPFL and INPFL advances. Thousands of AFL soldiers and commanders

deserted their posts or were evacuated to Sierra Leone by the ECOMOG peacekeepers.

The situation in the refugee camps grew desperate in March 1991, when the RUF

rebels, backed by Charles Taylor, invaded Sierra Leone. The Sierra Leone Army

(SLA) was routed by the rebels’ advance, and the Liberian refugees soon came under

attack. Liberia’s ambassador to Sierra Leone, Albert Karpeh, met with the Sierra

Leone government to discuss how to protect the Liberian refugees. Karpeh was a

former AFL General and had close ties with Liberian soldiers in the refugee camps.

He suggested that the Sierra Leone Army provide weapons to the former soldiers so

they could form self-defense groups and protect the refugees from the brutal rebel

attacks.

Sierra Leone President Joseph Momoh agreed to supply arms to the Liberian

soldiers, but only if the Liberians were led by commanders from the SLA (C13, C18).

Momoh feared that if Karpeh had direct command of the forces, the group would

be too autonomous from the Sierra Leone government and try to re-enter Liberia,

6.2. ULIMO AND ITS OFFSPRING: SELF-DESTRUCTION 185

thereby drawing Sierra Leone even further into Liberia’s mess. So long as the group

lacked strong leadership and command structures, the Sierra Leone government could

maintain influence over its activities (C44). Desperate for support, Karpeh agreed

to the plan and his Liberian forces, known as the Liberian United Defense Force

(LUDF), helped push back the RUF rebels to the Liberian border.

Despite Sierra Leone’s warnings, the LUDF was determined to return to Liberia.

To increase the group’s membership, Karpeh struck an agreement with Alhaji G.V.

Kromah, the leader of a Liberian Muslim organization based in Guinea with many

ethnic Mandingo followers. Although Karpeh and the LUDF commanders were eth-

nic Krahn, the Mandingos were natural allies. The Mandingos had been granted

citizenship and preferential economic opportunities during President Doe’s Krahn-

dominated government, and both groups were now targeted by Taylor’s NPFL forces.

Alhaji Kromah’s followers relocated to Sierra Leone and the combined group changed

its name to ULIMO, with Ambassador Karpeh as leader and Alhaji Kromah as group

spokesman.

The increased manpower gave the group confidence, but Ambassador Karpeh

could not convince the Sierra Leone government to provide support for a return

to Liberia. The government reiterated its requirement that the group remain in

Sierra Leone and work alongside the SLA if it hoped to continue receiving assistance.

Despite these warnings, in late 1991 ULIMO forces crossed the Mano River bridge into

Liberia and attacked NPFL positions. Fierce fighting lasted all night, but a shortage

of ammunition forced ULIMO to retreat. Two dozen ULIMO fighters, including top

commander Free Karla, died in the attack (C12, C18).

The failed operation forced ULIMO to reconsider its plans and restructure its

forces. Alhaji Kromah, taking advantage of the disarray, orchestrated the assassina-

tion of Ambassador Karpeh and installed himself as group leader (C02, C07, C09,

C12, C13, C44). The Krahn commanders in the group, loyal to Karpeh, were shocked

by the move, but eventually accepted Kromah as leader because his international con-

nections offered the best chance for returning to Liberia. The commanders, however,

remained wary of Kromah’s motives and his promises.



When ULIMO crossed into Liberia in early 1992, the group was aware that Sierra

Leone’s support would come to an end. ULIMO’s soldiers had already received

weapons from the Sierra Leone government, but continued access to ammunition

would be a major challenge. In advance of the invasion, the group purchased am-

munition with $5,000 from Alhaji Kromah’s personal finances and some money that

Kromah had gathered from small contributions from the Liberian diaspora (C04, C07,

C41, TRC0A7). Once in Liberia, the group focused its efforts on capturing ammu-

nition from NPFL bases. ULIMO seized large arms caches at Israel Camp and Lofa

Bridge, but the group faced chronic shortages (C12, C13). Despite these challenges,

ULIMO overran NPFL positions and advanced across the northwest.

ULIMO’s forces were divided into two battalions, Alligator and Zebra, with a

complicated command structure that reflected the unsteady coalition of Krahn and

Mandingo forces, depicted in Figure 6.4. Units were initially “mixed” with a balance

of Krahn and Mandingo commanders (C02, C06, C41). But the Mandingo forces

increasingly concentrated themselves in Zebra battalion while the Krahn forces con-

centrated in Alligator battalion. Authority in the group followed mostly along ethnic

lines, with the Mandingo forces reluctant to take orders from the Krahn commanders

who dominated the top military positions (C02, C13).

ULIMO’s access to weapons increased in August 1992 after the group captured the

Po River Bridge, 25km from Monrovia. Within striking distance of the capital city,

ULIMO encouraged AFL soldiers to abandon their posts and join the group. AFL

armories were emptied as soldiers took whatever weapons and ammunition they could

carry and crossed the lines (C11, C13, C16). When the NPFL launched Operation

Octopus in October 1992, a desperate ECOMOG called upon ULIMO for assistance

(TRC0A7). ULIMO fought alongside the peacekeepers to push back the NPFL and

even complied with ECOMOG’s request to withdraw from the city when the fight-

ing subsided. Operation Octopus opened the door to a partnership between Alhaji

Kromah and the ECOMOG peacekeepers, as well as with the Interim Government of

National Unity (IGNU). ECOMOG and IGNU supplied ammunition and other sup-

plies to Kromah to assist ULIMO’s offensives against the NPFL. In exchange, ULIMO


was expected to hand over captured territory to the peacekeeping forces (C07, C41,

C43).

ULIMO controlled a territory rich in diamonds, timber, and rubber, but the or-

ganization failed to benefit from these resources. Diamond and rubber companies

ceased their operations at the start of the war and did not return while ULIMO was

in power. Timber companies in Lofa County approached Kromah to negotiate part-

nerships, but Kromah could not provide protection from looting by ULIMO forces,

even though the local soldiers shared Kromah’s Mandingo ethnicity. The companies

were forced to evacuate (C45).10 Some ULIMO commanders and soldiers benefitted

individually from the resources in ULIMO territory, but none of this revenue reached

the group’s leadership, who faced acute shortages of financing and military supplies.

The diamond fields around Weasua and Lofa Bridge, for example, were controlled by

ULIMO’s top commander, Thomas Karla (C05, C13). Civilians and ULIMO soldiers

shoveled gravel and sifted for diamonds in small alluvial pits. Anyone lucky enough

to find a diamond would try to smuggle it to Monrovia, where the valuable stone

could be sold to a diamond dealer (B45, C13).

In an attempt to gain control over the diamond trade, the leadership appointed

a civilian official, Moses Neneh, to oversee the diamond mines and collect stones

on behalf of ULIMO. After a brief journey to the diamond areas, Neneh returned

to headquarters empty-handed and reported that he would be killed if he forced

soldiers to hand over their diamonds (C13). No other systematic attempt was made

to control the diamond trade, and the lack of resources continued to limit ULIMO’s

organizational capabilities (C07).

In general, ULIMO commanders operated with a high degree of autonomy and

were difficult for the leadership to control. For example, Ousman Konneh, better

known as “Pepper and Salt”, was dispatched to Lofa County with a company of some

100 soldiers to capture arms and ammunition. But rather than returning to base,

10One might wonder why the logging companies did not give Kromah a cash advance so thatKromah could provide spot payments to the troops. The owner of a logging company interviewedfor this project did not believe Kromah could be trusted to live up to his promises without firstdemonstrating his capability to control the troops (C45). This created a Catch-22 in which a cashadvance might have increased ULIMO discipline, but a lack of discipline prevented Kromah fromobtaining the cash advance.


“Pepper and Salt” remained in the area and recruited heavily from the local Mandingo

population, increasing his forces nearly ten-fold to battalion-strength. His troops

preyed on civilians and looted from timber companies, despite calls from ULIMO

superiors to behave with restraint (C13, C44, C45).11

Alhaji Kromah’s relationship with ECOMOG was based on the condition that

ULIMO relinquish territory to the peacekeepers in exchange for ammunition. ULIMO

forces, however, became increasingly reluctant to withdraw from newly captured ar-

eas. In 1993, frustrated with ULIMO and anxious to increase military pressure on

Charles Taylor, ECOMOG encouraged the formation of a new rebel group, the Liberia

Peace Council (LPC), led by George Boley, a former member of Doe’s government,

then living in the United States. George Boley had previously tried and failed to

secure support for a rebellion. Between 1990 and 1993, Boley petitioned for meet-

ings with representatives from Burkina Faso, Togo, Cote d’Ivoire, and other regional

states but was denied any hearing (TRC160).

After receiving ECOMOG support, the LPC’s ranks were filled by ethnic Krahn

members of ULIMO and AFL who deserted their posts during a brief ceasefire. These

forces were recruited by LPC chief of staff Teh Quiah, a highly trained member of the

AFL, who lacked any personal ties to George Boley (TRC163). The LPC’s mission

was to capture the valuable port at Buchanan as well as the timber-rich areas of the

southeast, which supplied most of Charles Taylor’s revenues, and then relinquish these

areas to the peacekeeping troops. George Boley had no access to financial resources

and had never met most of the commanders in the group. Like ULIMO, LPC forces

preyed on civilians and were only loosely controlled by the LPC leadership.

Alhaji Kromah and George Boley lacked the resources to adequately supply their

troops, so ULIMO and LPC commanders traded directly with ECOMOG commanders

to meet their needs. The Mandingo commanders in ULIMO established relationships

with Guinean ECOMOG forces, while the Krahn commanders partnered with the

Nigerian peacekeepers (C11, C43). In Kakata, for example, ULIMO commander

Philip Wlue, better known as “General Rambo”, traded rubber and looted items for

11Nearly all civilian informants in areas controlled by “Pepper and Salt” reported widespreadabuse and looting (B08, B11, B12, B13, B14, B15, B19, B22, B25).


ammunition from Colonel Onia, a Nigerian ECOMOG commander (C09). A valuable

item like a generator might fetch 3-4 boxes of ammunition (C41). A truckload of

rubber would earn $300 (Ellis 1999: 167). In the southeast, the LPC purchased

ammunition with timber looted from logging companies, as well as with rubber and

gold (C42; Ellis 1999: 167; Reno 1999: 105). LPC forces in Buchanan supposedly

earned $45,000 by selling looted iron ore (C07).

Commanders did not expect to receive rewards from the leader after the war

ended. No ULIMO commanders received positions in post-war governments from

Alhaji Kromah, although some members of ULIMO were later incorporated into

Charles Taylor’s security forces (C18). The LPC’s chief of staff, Teh Quiah, fared

better than his ULIMO counterparts. LPC leader George Boley appointed Quiah

as Deputy Minister of Defense during a short-lived transitional government in 1995.

This reward, however, resulted from Quiah’s influence over the LPC’s soldiers rather

than a credible promise from Boley.


Tensions between Alhaji Kromah and his top commanders came to a head in 1994 over

positions in a transitional government. The Cotonou agreement allocated two seats

on a five-member executive council to ULIMO. These seats were given to Thomas

Ziah, an ethnic Krahn, and Mohamed Sheriff, a Mandingo. Distrustful of Kromah,

the Krahn commanders wanted Ziah to form a coalition with the civil society rep-

resentative on the council (C02, C12). Alhaji Kromah, on the other hand, wanted

ULIMO’s representatives to vote together and elect a Mandingo as chairman of the

council (C06, C13, C44).

To solidify his power, Alhaji Kromah ordered his Mandingo forces to disarm the

Krahn military commanders at ULIMO’s headquarters in Tubmanburg. Commanders

“Pepper and Salt” and “Jungle Jabah” led the operation, which quickly descended

into fighting (C02, C05, C06). Top commander Thomas Karla was severely wounded

and company commander Sarkoh Selee, among others, was killed (C02, C09, C12).

The Krahn forces fled Tubmanburg and made their way through the bush to Bong


Mines, which was controlled by the Krahn commander Amos Chayee.

The Krahn forces regrouped under the leadership of Roosevelt Johnson and be-

came known as ULIMO-J, while the Mandingo forces remained under Kromah’s com-

mand and became known as ULIMO-K. Intense fighting erupted between the two

factions as ULIMO-J forces recaptured Tubmanburg and nearby diamond fields, but

failed to solidify their control over the surrounding area. For the remainder of the

war, ULIMO-J and ULIMO-K fought each other fiercely, at the expense of the civilian

population.


ULIMO attracted its initial recruits in the Liberian refugee camps of Sierra Leone.

Those who lacked military experience received one month of basic training at a mili-

tary base in Kenema, before deploying to fight the RUF rebels alongside SLA forces.

Training was conducted by ULIMO commanders who had previous experience in

the AFL (C06, C44). During intense fighting, training was shortened to 2-3 weeks

(C06, C48). After ULIMO crossed into Liberia, the main training base moved to

Mano River Kongo, with additional training facilities at the ULIMO headquarters in

Tubmanburg.

ULIMO’s centralized training program was undermined by group commanders,

who attracted their own recruits as a way of increasing their power and influence in

the organization. The most extreme example was “Pepper and Salt”, who recruited

hundreds of untrained men after reaching Lofa County (C02, C41). Such decentral-

ized recruitment was common practice among ULIMO’s commanders and quickly

deteriorated the military standards in the organization (C09).

Soldier discipline was also compromised by a lack of organized food distribution

and commander oversight. While some commanders collected food from village chiefs

and distributed the supplies to their soldiers, most commanders let soldiers fend

for themselves (B18, C09). Even when civilians reported soldier abuse, comman-

ders rarely punished soldiers for misbehavior.12 Abuse was widespread throughout

12A ULIMO commander claimed that he once executed a soldier for looting $1,500 (C09), butmost civilian informants claimed they never witnessed a ULIMO soldier punished for misconduct


ULIMO territory. The Truth and Reconciliation Commission accuses ULIMO of

looting, killing, and destruction throughout their territory, including in towns such

as Tienne, Goe, and Bangoramah. In July 1993, ULIMO soldiers allegedly killed

some 300 ethnic Lorma civilians in Voinjama (TRC 2009: Vol. II, p. 227-8). After

ULIMO’s split in 1994, large swaths of ULIMO territory were destabilized by the

competing ULIMO-J and ULIMO-K forces, and soldiers relied even more heavily on

looting as a means of acquiring food.

LPC soldiers were also required to fend for themselves, although units sometimes

designated a “Mess Sergeant” to coordinate food supplies and cook the looted food

(C42). This autonomy from the command structure allowed many LPC soldiers to

become involved in communal violence. In Sinoe County, many ethnic Sarpo joined

the LPC to exact revenge on their Kru neighbors, whom they accused of looting and

abuse during NPFL’s rule. The larger conflict also intensified family feuds dating

back to the 1930s, and the LPC provided an opportunity to settle old scores (Ellis

1999: 129). In this way, the objectives of the armed group were manipulated to serve

the private interests of its members (Kalyvas 2006).13

The ill-disciplined behavior of ULIMO soldiers cannot be explained by their eth-

nic diversity. After ULIMO’s split, soldiers reorganized into ethnically homogeneous

units, but this increased homogeneity did not result in greater discipline (C04).

ULIMO-J and ULIMO-K soldiers continued to prey upon civilians and faced lit-

tle risk of punishment from their co-ethnic commanders. The LPC was dominated

by exclusively Krahn units but also struggled to control the predatory behavior of

its troops. Rather than ethnicity, soldier behavior resulted from poor military train-

ing and the failure of commanders to organize food supplies and discipline soldier

misconduct.

(B08, B18, B22, B25, B51, B54). Only one informant, who was living in ULIMO’s headquartersin Tubmanburg, claimed that ULIMO officials would punish soldiers after receiving reports of badbehavior (B50).

13Some observers believe the LPC used gratuitous violence as an intentional strategy to depopulatethe southeastern region. Forced to flee their homes by the predatory LPC, civilians sought shelterin key points held by ECOMOG (Ellis 1999: 103). This strategy increased ECOMOG’s leverage,while preventing the NPFL from recapturing the depopulated areas. This strategy was costly forthe LPC, as the group was unable to generate revenue or secure food in its territory. But the LPC’scampaign of chaotic violence ensured a continued source of arms from its ECOMOG allies.


6.2.5 Discussion

ULIMO and its spin-offs were unable to control the opportunistic behavior of their

members, and the groups’ experiences fit well with the theory of commander incen-

tives. Although Alhaji Kromah was not installed by an external patron, he was

distrusted by his commanders and could provide few incentives for cooperation (Hy-

potheses 1, 3, 4). Many ULIMO commanders devoted their attention to mining and

looting rather than carrying out offensives against the NPFL (Hypothesis 2). While

Charles Taylor and the NPFL made millions from timber and rubber, ULIMO’s in-

ability to control its forces prevented the group from generating any revenue. Lacking

partnerships with foreign firms or governments, ULIMO remained dependent on the

whims of ECOMOG commanders whose preferences diverged from those of the group

(Hypothesis 9).

After ULIMO’s initial advance to Po River, the group stagnated and failed to cap-

ture new territory. Alhaji Kromah could not prevent his troops from deserting to join

LPC, and his lack of credible promises triggered a split in the group during the 1994

peace process (Hypothesis 8). Frustrated with ULIMO, ECOMOG recruited George

Boley to lead the LPC, despite his lack of qualifications (Hypothesis 10). Boley pro-

vided little to his troops, who preyed on civilians, but rewarded his top commander

with the position of Deputy Minister of Defense in a short-lived transitional govern-

ment. This reward, however, reflected the commander’s power in the group, rather

than a credible promise, and did not motivate the commander to train, discipline, or

feed LPC forces during the war.

ULIMO’s behavior challenges some of the competing theories, but fits well with

others. Theories of territorial stability cannot explain the widespread predation in

ULIMO territory (Competing Hypothesis 1), while theories of ethnic composition

cannot explain the predatory behavior of the ethnically-homogeneous forces in LPC,

ULIMO-K, or ULIMO-J (Competing Hypothesis 4). Theories of costly signaling

perform better, since the various ULIMO factions faced some pressure to prove their

worth to their patron ECOMOG (Competing Hypothesis 3). The theory of member

selection also does well (Competing Hypothesis 6). Although ULIMO had fewer

resources than NPFL, the group’s access to diamonds and loot may have attracted

6.3. LURD: DISTRUST AND DYSFUNCTION 193

opportunistic members who were difficult for the leadership to control, as indicated

by the leadership’s failed efforts to generate revenue from the diamond trade.

ULIMO and its offspring were a nuisance for Taylor’s NPFL, but ultimately failed

to prevent his rise to power. Within a few years, former ULIMO commanders would

regroup and try again to defeat Taylor, this time under the banner of Liberians United

for Reconciliation and Democracy (LURD).

6.3 LURD: Distrust and Dysfunction

LURD leader Sekou Conneh was appointed by the president of Guinea, LURD’s

only source of support. Conneh could offer few incentives to his commanders, who

distrusted him. LURD forces preyed on civilians and fought amongst themselves for

valuable loot. The group nearly split over the allocation of positions in the post-

conflict government, but remained unified due to pressure from the United Nations.

6.3.1 Formation

Soon after the 1997 elections brought Charles Taylor to power, Liberian exiles began

seeking support to overthrow his repressive, although democratically elected, regime.

A violent crackdown on Roosevelt Johnson and his supporters on 18 September 1998

sent a wave of former ULIMO members to refugee camps in Sierra Leone, Guinea,

and Cote d’Ivoire. Prominent dissidents formed an “executive council” in Guinea and

began contacting former ULIMO commanders and searching for support for a new

rebellion. The council elected businessman Mohammed Jumandy as its leader based

on his promises of securing $1 million in financing for the group, but soon removed

Jumandy when these funds failed to materialize (C09, C46, Brabazon 2003).

After Jumandy’s exit, other members of the council jockeyed to position them-

selves as leader, each with some base of support (C30). Ethnic Mandingos in the group

supported Kabineh Ja’neh, a respected Mandingo lawyer, while the ethnic Lormas

followed Lavala Supuwood, who was Minister of Justice in Charles Taylor’s NPRAG

government before he split off to form NPFL-CRC. The ethnic Krahns were divided


among Jackson Doe, the runner-up in the rigged 1985 presidential elections, and two

relatives of the late President Samuel Doe—his brother Chayee Doe and his cousin

George Dweh. Other members of the council, such as the U.S.-trained psychologist

Thomas Yaya Nimely, had smaller groups of followers and struggled to gain influence

in the group.

The council’s initial attempts to secure support were unsuccessful. The President

of Guinea, Lansana Conte was not interested in supporting a rebellion (C46), and the

government of Cote d’Ivoire remained an ally of Charles Taylor. The situation looked

more promising in Sierra Leone, where small groups of Liberian ex-soldiers were fight-

ing to push back the RUF rebels and receiving weapons through ECOMOG General

Maxwell Kobie (C10, C29). Kabineh Ja’neh and several other council members went

to Freetown to try to recruit these forces and secure financing and military supplies.

The newly-reinstated Sierra Leone government, however, did not want another insur-

gency launched from its territory and arrested the men for mercenary activities (C02,

C03, C18, C30). After more than a month in prison, Ja’neh and the others returned

to Guinea to craft another plan.

LURD’s fortunes brightened in 2000 when a group of Taylor-backed rebels at-

tacked southeastern Guinea and briefly captured the town of Guekedou. Liberian

ex-soldiers in the refugee camps received weapons from local Guinean military com-

manders and assisted in recapturing the town and defeating the insurgency. Swayed

by these events and hoping to prevent a repeat, Guinean President Lansana Conte

agreed to provide military assistance to the LURD rebels. This assistance, how-

ever, would be provided on the condition that Sekou Damate Conneh—a former used

car salesman with no political or military experience—become the chairman of the

executive committee (C09, C29, C46). President Conte selected Sekou Conneh as

leader because of Conneh’s wife, Aisha, who was the President’s spiritual advisor and

adopted daughter.14 Sekou Conneh, unlike the other members of the committee, did

not have a base of support among the members.

The executive council was not happy with Sekou Conneh’s appointment as leader,

14Aisha was initially married to Sekou Conneh’s brother. She married Sekou Conneh after herhusband’s death, according to the traditional custom. Aisha became President Conte’s spiritualadvisor in the 1990s after warning him of a coup attempt based on a dream she had.


but most of LURD’s members grudgingly accepted him as the only means of receiving

external support (C07, C09, C10). Some members plotted to remove Conneh from

power as soon as the group reached Monrovia. Others, such as Thomas Yaya Nimely,

left the group in frustration and tried to attract support for their own insurgency

(C29). With military support and a force of former ULIMO soldiers, LURD crossed

into Liberia in July 2000 and briefly captured Voinjama. The incursion was quickly

repulsed by Taylor’s forces, which then ransacked neighboring towns in Guinea (A02).

President Conte increased the military support for the group and, after several false-

starts, LURD expanded beyond its headquarters in Voinjama in 2002 and captured

much of northern Liberia.


Aside from a few weapons captured from government forces, LURD relied exclusively

on weapons and ammunition provided by the Guinean government. These shipments,

though often infrequent, gave LURD a weapons advantage over Taylor’s government

forces, which struggled under a U.N. arms embargo. The Guinean Ministry of Defense

was directly involved in supplying LURD with weapons and ammunition, which were

relatively inexpensive for Guinea due to the military aid provided by the United States

(ICG 2002a). Military supplies were diverted to LURD either through the Guinean

military base at Macenta or through Katex Mining Company, a Conakry-based firm

with connections to President Lansana Conte (UNSC 2003a: 30).

LURD received these weapons at its headquarters in Voinjama, where the muni-

tions were transported to the various front lines.15 Delivering supplies to the forces at

Tubmanburg required an arduous 4-5 day trek through the jungle and relied heavily

on forced labor. On at least one occasion, the Guinean military sent munitions by

ship to LURD forces in Robertsport. Guinean assistance also included occasional

artillery support and medical attention for wounded soldiers.

The LURD rebels did not have to pay for Guinean military supplies, but the group

did have to fulfill certain obligations to its patron. Guinea insisted that Sekou Conneh

15Information for this paragraph from B17, B26, B28, C01, C03, C10, C14, C16; HRW 2003b;UNSC 2003a.


remain leader of the group, despite the tensions this caused among LURD’s military

and political leadership. LURD was also expected to secure the border and prevent

any repeat of the 2000 rebel attacks in Guinea. By keeping Charles Taylor tied up

militarily, LURD became, in essence, a paramilitary for the Guinean government.

LURD’s top commander was its chief of staff, Charles Dent, an ethnic Krahn and

former ULIMO commander. When Dent was killed in an ambush in 2001, the top

military position passed to Prince Seo, another Krahn ULIMO commander. LURD’s

forces were organized into a primary battalion under the command of Oforie Diah, also

Krahn, and a number of smaller forces that lacked a clear command structure.16 For

most of the war, LURD deployed its troops in small squads that were constantly on

the move. These squads attacked and retreated in a “hit and run” strategy designed

to confuse and over-extend the government forces (C02). Only in larger towns did

LURD maintain a constant presence. Figure 6.5 depicts LURD’s command structure.

Before the invasion, at least four international timber companies were active in

areas that would later fall into LURD hands (UNSC 2001: 73). The group’s territory

also included the 100,000 acre Guthrie rubber plantation. Partnerships with these

firms would have generated enormous revenue for LURD and increased its access to

military supplies. But these companies were skeptical of LURD’s ability to provide

security and, almost immediately after LURD’s invasion, a series of incidents in which

LURD soldiers looted property and machinery forced all companies to cease opera-

tions in LURD areas (UNSC 2003a: 13). Timber and rubber remained unexploited

throughout the war.

LURD territory also included the diamond fields near Lofa Bridge, Kongo Camp,

and Weasua. Upon capturing these areas, LURD soldiers went to work shoveling

gravel and sifting for diamonds. Stones found in these areas were smuggled out of

Liberia and sold to Guinean dealers in Conakry (UNSC 2001: 79). Diamond revenue

would have provided LURD with greater autonomy from Guinea (UNSC 2001: 34).

The LURD leadership, however, could not gain control over the diamond trade and

16LURD members typically claim that the forces were divided into two battalions, but none of theinformants could provide consistent information on who led the second battalion or how its forceswere organized.


Alhaji G.V. Kromah

Alligator Battalion Zebra Battalion

Lofa Company(Battalion-Strength)

Commanding GeneralThomas Karla

Field CommanderJoe Harris

Military ChairmanRoosevelt Johnson

Prince Seo, a.k.a “General Crazy Devil”

Mohammed Jabateh, a.k.a “Jungle Jabah”

Ousman Konneh, a.k.a “Pepper and Salt”

Note: Arrows indicate chain of command and influence. Dotted arrows indicate weaker influence.

Figure 6.4: ULIMO command structure.

Sekou Conneh

Primary Battalion

Chief of Staff

Prince Seo, aka “Crazy Devil” (2001-3)

Oforie Diah

Various Forces

Charles Dent (2000-1)

Seeya Sheriff, aka “Cobra” (2003)

(No clear command structure)

Note: Arrows indicate chain of command and influence.

Figure 6.5: LURD command structure.


worried that diamond mining would distract their troops from fighting. Senior LURD

officials issued a ban on diamond mining in LURD territory, but this did little to

discourage the motley crew of soldiers and civilians in the diamond fields (B48, C30,

C32; ICG 2002: 7; Brabazon 2003).

With no money or food provided by the organization, LURD soldiers and com-

manders preyed on civilians. LURD members stripped buildings of their zinc roofs

and electrical wiring, and looted anything of value. Large tracts of the countryside

were depopulated as civilians fled to more secure areas. Facing severe food shortages,

the LURD leadership in Voinjama attempted in mid-2001 to provide security for civil-

ians by executing at least one soldier and providing a “political education” to some

of the forces; these efforts failed to stem the predatory behavior of the troops (ICG

2002: 9-10). Predation was so pervasive that LURD members sometimes fought each

other over valuable looted items. The most notorious incident occurred in early 2003

at LURD’s military base in Tubmanburg. Two high-level commanders, Mohamed

Kamara, aka “K-1”, and Musa Kamara, aka “Black Marine”, both claimed owner-

ship over a looted vehicle. “K-1” resolved the dispute by killing “Black Marine”; he

was not punished by the LURD leadership (C03, C10, C46).17

A high level of distrust characterized the LURD organization, and top comman-

ders knew that Sekou Conneh’s promises were not credible. In advance of LURD’s

invasion of Monrovia in June 2003, Sekou Conneh ordered the Guinean military to ar-

rest the group’s top commander, Prince Seo, because Conneh feared the commander

would become too powerful if he captured the capital city (C2). Seo was incarcer-

ated in Macenta, Guinea, while LURD forces repeatedly attacked Monrovia under the

command of Seeya Sheriff, aka “General Cobra”, a Mandingo commander whom Con-

neh viewed as less of a threat to his position. After Charles Taylor was forced from

17Various conspiracy theories surround this event. Other versions of the story claim that “BlackMarine” was killed because he often argued with Conneh and refused to implement orders (C09,C14, C18), or because “Black Marine” was going to tell LURD members that Sekou Conneh hadreceived some money for the group but was keeping the funds for himself (C16, C18, C30). Thesestories may have some truth in them and reveal the tensions that existed between Conneh andLURD commanders, even those that shared Conneh’s Mandingo ethnicity. But the most crediblereports indicate that the immediate cause of “Black Marine’s” death was a dispute over a lootedvehicle.


power, Sekou Conneh filled LURD’s positions in the transitional government with

family relatives and cronies; none of LURD’s top commanders received positions in

the post-war government.

Intense insecurity and predation prevented resource extraction in LURD territory

during the war, but the situation changed with the deployment of peacekeepers.

Before disarming in late-2003, a LURD commander, Sumo Dennis, relocated to the

Guthrie Rubber Plantation, along with some 260 soldiers. This group was later joined

by former combatants from MODEL and the government forces. The plantation was

run by Sumo Dennis and a handful of other commanders, who formed the High-

Powered Ruling Council (HPRC). An army of ex-combatants harvested rubber, which

the HPRC sold illegally to the Firestone corporation. Mid-level managers in this

operation received salaries of approximately $1,000 per month, with much higher

rewards for HPRC members (C11, C16). The ex-combatants occupied the Guthrie

plantation until U.N. peacekeepers forced their eviction in 2006.

The deployment of peacekeepers also allowed LURD commanders to profit from

the rubber trade in Bong County. In late-2003, a number of small, family-owned plan-

tations resumed operations in the area around Totota. Trucks filled with rubber began

traveling along Liberia’s main highway, destined for Firestone buying stations near

Monrovia. In the months leading up to the U.N. disarmament program, LURD com-

manders demanded taxes of $300 for each rubber truck that passed through LURD

territory and into government-controlled areas. Even with a pass, truck drivers had

to hand over small payments at roadblocks, but no longer feared that their trucks

would be looted (B37, B39).


Although the group remained unified during its three-year military operation, LURD

faced a split in early 2004 after the deployment of U.N. peacekeepers. Sekou Con-

neh filled LURD’s positions in the transitional government with his relatives and

allegedly sold some of the offices to the highest bidders (C30, C46; IRIN 2004a).

Forty LURD commanders signed a petition for a change in leadership and threw


their support behind Sekou’s wife, Aisha Conneh, who promised greater rewards for

the commanders (C30; IRIN 2004a). The United Nations and Liberia’s Transitional

Government, however, were unwilling to recognize a change in LURD leadership,

since Sekou Conneh had signed the peace agreement on behalf of LURD (C46). After

several weeks, the attempted coup failed and Sekou Conneh maintained his position

as group leader. LURD commanders and soldiers briefly revolted again in April 2004

to protest Conneh’s broken promises, but eventually handed their weapons over to

the United Nations (C01; IRIN 2004d).


LURD failed to provide any significant training for its forces. Recruits received one

week of military training, if they received any training at all (C10). Most recruits

learned the basics of military life by following around more experienced soldiers, rather

than in a structured training program. As one LURD soldier explained, “If you were

smart, you could pick up what you needed to know” (C14). Soldiers did not receive

much feedback on their military performance, but harsh punishments could be meted

out arbitrarily by commanders.

LURD relied mostly on voluntary recruitment throughout the war. But while

LURD abducted only some of its recruits, most joined the group to gain some pro-

tection from its predatory activities. James Pugel’s (2007) survey finds that 22% of

LURD respondents claim they were abducted by the group, while 34% claim they

joined the group to protect their family; another 20% of respondents said they joined

the group because they were scared. According to these figures, LURD recruitment

practices were nearly identical to those of the NPFL.

Commanders did not organize food distribution for their soldiers, who were left

to fend for themselves. LURD soldiers did not share many ethnic ties, but social

bonds formed within LURD squads and platoons. In some cases, LURD platoons

took on the appearance of sports teams, with each soldier wearing a similar T-shirt

and chanting songs in unison as they marched. Figures 6.6 and 6.7 depict the close

bonds that existed among members of LURD platoons. Commanders, however, failed


to use these bonds to create social incentives for disciplined behavior among their

troops. Reports of LURD abusiveness, including looting, rape, and killing, were

nearly universal among the civilian informants for this project, and are supported by

the Truth and Reconciliation Commission report (TRC 2009).

6.3.5 Discussion

The predation in LURD territory hurt not only civilians but also the organization

itself, and this suboptimal behavior can be explained by the theory of commander in-

centives. Despite his lack of credibility or experience, Sekou Conneh became LURD’s

leader due to the group’s dependence on Guinea for support (Hypotheses 9 and 10).

Sekou Conneh, who had access to guns but no financial resources or credible promises,

provided few incentives for commander cooperation and had little alternative but to

allow his forces to loot food and other goods (Hypotheses 1, 3, 4). This predation

occurred even in vicinity of the group’s headquarters, and control was further weak-

ened in areas that contained lootable resources (Hypotheses 2, 7). Only after the

deployment of peacekeepers could LURD members earn profits from the rubber areas

they controlled.

LURD commanders plotted to remove Sekou Conneh from power, but found little

opportunity during the war due to the group’s reliance on Guinean support. Cut-

off from the spoils of the peace process, the commanders staged a last-minute coup

attempt against Conneh (Hypothesis 8), but failed to gain international support for

their efforts. Facing a robust peacekeeping force, the commanders had little choice

but to disarm.

Competing theories have mixed success when applied to LURD’s behavior. As

with ULIMO, theories of territorial competition cannot explain the predation that

occurred in stable parts of LURD territory (Competing Hypothesis 1). Theories of

costly signaling, ethnic composition, and member selection do much better. LURD

may have used predation as a signal to Guinea that the weapons were being used to

hurt Taylor’s regime (Competing Hypothesis 3). Alternatively, predation could have

been the result of ethnic diversity, which hindered monitoring and coordination, or



Figure 6.6: A LURD platoon chants a song while on the move, 2003.


Figure 6.7: A squad of LURD soldiers poses for the camera, 2003.

6.4. MODEL: ALMOST DOESN’T COUNT 203

the selection of opportunistic members who sought looting opportunities (Competing

Hypotheses 4 and 6).

After years of halted offensives, LURD broke out from Lofa County and reached

the outskirts of Monrovia. But before the group could initiate its final assault on Tay-

lor’s government, the MODEL rebels emerged and rapidly expanded across Liberia’s

southeast.

6.4 MODEL: Almost Doesn’t Count

MODEL leader Thomas Yaya Nimely was virtually unknown by his troops but man-

aged to secure support from the government of Cote d’Ivoire. Yaya Nimely could

offer few incentives and was distrusted by some of the group’s top commanders. Al-

though the common ethnicity among group members increased discipline in some

areas, tensions existed in the group and MODEL forces were often predatory. During

the group’s six-month lifespan, Yaya Nimely faced two challenges to his leadership

but managed to stay in power due to external intervention.

6.4.1 Formation

Thomas Yaya Nimely left LURD’s executive council to seek financing for his own

insurgency. He was accompanied by two experienced fighters, Arthur Baygboe and

Philip Pardea, but had few other connections to soldiers. Yaya Nimely made the

rounds in Abidjan, petitioning businessmen to support his cause, but found no sup-

port (C32). Although the 2000 elections in Cote d’Ivoire brought to power a gov-

ernment that was less friendly to Charles Taylor’s regime, newly-instated Ivorian

President Laurent Gbagbo did not want to damage regional relations. When mem-

bers of the government heard about Yaya Nimely’s appeals for support, they asked

Yaya Nimely to leave the country (C29). Yaya Nimely, a U.S.-trained psychologist,

returned to his home in Philadelphia, while Arthur Baygboe and Philip Pardea re-

mained in Cote d’Ivoire.

The opportunities for starting an insurgency increased in September 2002 after a


coup attempt triggered a civil war in Cote d’Ivoire. The Ivorian government faced

rebels supported by Charles Taylor and, once again, Liberian refugees were caught

up in the violence. The “Peace Town” refugee camp outside Guiglo, suffered attacks

from both rebel and government forces and young men were forcibly conscripted to

fight (A01, C23, C25; HRW 2003a). One of the Liberian soldiers in the refugee camp,

Eric Digbeson, walked to the nearby Ivorian military base in early December 2002

and met with Ivorian commander Mohammed Doumuya to ask for support (C23,

C25, C31). Digbeson was soon joined by other Liberian soldiers, including Arthur

Baygboe and Philip Pardea.

After two days of discussions, the Ivorian military provided uniforms and weapons

to some 75 Liberian soldiers on the condition that they obey orders from Ivorian

commanders and fight alongside the government forces (C23, C25, C29, C31). The

group, led by Digbeson, became known as Liberians Mobilized in Action (LIMA) and

was immediately deployed to the frontline in the nearby town of Blolekin; Digbeson

was killed in action four days later (C23, C31). Arthur Baygboe assumed command

of the group, and LIMA pushed the rebels back to Toulepleu.

LIMA’s activities encouraged Thomas Yaya Nimely to return to Cote d’Ivoire.

Before leaving his home in the United States, Yaya Nimely convened several meet-

ings of the Liberian diaspora community, but failed to secure support for an armed

group under his leadership (C29). Once in Cote d’Ivoire, Yaya Nimely focused his

attention on expanding LIMA’s access to Ivorian military supplies, with the goal of

using these supplies to invade Liberia. The government agreed to increase its support

for the LIMA militia, but reiterated that the group was expected to operate along-

side government forces. LIMA’s membership group and the group was soon joined

by Amos Chayee, a former ULIMO commander who made his way to “Peace Town”

from the refugee camps in Ghana.

With a command structure in place, the group assumed the name MODEL and

began stockpiling the military supplies provided by the Ivorian government. By the

time MODEL crossed into Liberia in March 2003, LURD was already threatening

the capital city. MODEL’s strategy emphasized rapid military victories to catch up

with its rebel competition. MODEL forces moved together to take large towns, and


few soldiers were left behind to stabilize captured areas. Taylor’s government forces,

preoccupied by fighting LURD in the northwest, put up little resistance to MODEL’s

advance, except for intense battles in the critically important towns of Tappita and

Buchanan. By late-July 2003, MODEL controlled three ports, including Buchanan,

and vast timber concessions.


After invading Liberia, MODEL continued to rely on support from the Ivorian gov-

ernment. MODEL soldiers in Cote d’Ivoire received supplies from the Ivorian military

and then transferred the munitions to the front lines in Liberia. Although these sup-

plies were provided free of charge, they came at a heavy cost: many of MODEL’s

soldiers had to remain in Cote d’Ivoire to continue fighting alongside government

troops (C22, C23, C25, C32). Perhaps 20% of MODEL’s forces stayed behind in

Cote d’Ivoire during the initial invasion.18 Any assistance received from the Ivorian

government had to be divided between MODEL’s forces in Cote d’Ivoire and those

fighting in Liberia. For example, if five boxes of ammunition were delivered to the

Liberian forces, two boxes would be allocated for fighting in Cote d’Ivoire while the

remaining three boxes would be shipped to the forces in Liberia (C23).

This division of labor created tensions within the group. At one point, the Ivorian

military called on MODEL to participate in an attack on the strategically impor-

tant town of Man. Philip Pardea, MODEL’s deputy chief of staff, gathered some

150 fighters and left MODEL’s base to join up with the government forces. This

action contradicted orders from Yaya Nimely, who wanted Pardea to remain near

MODEL’s headquarters because the operations threatened Yaya Nimely’s control over

the group’s military supplies. Yaya Nimely dispatched troops to prevent Pardea’s de-

parture, without success. Pardea later surrendered to French peacekeepers to avoid

reprisals from the rebel leader (C25, C29, C31).

18The relative number of soldiers who remained in Cote d’Ivoire is difficult to estimate, but twoinformants indicated that MODEL initially fielded four companies in Liberia, while one companyremained in Cote d’Ivoire (C22, C23). The percentage of forces in Cote d’Ivoire likely decreasedover the course of the war as MODEL recruited within Liberia and ultimately formed as many asten companies of 100-300 men each (C20, C31).


MODEL’s top commanders were Amos Chayee, who had close ties to many of the

troops, and Arthur Baygboe, who shared ties with Yaya Nimely. Also influential was

Thomas Baygboe, aka “Boi Bloaju Boi”, the group’s military spokesman and brother

of Arthur Baygboe. The forces were organized into 7-8 company-sized detachments of

100-300 men each. The most prominent company commanders in Liberia were Paye

D. Dowah, aka “Garang”, Junior Vourjolo, and Kai Farley, who assumed the position

of Commanding General when Arthur Baygboe was killed in fighting in Buchanan

in August 2003. MODEL’s forces in Cote d’Ivoire were led by “Alpha One”, under

the supervision of Amos Chayee and the group’s leadership. Although all of these

commanders shared Yaya Nimely’s Krahn ethnicity, none except Arthur Baygboe had

any pre-war ties with the leader. Figure 6.8 depicts MODEL’s command structure.

With the capture of Pleebo and Harper in May 2003, MODEL captured valu-

able timber concessions as well as a border crossing and port where timber could be

exported.19 Harper’s port was managed by two timber companies, Maryland Wood

Processing Industries and Togba Timber Company, which exported round logs and

sawn timber from the port. MODEL’s continued advance created other opportu-

nities for corporate partnerships. In June 2003, MODEL captured Greenville port

and the facilities of the Inland Logging Company, and in July the group captured

Buchanan port and the offices of Liberia’s largest timber company, Oriental Timber

Company (Global Witness 2003b: 21). These logging companies, each with connec-

tions to Taylor’s government, were understandably nervous about the rebel advance

and evacuated before MODEL’s arrival.20 MODEL responded by looting several of

these operations, causing, for example, an estimated $4 million in damage to the

Togba Timber Company (TRC 2009: Vol. III, Title III, p. 22).

Despite their connections to Charles Taylor, some of these companies were open

to a new rebel partnership, but did not find a suitable partner in MODEL’s leader,

Thomas Yaya Nimely. In the hopes of finding a stronger leader, the President of Mary-

land Wood Processing Industries (MWPI), Abbas Fawaz, contacted former ULIMO-J

19The border town of Pleebo provides easy land access to the large port at San Pedro, Coted’Ivoire.

20These timber companies facilitated arms shipments for the Taylor government and also providedsubstantial cash payments to Taylor and his close associates. See Global Witness (2003a).


leader Roosevelt Johnson, who shared close ties with MODEL’s top commanders and

had aspirations of seizing control of the group (C29, C31). Fawaz reportedly provided

some vehicles and supplies to the group and transferred some $30-40,000 to Johnson

as an initial show of good faith.21

Yaya Nimely responded to this coup attempt in July 2003 by having the Cote

d’Ivoire government attempt to arrest Johnson along with MODEL’s chief of staff,

Amos Chayee, and other commanders who might favor Johnson as leader. Ivorian

security forces raided the hotel in Abidjan where Johnson was staying, triggering a

firefight with Johnson’s security guards. Johnson escaped and returned to exile in

Nigeria, while other members of his group were arrested (C29). The deal with MWPI

collapsed and MODEL failed to establish any commercial alliances.

MODEL commanders received some food and supplies from Yaya Nimely’s re-

lationship with the Ivorian government. But commanders did not receive any cash

incentives and engaged in systematic looting for profit in larger towns such as Harper

and Buchanan. Most commanders in the group had little expectation of future re-

ward, but even these low expectations were disappointed during the peace process.

When Yaya Nimely allocated MODEL’s positions in the transitional government in

early 2004, none of MODEL’s commanders received a position.

Similar to LURD, the MODEL rebels only began profiting from the resources

in their territory after the deployment of peacekeepers. MODEL fighters illegally

occupied the Sinoe Rubber Plantation in late-2003 and began harvesting rubber.

With less risk of looting, large-scale export operations became possible in MODEL

territory. By February 2004, one month after peacekeepers deployed in Harper, the

Togba Timber Company began exporting round logs in collaboration with MODEL’s

forces (IRIN 2004b). The deployment of peacekeepers created the security necessary

for resource extraction that MODEL was unable to provide during the conflict.

21Allegations that Fawaz provided support to MODEL come from only two sources within MODELand could not be independently confirmed. Another informant, however, noted independently thatRoosevelt Johnson received $30-40,000 from an unknown source, which he used to distribute pay-ments and Thuraya satellite phones to various MODEL commanders (C32).



In addition to Roosevelt Johnson’s attempted coup, Thomas Yaya Nimely was chal-

lenged in early 2004 by MODEL commanders who were infuriated by their lack of

positions in the transitional government. Yaya Nimely was approached in his home

by a disgruntled delegation that included the group’s military spokesman Boi Bloaju

Boi, powerful commander “Garang”, and intelligence officer Wallace Dennis. The

men proceeded to trash Yaya Nimely’s house and accused him of handing out posi-

tions to his friends “from America” who played no part in the group’s success (C23,

C25, C31, C48). Yaya Nimely justified his decision, saying he allocated the positions

to “qualified people” who had the necessary educational background (C29).

The mutineers signed a petition and appointed Boi Bloaju Boi as the new leader

of MODEL, with the authority to distribute positions in the government. The United

Nations and regional states, however, would not accept this change in leadership, since

Yaya Nimely had signed the 2003 peace agreement (C29). The mutineers backed

down, and Yaya Nimely granted several mid- and low-level positions to the three

MODEL members who instigated the coup attempt. Wallace Dennis became the

Deputy Director of VIP protection at the National Security Administration; “Garang”

was appointed Deputy Managing Director for Administration at the National Port

Authority; and Boi Bloaju Boi became Assistant Minister of Finance. Once again,

Yaya Nimely maintained his position in the group through external intervention.


MODEL maintained slightly higher training standards than the LURD rebels. MODEL’s

top training commander was a former AFL officer who had been trained by the United

States during the Doe regime (C22). The training commander organized a month-

long basic training program for new recruits, initially based in Cote d’Ivoire. When

MODEL crossed into Liberia, the group decentralized its training activities to the var-

ious companies. Each company was assigned an “Assistant Instructor” who trained

new recruits and reported to the training commander at headquarters. The demands

of battle, however, soon undermined MODEL’s training standards, and commanders


scrapped the month-long program in favor of an abridged version. As MODEL ad-

vanced across Liberia, its recruits were sent to the frontline after only two weeks of

training (C22, C23, C25).

MODEL attracted recruits in ways similar to the other rebel groups. Very few

of MODEL’s recruits were abducted, but many joined for safety reasons. In James

Pugel’s (2007) survey, only 12% of MODEL respondents report being abducted by

the group, while 42% joined to protect their family and 23% joined because they were

scared. Aside from the lower percentage of abductees, these figures are similar to the

responses from NPFL and LURD members.

The MODEL leadership organized some food shipments to the forces in Zwedru,

but for the most part MODEL commanders either organized contributions from vil-

lages or allowed soldiers to loot (C22, C35). MODEL discipline was strongest in

Grand Gedeh, the Krahn homeland, where MODEL soldiers had ties with local com-

munities. For example, one of the few NGOs to visit MODEL territory after the

ceasefire noted that Krahn villages in Grand Gedeh appeared to be relatively un-

harmed, with cattle and food available for civilians, while other areas in the group’s

territory were nearly depopulated by predation (Merlin 2003). Once the ceasefire was

in place, MODEL commanders in Buchanan and elsewhere began organizing food for

their soldiers and cracked down on soldier misconduct (C20).

6.4.5 Discussion

MODEL’s short lifespan resembled in many ways the early days of the NPFL, thirteen

years before. MODEL’s offensive was accompanied by looting, but not the intense

predation that characterized ULIMO and LURD. And like the NPFL, MODEL made

contact with timber companies and other potential corporate partners. But these

negotiations failed to bear fruit for MODEL, and the group never gained access to

substantial revenues or arms. Timber companies attempted to install a stronger

leader, Roosevelt Johnson, who shared close ties with powerful commanders in the

group.

If this coup had succeeded, MODEL might have turned out differently. With


increased financing, Roosevelt Johnson could have strengthened the incentives for

discipline among the forces. These efforts, however, were thwarted by the government

of Cote d’Ivoire, who found a useful partner in Yaya Nimely. To maintain access

to weapons and basic supplies, Yaya Nimely agreed to allow much of MODEL’s

forces to remain in Cote d’Ivoire, diverting manpower from the group’s objectives in

Liberia. MODEL commanders looted for food and personal gain, and the group never

succeeded at enforcing discipline among its members. When the peace process began,

Thomas Yaya Nimely failed to reward MODEL’s commanders for their efforts.

MODEL’s experience illustrates the logic of the theory of commander incentives.

The government of Cote d’Ivoire, with preferences that conflicted with the Liberian

commanders in the refugee camps, installed a relatively weak leader who could of-

fer few incentives to the troops (Hypotheses 1, 3, 4, 10). And although the group

remained unified, top commanders attempted a split during peace talks (Hypothesis

8). Nevertheless, the theory fails to explain some aspects of MODEL’s behavior, such

as the relative discipline of MODEL forces in ethnic Krahn areas.

Competing theories can offer partial explanations for MODEL’s behavior, but

also suffer from shortcomings. Theories of territorial stability can explain the relative

security near MODEL’s base at Zwedru and the predation in contested towns such

as Buchanan and Harper; but fail to explain MODEL’s predatory behavior beyond

these unstable areas (Competing Hypothesis 1). Theories of costly signaling explain

MODEL’s abuses, but not the discipline near its headquarters (Competing Hypothesis

3). Theories of ethnic composition can explain MODEL’s discipline in ethnic Krahn

villages and predatory behavior outside of these areas (Competing Hypothesis 5).

But ethnic composition struggles to explain the generally predatory behavior of the

ethnically-homogeneous forces (Competing Hypothesis 4).

The theory of member selection offers more nuanced predictions for MODEL

(Competing Hypothesis 6). The resource-poor group’s initial reliance on commit-

ted Krahn soldiers from Ivorian refugee camps could explain the group’s discipline

in Krahn areas. But the group’s need to quickly expand to compete with the LURD

rebels soon flooded the group with opportunists, which explains the group’s preda-

tory behavior outside of the Krahn areas. This explanation, however, depends on the

6.5. ASSESSING THE THEORETICAL MODEL 211

assumption that committed recruits remained in Krahn areas while new recruits were

deployed elsewhere. The high-degree of mobility among MODEL’s forces, however,

does not support this assumption.

6.5 Assessing the Theoretical Model

The experiences of Liberia’s rebel organizations support and add nuance to the the-

ory’s causal mechanisms. The leaders of NPFL, LURD, and MODEL came to power

through their connections with external patrons. Only Alhaji Kromah became a rebel

leader without external influence, but he too was accepted as leader because of his

potential for attracting support. When ECOMOG grew frustrated with Kromah, the

peacekeepers recruited George Boley to lead a new rebel group, in the hopes that he

could be more easily controlled.

The motives of the external patrons determined the amount of resources available

to Liberia’s rebel leaders. Resource companies demanded security from the NPFL

and provided large sums of cash and weapons to augment Taylor’s ability to provide

it. Taylor used these resources to give cash incentives to his commanders, who in

turn exerted some degree of control over their soldiers. Liberia’s neighbors and the

ECOMOG peacekeepers, on the other hand, were more interested in groups that

could be manipulated and controlled. These patrons provided mainly weapons and

ammunition to rebel leaders who had few financial resources and were not trusted

by their troops. Lacking the ability to create incentives for discipline, these leaders

allowed their forces to loot. Commanders provided little training and failed to enforce

discipline among their soldiers.

The risk of factionalization among Liberia’s rebel groups seems, at first glance,

to contradict the theory’s predictions. The theory predicts that Charles Taylor’s

NPFL faced the lowest risk of factionalization, yet the group suffered two splits dur-

ing its lifespan. ULIMO, LURD, and MODEL are predicted to face a high risk of

factionalization, yet only ULIMO suffered a split. A closer look at the circumstances

surrounding these splits, however, reveals the value of the theoretical model. The

first NPFL split occurred at the start of the war, before Charles Taylor received the


first payments from his corporate backers. Taylor had family ties with some of the

commanders, but others were skeptical of his promises. When weapons failed to ma-

terialize, Prince Johnson decided Taylor’s promises were not credible and thought

he could increase his payout by going it alone. Unable to provide strong incentives,

Taylor could not prevent this split. MODEL, on the other hand, nearly split after a

timber corporation tried to install a stronger leader, but remained unified thanks to

a last-minute intervention from the Ivorian government.

The theory also predicts factionalization to be more likely during peace processes,

which finds strong support in the Liberia case study. The NPFL-CRC split in 1994

occurred because top NPFL officials received valuable positions in a transitional gov-

ernment, but feared Taylor would renege on the agreement. These officials split from

the group because Taylor could not promise them better offices in the future. ULIMO

split after a dispute between Alhaji Kromah and the Krahn forces regarding the tran-

sitional government. LURD and MODEL both faced attempted splits after the 2003

peace agreement due to the lack of rewards provided by the group leaders. With-

out a robust peacekeeping mission in the country, these groups would have likely

factionalized and Liberia would have descended once again into violence.

Liberia’s rebel groups reveal the importance of external patrons and commander

incentives. But these organizational challenges are only important if they affect the

situation for civilians. The next chapter uses qualitative and quantitative data on

civilian security during Liberia’s war to examine the theory’s spatial predictions and

compare them with predictions from alternative theories.

6.5. ASSESSING THE THEORETICAL MODEL 213

Thomas Yaya Nimely

Chief of Staff

7-8 Companies

Amos Chayee

(March-July 2003)

Commanding General

Arthur Baygboe (March-August 2003)

Kai Farley (August 2003)

Note: Arrows indicate chain of command and influence. Dotted arrows indicate weaker influence.

Figure 6.8: MODEL command structure.

Chapter 7

Civilian Security during Liberia’s

War

This chapter connects the internal politics of Liberia’s rebel organizations to the situa-

tion for civilians in their territories. The chapter begins with a qualitative description

of civilian security based on extensive fieldwork and then proceeds to a spatial, quan-

titative analysis. Using satellite imagery, I measure changes in crop area as a proxy

for civilian security. As predicted, the NPFL rebels provided security to civilians in

major towns but preyed on civilians in remote or unstable areas, while ULIMO, LPC,

and LURD were indiscriminately predatory. The MODEL rebels provided some de-

gree of security near their headquarters, most likely due to the extensive ethnic and

social ties that existed between MODEL soldiers and ethnic Krahn villages.

7.1 Life in Rebel Territory

Near the NPFL’s headquarters at Gbarnga and in other major towns, civilians bene-

fitted from relatively high security and were able to continue farming and trading at

pre-war levels. But civilians suffered from abusive NPFL soldiers near the frontlines

and in remote areas. The other groups—ULIMO, LPC, LURD, and MODEL—were

unable to provide even this modest level of security. Civilians were used as forced la-

bor by ULIMO and LURD, and suffered from rapacious looting by all of these groups.

214

7.1. LIFE IN REBEL TERRITORY 215

LURD was perhaps the most abusive of Liberia’s rebels. Many civilians fled for safer

areas; the few who remained had to rely on food found in the forest.

7.1.1 Life in NPFL Territory: Better Stay in Town

In June 1990, NPFL fighters marched up the main highway and reached the town of

Madina, 75 km northwest of Liberia’s capital Monrovia. Roosevelt,1 barely a teenager

at the time, joined the crowd of townspeople as they gathered on the road to hear

the rebels’ announcement.2 The commander of the rebel detachment explained they

were freedom fighters and that civilians had nothing to fear. The NPFL would only

harm government soldiers, the commander asserted, along with the ethnic Krahn and

Mandingo who supported the dictator Samuel Doe.

Over the next year and a half, life remained fairly normal for Roosevelt and the

other citizens of Madina. People continued to farm and sell their goods in the local

market at the edge of town. Every week the town chief made the rounds, collecting

food from each family and handing over the supplies to the local NPFL commander.

Roosevelt heard reports of hunger in Monrovia, but food was not a problem in Madina.

Travel along the roads, however, was risky. Looting and abuse was known to occur

in remote areas and near the frontlines (B45, B51, B53).

Madina’s situation was typical of many parts of Liberia under NPFL control.

When the group captured an area, civilians continued to farm their lands and sell their

produce in local markets. Within larger towns, civilians enjoyed a considerable degree

of protection from looting. Traders could sell their goods openly, and when NPFL

soldiers wanted something, they generally paid for it. Soldiers would occasionally

loot or harass civilians in the towns, but civilians could issue a complaint to the

local commander, who would then punish the soldier, often severely.3 If civilians

had disputes among themselves, they could plead their case to a policeman or take

their claims to courts set up by the NPRAG, the NPFL’s civilian administration

(C08). The NPFL even repaired the roads in certain parts of their territory, which

1Not his real name.2The narrative of Roosevelt’s experiences is based on interview B26 unless otherwise noted.3B25, B29, B30, B35, B37, B40.

216 CHAPTER 7. CIVILIAN SECURITY DURING LIBERIA’S WAR

encouraged trade (Ogunleye 1995: 138).

Under NPFL rule, certain towns became hubs of economic activity. A Nigerian

visitor in NPFL territory describes the relative prosperity of some areas: “A first-time

visitor to Kakata at night would think he was in pre-war Monrovia on Christmas day.

The multitudes on the streets were teeming... A number of new film houses and beer

parlours had sprung up” (Ogunleye 1995: 136). While people in Monrovia faced food

shortages and other deprivations, a few towns in NPFL territory offered luxuries and

access to relatively low-cost food.

NPFL commanders occasionally went to great lengths to protect civilians. In

Kpotomai, a village in Lofa County, two NPFL soldiers found suspicious powders in

the house of a village elder. The soldiers accused the old man of trying to poison

them, since the soldiers ate their meals in the village, and planned to execute him.

The elder pleaded his case to the NPFL commander and explained that the powders

were actually medicine; he was responsible for storing drugs for the village as the

nearest hospital was far away. The commander convened a town meeting, during

which the villagers vouched for the old man. The commander ordered that the man

be released and the soldiers did not hassle him again (B12).

The NPFL benefited by protecting and encouraging local production and trade.4

NPFL units stationed in villages received their food from local communities. The

chief would collect a share of rice from every household, and women from the village

would prepare meals for the soldiers. In larger towns, market sellers were required

to pay a daily tax of about $0.10, a large part of which went to the local NPFL

commander, and traders were expected to give approximately 50% of their goods to

soldiers who manned “gates” at the entrance to each town.

Despite these incentives, the NPFL failed to protect civilians outside of the major

towns. Civilians traveling along roads in NPFL territory faced a constant threat

of expropriation and harassment. Far from the “law-abiding” commanders, NPFL

soldiers had less fear of being caught and punished (B29, B30, B35, B37). The worst

offenders were NPFL soldiers returning from the front lines. On their journey back

4Information on NPFL taxation of market activity in this paragraph comes from B17, B19, B22,B23, B34, B43, B46, B53, C17, C45.


to headquarters these soldiers preyed on unfortunate civilians that happened to cross

their path.

Civilians adopted strategies to cope with this predation and abuse. Local trade

became increasingly dominated by women, since they could cross NPFL roadblocks

with less risk of harassment than men.5 Many of these women formed personal

relationships with NPFL commanders and soldiers, which allowed them to travel

more safely and pay fewer taxes than their competitors. These “soldiers’ wives”

could also sell looted items on behalf of their rebel “husbands”. This strategy of

forming personal relationships with rebel soldiers became increasingly important for

survival as the war progressed.6

For its part, the NPFL implemented a pass system aimed at curbing the looting

that occurred along the roads (B06, B15, B17, B45).7 Civilians traveling between

towns registered with a local NPFL official to receive a pass, which deterred soldiers

from looting goods that could otherwise be taxed by NPFL commanders. The passes

were also important for ensuring a stable food supply for NPFL soldiers based in the

large towns. As a woman trader in Voinjama explained, “soldiers wanted to eat, so

they had to give us a chance [to make it to town without being looted]” (B06).

The NPFL’s rapid advance across Liberia created a humanitarian emergency with

few aid organizations equipped to handle the crisis. Aid operations were in disarray

until the deployment of ECOMOG opened up the port and airport for aid deliv-

eries, and a few organizations such as MSF, CRS, and ICRC provided assistance in

Monrovia (Scott 1995). In late-1990 these organizations tentatively expanded beyond

Monrovia, but largely suspended their operations in 1993 when ECOMOG enforced

an economic blockade of NPFL areas. Only MSF continued to operate in Greater

5Information for this paragraph from B08, B17, B22, B23, B43, B45, B50, B51, B52, B53, B54.6The economic importance of women during conflict has received some attention in the anthro-

pology literature. See Mats Utas (2005) for a description of the role of women during Liberia’sconflict and Chris Coulter (2009) and Christiana Solomon (2006) for studies of the involvement ofwomen in Sierra Leone’s war.

7The pass system also allowed the NPFL to better control the civilian population. To receive apass, civilians had to present themselves to the local G-2 office (NPFL intelligence) and be interro-gated by an NPFL official. Only once the official was satisfied that the civilian was not Krahn orMandingo, or posed a threat to the NPFL, would a pass be issued (B45). Also see Ogunleye (1995:130) for a description of the fear and risks that might accompany a visit to the G-2 office.


Liberia, and suffered an ECOMOG air attack on one of its convoys (Ellis 1999: 100).

Charles Taylor’s NPRAG administration maintained contact with MSF and played

some role in coordinating the group’s activities. During 1993 MSF opened feeding

centers in several locations in NPFL territory. MSF’s official policy was that no

food be distributed to soldiers at the feeding centers, but MSF staff funneled food

shipments to the NPFL by inflating reports of the numbers of children in need. For

example, if a feeding center provided support to 800 children, the manager of the

center would request supplies for 1,000 children and deliver the difference to the NPFL

(C40). In exchange for providing security, NPFL benefited from MSF’s humanitarian

operations.

The NPFL occasionally delivered its own humanitarian aid to civilians. In 1990

the NPFL captured a ship in Buchanan port loaded with rice. The NPFL assigned

quotas to villages in the surrounding area and local chiefs and other delegates made

the journey to Buchanan to claim their rations, free of charge.8 This period of abun-

dance, however, was short-lived and the NPFL soon returned to taxing civilians for

food and other vital supplies.

As the conflict intensified in 1995, even the limited humanitarian assistance in

Taylor’s areas had to be suspended. Although an estimated 1.8 million civilians

needed humanitarian assistance, aid organizations could only operate in the few zones

of ECOMOG control (UNSC 1995a). Aid organizations gradually expanded their

operations in early 1996, only to see their gains reversed by the return to intense

fighting in April (UNSC 1996b). Civilians continued to struggle until 1997, when

disarmament allowed for a dramatic expansion of aid operations throughout Liberia.

7.1.2 Life in ULIMO Territory: Run for the Hills

Back in Madina, Roosevelt’s situation changed in 1992 when the town was captured

by the ULIMO rebels. Fighting erupted along the road and, despite assurances from

the ULIMO commander, people began to fear for their safety after reports of looting

by ULIMO soldiers in nearby villages (B46). Soon after crossing into Liberia, ULIMO

8B34, B37, B38, B40, B42, B43.


faced shortages of food and supplies. ULIMO soldiers rounded up villagers and looted

all food and valuable goods found in civilian houses. To prevent depopulation, the

soldiers forcibly relocated villagers to larger towns, where they could be supervised

(B14, B46, B51, B52). These civilians were forced to carry loads of food and looted

items, or sent out with ULIMO escorts to hunt and fish for the soldiers (B12, B54).

Even in major towns, civilians could be looted by ULIMO forces with impunity

(B18, B22, B25). Civilians in many areas survived by foraging for bush yams and

other wild foods (B12, B21, B55). Desperate families established links with the rebels

by encouraging young females to become a soldier’s “wife”. Through these female

relatives, families could gain access to food and some degree of security (B46).

What little trade occurred was dominated by ULIMO members.9 Trains of civil-

ians carried looted goods to the Guinea border on behalf of ULIMO commanders and

suffered brutal treatment along the way. In Voinjama, for example, a man died while

trying to carry an engine block that ULIMO soldiers had loaded onto his back (B15).

The profits from this trade were used by “soldiers’ wives” to purchase foodstuffs in

Guinea, which were then sold in small markets. Goods were purchased mainly by

ULIMO soldiers and prices were extremely high. For example, a bag of salt, normally

priced at LD30 ($0.50), sold for LD150 ($2.50) or more. ULIMO was hurt by the

lack of market activity in its territory. The collapse of farming meant that ULIMO

could not purchase food or tax local production. Looting supplied some food, but the

group struggled to keep its soldiers fed. At times, desperate ULIMO soldiers went

into abandoned fields and harvested rice and cassava themselves (C09).

Security for civilians declined even further in 1994 when ULIMO broke apart.

Madina and other towns in Grand Cape Mount and Bomi counties changed hands

between the competing factions, ULIMO-J and ULIMO-K. Civilians were indiscrim-

inately killed as their villages were systematically stripped of zinc roofing and any

items that could possibly be traded for ammunition or profit (B26, B42, B46, B54).

Life was no better in Liberia’s southeast, where the LPC rebels, composed of ULIMO

defectors, preyed on civilians to such an extent that their territory was virtually

depopulated. The only hope for civilians was to flee to ECOMOG-controlled areas

9Information for this paragraph comes from B17, B19, B20, B21, B25, B46, B50, B54.


where some markets still functioned and food was supplied by aid organizations (B40,

B52, B53).

Aid organizations made only sporadic forays into ULIMO territory, at great peril.

ULIMO looted food aid when it captured Klay in mid-1992 (B52). In December

1993, ULIMO looted and destroyed the UNHCR office in Lofa County, causing the

suspension of all humanitarian operations in the region (UNSC 1994a: 5). Aid or-

ganizations returned to ULIMO-J and -K areas during a ceasefire in 1995, but could

only deliver aid in small quantities by helicopter because rebel roadblocks made it

impossible to travel by road (UNSC 1995b: 6-7). Oxfam made a brief foray into

Lofa County in late 1995, but suspended operations once again after its convoy was

seized by ULIMO-K forces (UNSC 1996a: 7). The only substantial aid was provided

in ECOMOG “safe havens” at Bo Waterside, Tiene, Tubmanburg, and Klay, where

the presence of substantial peacekeeping forces could ensure the aid was delivered to

civilians (UNSC 1995b: 6-7).

Aid organizations were tempting targets for looting because they provided ULIMO

with communications equipment, vehicles, and other valuable supplies (UNSC 1994b:

9). But ULIMO was ultimately hurt by this predatory behavior. The lack of hu-

manitarian aid in ULIMO territory prevented the group from gaining access to food

and medical attention. Soldiers increasingly turned to looting rather than military

offensives, and the plight of civilians hurt ULIMO’s image among international actors

such as the United Nations. ULIMO’s group goals were sacrificed for the individual

incentives of its soldiers and commanders.

After ULIMO’s split, Roosevelt and his family fled Madina to seek shelter in the

surrounding forest. He foraged for bush yams and other food but had to be careful to

avoid capture by the rebel forces. After a month, Roosevelt made his way on foot to

Monrovia, where some humanitarian aid was available. He returned to Madina only

after a ceasefire and the deployment of peacekeepers offered some degree of protection

from the predatory rebels.


7.1.3 Life in LURD Territory: Carrying a Heavy Load

An uneasy calm returned to Madina after the 1996 peace agreement, but proved only

temporary. In 2002 the LURD rebels swept through the town. Roosevelt tried to

flee, but was captured by the soldiers. Roosevelt and others were forced to carry

ammunition and supplies to LURD forces on the frontlines and he spent days hauling

back-breaking loads down the highway, pushed onwards by threats from the soldiers.

Unable to stand the stress, many of Roosevelt’s friends joined the rebels in hopes

of acquiring food and gaining protection from the unpredictable cruelty of the rebel

forces.

LURD’s reputation for predation and abuse spread even faster than LURD’s forces

across Liberia. Civilians viewed LURD as a reincarnation of ULIMO, and were not

anxious to relive those desperate times.10 LURD’s initial advances in 2000-2001 were

accompanied by looting, which further cemented their reputation. Entire villages

depopulated in anticipation of LURD’s arrival and either fled to displacement camps

or took cover in the surrounding forest.11

Once LURD secured control over an area, the group instituted a system of forced

labor for the remaining population.12 LURD soldiers went on patrols for civilians

hiding in the forest. These civilians, along with those from small villages, were forcibly

relocated to Voinjama or other major towns under LURD control.13 LURD soldiers

stripped buildings of their zinc roofs and electrical wiring, and looted anything of

value. Men and women were forced to carry these items to the Guinea border, where

“soldiers’ wives” exchanged the goods for money and food. Civilians also carried

ammunition and military supplies to the front line, up to 5 days’ walk, and were forced

to gather food for the LURD forces in their areas. Throughout LURD territory, food

and any items of value were immediately looted by the nearest soldier. In Madina,

10Many Liberians still confuse LURD with ULIMO, which is not entirely inaccurate. All of LURD’stop military commanders and political officials, excluding Sekou Conneh, were former members ofULIMO.

11B19, B20, B21, B22, B51, B52, B54.12Information for this paragraph comes from B03, B07, B13, B15, B19, B26, B27, B42, B46, B49,

B51, B53, B55, C1413Some senior military officials claim this strategy was to protect and provide support for civilians

who “could not flee” (B10, C02).


Roosevelt survived by foraging for cassava in abandoned farms. If LURD soldiers saw

him with food, the soldiers would beat him and force him to show where the food

came from. In other areas, people relied on wild food found in the forest.

Compounding their desperate situation, civilians in LURD territory had no access

to humanitarian aid. One of LURD’s earliest attacks in Liberia involved looting the

U.N. office in Kolahun, capturing 10 vehicles and holding the staff hostage for several

weeks (C09). Aid agencies throughout the northwest suspended operations and their

offices were looted immediately upon LURD’s arrival. By April 2003, aid organiza-

tions could only function in four of Liberia’s 15 counties, none in LURD territory

(UNSC 2001). When LURD reached Monrovia in June 2003, aid organizations strug-

gled to operate even within the confines of the city center. During its three attacks

on Monrovia, the rebel group managed to loot 50 vehicles from the compounds of

aid organizations (IRIN 2003a). Aid only tentatively extended into LURD areas in

late-2003, after the deployment of peacekeepers. Aside from a few vehicles acquired

through looting, LURD failed to benefit from humanitarian assistance, and civilians

suffered from its absence.

Market activity ceased in LURD areas except for the trade controlled by the

“soldiers’ wives”.14 Small markets run by the “wives” in towns such as Voinjama

and Tubmanburg offered basic goods such as rice, salt, and cigarettes, which were

purchased by LURD soldiers. The women earned a profit from these sales, which

they shared with their rebel “husbands”. All rebel soldiers, even those without wives,

benefited from the trade because it augmented their diet and ensured a stable food

supply. But even with a rebel “husband”, women traders faced risks of looting and

rape. LURD soldiers only refrained from abuse when there was an imminent threat of

punishment by the rebel “husband”. The women’s security depended on the power—

and close proximity—of their rebel partner.

Facing severe food shortages, the LURD leadership in Voinjama attempted in

mid-2001 to protect market activity. LURD soldiers began to receive a “political ed-

ucation” that stressed the necessity of avoiding civilian casualties. At least one LURD

soldier was executed for killing a civilian, and forums for hearing local grievances were

14Information for this paragraph comes from B01, B02, B03, B15, B17, B25, B28, B42, B47


created. Civilians in Lofa County at the time reported that abuse decreased, but only

briefly. By the end of 2001 human rights violations were rampant once again (ICG

2002: 9-10). LURD also instituted a pass system similar to the one created by the

NPFL, but the passes failed to improve civilian security. Even with a pass, valuable

items would be looted at the nearest roadblock (B15, B17).

The situation improved for civilians in some parts of LURD territory in August

2003 when LURD captured Monrovia’s port.15 LURD soldiers looted huge stores of

rice and consumer goods from the port, which they transported and sold throughout

Bomi and Cape Mount counties. Local markets were flooded with goods, pushing

prices down below the cost of manufacture. Soap, which costs LD40 ($0.70) during

peace time, sold for LD5 ($0.09). A 50kg bag of rice, normally $25-30, could be

purchased for $3. LURD soldiers, saturated with goods, even handed out rice for

free in some areas. In other locations, civilian porters during this time received one

cup of rice as compensation for their labor. Starving civilians in Monrovia crossed

into LURD territory and, for the first time since LURD arrived, civilians in Bomi

and Cape Mount counties faced little risk of looting. Roosevelt barely scraped by

during the two-year occupation of the town, but finally found some relief after LURD

captured the port and when the return of peacekeepers allowed humanitarian aid to

reach the village.

7.1.4 Life in MODEL Territory: Ethnic Ties Help

While the LURD forces preyed on Madina and other towns in Liberia’s northwest,

MODEL launched an invasion in the southeast. MODEL received some food ship-

ments from Cote d’Ivoire at its Zwedru base, but relied on civilians to provide food

for its troops elsewhere in its territory.16 Soldiers searching for food soon turned to

looting, and large towns were especially hard hit. Buildings in Buchanan and Harper

were stripped of anything of value—even doorknobs and toilets—and the towns were

largely depopulated (IRIN 2003b, 2004b). MODEL’s looting caused displacement

15Information for this paragraph from B26, B27, B44, B45, B46, B48, B51, B54.16C22, C25, C31, C32, C35.


and a temporary collapse in outlying areas as well. Villages in Sinoe County, for ex-

ample, were emptied as terrified civilians hid in the forest during the first few months

of MODEL control. The situation only improved in October 2003, two months after

the ceasefire, when humanitarian aid began reaching into MODEL territory (Merlin

2003).

Some parts of MODEL territory were spared the group’s predatory behavior.

Krahn villages in parts of Grand Gedeh County, especially near Zwedru, were rela-

tively untouched by looting. Civilians stayed in their homes and continued to farm,

trade, and herd cattle (Merlin 2003). This security most likely resulted from the

ethnic and family bonds shared between these villages and MODEL soldiers, nearly

all of whom originated from the area.

At least one other area of MODEL territory also escaped the group’s abuses. A

UNHCR team visiting Pleebo, a border town in Maryland County, in September 2003,

reported that the town’s markets were “bustling” with cross-border trade from Cote

d’Ivoire. In addition, the town maintained a sizable population and even women and

children could be seen on the street (IRIN 2003b). The likely reason for this security,

however, was the absence of MODEL forces, which had only briefly passed through

the town on their way to Harper, 25km away. When the UNHCR team visited Harper

later the same day, they found the town nearly deserted and thoroughly looted by

the MODEL soldiers stationed there (IRIN 2003b).

MODEL struggled to restrain its soldiers even after the August 2003 ceasefire.

Persistent predatory behavior in Buchanan led MODEL leaders to create a civilian

council in the town. The five-member council of elders communicated directly with

MODEL’s commanding general, Kai Farley, and reported all cases of soldier miscon-

duct and looting (C20, C32). The situation improved, but civilian abuse remained

common even after the deployment of peacekeepers in December 2003 (IRIN 2004c).

No aid organizations operated in MODEL territory until after the August 2003

ceasefire. MODEL, however, generated some benefit from the aid organizations in

Cote d’Ivoire. The rebel group formed in the “Peace Town” refugee camp, where

UNHCR and several other organizations provided food and other support to the

refugees. After MODEL crossed into Liberia, soldiers visited the camp and received

7.2. REMOTE SENSING AND LOCAL FOOD PRODUCTION 225

rations and other support from the aid organizations and the refugee population.17

Civilians in Liberia had to wait until the start of WFP food shipments in September

2003 to receive any support. This aid increased MODEL’s access to food supplies

and other valuable resources, but was never systematically taxed or controlled by the

rebel leadership. Throughout the war, humanitarian aid had a negligible effect on

the group’s capabilities or behavior, and civilians struggled to survive.

7.1.5 Discussion

This qualitative evidence supports the theory of commander incentives. The NPFL,

with access to leader financing and credible promises, is predicted to provide greater

security for civilians, but this security depends on the presence of lootable resources

and effective monitoring (Hypotheses 1-6). The ULIMO and LURD rebels, with

little access to these resources, preyed indiscriminately on civilians, even in easy-to-

monitor areas such as the rebel headquarters (Hypotheses 6, 7). Competing theories

find mixed support. The predation that dominated unstable areas is consistent with

theories of territorial competition, but not the predation that occurred throughout

LURD and ULIMO areas. The relative security for Krahn civilians near MODEL’s

headquarters supports theories of ethnic composition, but not the predation from the

ethnically-homogenous MODEL soldiers outside of these areas, or the indiscriminate

predation from the ethnically-homogeneous ULIMO-J, ULIMO-K, or LPC. The re-

mainder of the chapter assesses these hypotheses through a quantitative analysis of

changing patterns of crop area during Liberia’s war.

7.2 Remote Sensing and Local Food Production

The Liberian conflict produced countless stories of atrocities, but few data exist that

can provide a systematic picture of rebel behavior. Satellite imagery collected during

the conflict, however, can shed light on the situation on the ground by revealing

changes in local food production, measured through crop land. Crop land is a proxy

17During a visit to the “Peace Town” refugee camp in August 2003 I met several MODEL fighterswho were staying briefly in the camp before returning to the front line. Also see HRW (2003a).


for civilian security because the incentive to clear crop land depends on the civilians’

experiences with the rebel group, as well as their expectations of the group’s future

behavior. Crop land declines in areas that have been depopulated by violence and

predation. Among the population that remains, a farmer will not spend the energy

to clear land if he expects his crops to be looted upon harvest—or if he might be

killed or displaced in the meantime.

In Liberia, civilians in insecure areas abandoned their farms and sought shelter

in refugee camps or hid in the surrounding forest and survived on foraged foods such

as bush yam. During stable times, Liberian farmers clear approximately the same

amount of land each year, limited by their physical ability to clear forest.18 Total

crop area is determined by the number of farmers able and willing to clear land in a

given area at a particular time. Over the course of the war, therefore, a decline in

crop land in t1 indicates that the area is less secure than it was in t0.

Analyzing crop area in a country such as Liberia presents certain challenges. First,

the majority of Liberians outside the capital city depend on small-scale, subsistence

agriculture which can be difficult to identify in a satellite image. Second, Liberia

experiences nearly constant cloud cover, which makes it possible to obtain satellite

images only during the brief dry season, January-March. Third, crop land can be

difficult to distinguish from the surrounding vegetation, or from cleared areas such as

villages, roads, and sports fields.

Fortunately these challenges can be overcome. Two satellite programs, LANDSAT

and SPOT, provide images with 20-30m resolution—sufficient for identifying the small

fields. Further, the dry season, when all cloud-free images are collected, is the time

when farmers slash and burn the areas they intend to plant, making the cleared crop

land easily identifiable against the surrounding vegetation. Because land is cleared

before the first rains of the season, crop area measurements are insensitive to annual

variation in rainfall or other climatic events. Finally, new land is cleared each year

in an eight-year fallow cycle. Layering multiple satellite images highlights changes in

land use and makes the shifting crops easily distinguishable from stationary villages

18Liberia is characterized by low population density and large tracts of uncultivated land. Ac-cording to interviews with farmers, land use is constrained by the labor required to clear forest fornew crop fields, not by limitations on the supply of land or the ability to purchase new land.


and other non-agricultural areas.

To measure changes in crop area in NPFL territory, I compare four LANDSAT

images from January 1991 to four images of the same region from January 1986,

before the start of the war. The coverage area includes neighboring agricultural areas

in Guinea and Sierra Leone. LURD territory is covered by three LANDSAT images

from March 2003, which are compared to three corresponding images from February

2001, before the first major rebel offensive. These images cover all of LURD territory,

as well as large tracts of government territory and neighboring areas of Guinea and

Sierra Leone. Figure 7.1 depicts the coverage area. Very few satellite images are

available for Liberia during 1992-2000, but four SPOT images, two from January

1993 and two from January 1995, can be combined with the LANDSAT images to

provide additional information on a small but significant area of Liberia, depicted in

Figure 7.2. This area was partially captured in 1994 by the LPC rebels and provides

an identification strategy to isolate the effects of rebel behavior by examining changes

over time in a fixed geographic area. Unfortunately, crop area cannot be measured

in ULIMO or MODEL territories due to insufficient satellite data.

For each image, a certain number of pixels are classified manually by the re-

searcher.19 These training pixels provide the input necessary for computer software

such as ENVI to categorize the remaining pixels in the image based on maximum

likelihood estimators. The resulting classified image is then adjusted iteratively until

no major errors can be determined through inspection (CI 2006). Specific types of

land cover can be difficult to identify in satellite imagery. To confirm my ability to

distinguish important land categories, I visited various locations in Liberia to collect

ground truth information and interviewed farmers on their land use practices. The

final result is a simple raster dataset that allows for crop area measurements in a GIS.

Assessing the accuracy of classification images is a vital part of land-use research

and the subject of a large literature in the earth sciences (for an overview see Foody

2002 and Stehman and Czaplewski 1998). I selected a random sample of pixels from

the classified image and then compared them manually to the original images (Hess

19A typical image might contain 12-30 classes, including, for example, “forest”, “dark forest”,“river”, “road”, and so on.


NP

FL

1986 v. 1991L

UR

D 2001 v. 2003

[Monrovia

Tubmanburg

Voinjama

Gbarnga

Buchanan

Sierra

Leon

eGuinea

-16.1%

+7.9%

+4.8%

Sierra

Leon

e

Guinea

+120.9%

-39.5%

[Monrovia

Tubmanburg

Voinjama

Gbarnga

Buchanan

-37.1%

-34.6%

Note:

Studyareas

depictedby

boldlines.

Figu

re7.1:

Chan

gein

croparea

inN

PF

Lan

dL

UR

Dterritories

(shad

ed)

and

neigh

borin

gregion

sof

Sierra

Leon

ean

dG

uin

ea.


[Monrovia

Tubmanburg

Voinjama

Gbarnga

Buchanan

1991 v. 1995

+17.2%

-18.3%

Note: Study area depicted by bold line.

Figure 7.2: Change in crop area, 1991 v. 1995, in territory that remained underNPFL control (shaded) and territory captured by LPC (dotted).


and Bay 1997). For this project, 600 randomly-selected pixels were manually assessed

for each pair of LANDSAT images and 300 random pixels were assessed for each pair

of smaller SPOT images, stratified between crop and non-crop areas.

LANDSAT pixels identified as crop land were 81-97% likely to reflect true crop

land, depending on the image, with an average accuracy of 88.7%; pixels classified

as non-crop were 95-99.25% likely to reflect true non-crop areas, with an average

accuracy of 98%. Among SPOT images, pixels classified as crop land were 72-96%

likely to reflect true crop land, with an average accuracy of 86.4%; pixels classified as

non-crop were 98-99.5% likely to reflect true non-crop areas, with an average accuracy

of 98.6%.

These error matrices were then used to estimate crop areas, standard errors, and

statistical significance using bootstrap simulations with 100,000 samples. This process

provides an estimate of measurement error for the image as a whole, which is then

used to estimate crop area at lower levels of aggregation, such as the village level. This

extrapolation is not ideal, since errors may be spatially concentrated in the image,

but provides a better estimate than the raw measurements, which have known biases.

Error propagation caused by these issues continues to pose a fundamental challenge

to spatial analyses (Haining 2003: 126).

7.3 Group-Level Comparisons

Group-level comparisons are a useful starting point for a quantitative look at the the-

ory’s predictions compared to existing theories of rebel group behavior. The theory

of member selection predicts that groups with access to economic resources are dom-

inated by opportunistic members who are more likely to abuse civilians (Competing

Hypothesis 6). In Liberia, only the NPFL had access to any significant economic

resources, though all of the rebel groups offered looting opportunities. The theory

of member selection, therefore, predicts that the LURD and LPC provided at least

as much security for civilians as the NPFL rebels, if not more. Theories of ethnic

composition expects higher levels of security from the homogeneous LPC rebels and

7.3. GROUP-LEVEL COMPARISONS 231

lower levels of security from the ethnically diverse NPFL and LURD rebels (Com-

peting Hypothesis 4). Theories of territorial competition expect higher security in

stable areas than in unstable areas for all groups (Competing Hypothesis 1). By

contrast, the theory of commander incentives, proposed here, predicts that the NPFL

provided more security than either LURD or LPC, due to Charles Taylor’s access to

financial resources and personal ties to commanders (Hypotheses 1, 3, 4). Table 7.1

summarizes these predictions.

Between 1986 and 1991, crop area is estimated to have increased 4.8% in the

NPFL areas covered by the study area. This difference, however, is not statistically

significant (p = 0.35) when compared to the null hypothesis of no change in crop

area.20 Crop area in neighboring regions of Guinea grew by 7.9% during this period

(p = 0.34), while crop area in neighboring regions of Sierra Leone decreased by 16.1%

(p = 0.09).

The situation in LURD areas between 2001 and 2003 was dramatically different.

Crop area decreased by 37.1%, which is significant at p = 0.028. Neighboring regions

of Guinea, where LURD forces were also active, decreased by 39.5% (p = 0.000), while

neighboring government-controlled areas of Liberia decreased by 34.6% (p = 0.016).

Crop area in neighboring regions in Sierra Leone—where a decade-long civil war ended

in 2002—increased by 120.9% (p = 0.000) during this period. Figure 7.1 depicts these

changes in crop area.

Crop area declined more in LURD territory than in NPFL territory, but these data

do not necessarily imply that LURD was more abusive than NPFL. The declines in

LURD territory could be caused by larger, systemic forces that were also responsible

for the large declines in neighboring government areas and in Guinea. Rather than

LURD abuse, the declines in crop area could be caused by fighting or the predatory

behavior of government forces. Indeed, unstable government areas that were attacked

by LURD and other armed groups, such as Nimba County, experienced 45.8% declines

in crop area (p = 0.008). Crop land in unstable LURD areas, such as the region

around Zorzor, declined by 65.7% (p = 0.0004).

20The significance values are best interpreted as the probability that the difference in crop area ispurely the result of measurement error.


A comparison of stable areas, however, reveals significant differences between

LURD and government territories. Crop area in relatively stable government-controlled

areas such as Bong County declined by only 6.2% during this period (p = 0.33).

In contrast, even the most stable LURD areas around Voinjama declined by 53.6%

(p = 0.076). In LURD territory, the lowest declines in crop area were in extremely

remote areas, only accessible by foot and relatively protected from looting. These

patterns suggest that the behavior of LURD forces contributed to the decline in food

production in the group’s territory.

Although fewer data are available for the LPC rebels, some comparisons can be

made. The study area depicted in Figure 7.2 was controlled by the NPFL until

1994, when the LPC captured part of the region. Between 1991 and 1995, crop area

declined 18.3% in LPC areas (p = 0.40) while neighboring NPFL areas benefitted from

an increase of 17.2% (p = 0.26). Due to the poor quality of the images, however, this

difference could potentially be explained by measurement error (p = 0.28).

These group-level comparisons contradict theories of member selection and ethnic

composition, which predict greater security from LURD and LPC. The data are

consistent with theories of territorial stability, but the large declines in stable LURD

areas suggests other factors are also important for explaining patterns of security

during conflict. The results are in line with the theory of commander incentives,

but key elements of the theory, such as the effects of lootable resources and imperfect

monitoring, cannot be examined at the group level. Analyzing variation within group

territories allows for a more thorough test of the theory’s predictions.

7.4 Within-Group Spatial Variation

To measure within-group variation, I aggregated crop area at the village level by

constructing Thiessen polygons around villages identified during the 2008 census.21

These polygons assign all crop pixels to the closest village, providing a measure of

21Very small settlements were combined with larger neighbors to ensure that all villages areseparated by at least 500 meters. Village locations did not change significantly in Liberia betweenthe pre- and post-war periods, so the data do not suffer from selection bias due to destroyed ornewly-formed settlements.

7.4. WITHIN-GROUP SPATIAL VARIATION 233

crop area for each village in each time period. Figure 7.3 depicts the percent change

in crop area in NPFL and LURD territories.

The theory of commander incentives has several spatial implications. Rebel leaders

who receive financing from external patrons are able to provide incentives and control

their subordinates, but this control depends on effective monitoring (Hypotheses 1 and

5). More effective monitoring decreases the chances that rebels will be predatory near

rebel bases, but this effect might be offset by the greater number of rebels near the

bases. By contrast, when leaders cannot offer strong incentives to their subordinates,

security is lower near rebel bases than in remote areas due to the higher concentration

of predatory rebels (Hypothesis 7). For all groups, security declines in areas that

contain lootable resources and in unstable areas (Hypotheses 2 and 5). In Liberia,

the most prominent lootable resource, aside from civilians’ assets, was rubber, due

to the thriving black market trade in which rebel soldiers and civilians sold rubber

to Nigerian peacekeepers (C09). A truckload of rubber could fetch $300 on the black

market (Ellis 1999: 167).

Table 7.2 lists the various security outcomes predicted by the theory in the

Liberian context. The theory predicts higher security in NPFL territory than in

LURD territory, especially near the rebel headquarters. Thus A > E and B ≥ F .

The LURD rebels are predicted to be more predatory near their headquarters than

far away, thus E > F . And rubber is predicted to correspond to lower levels of se-

curity for both groups, thus C > D and G > H. The theory does not offer a clear

prediction for the relationship between A and B.

Descriptive statistics are a useful starting point for analyzing the data. Table 7.3

lists the average percent changes in crop area at the village level. Villages are sorted

into four strata, depending on the distance from the rebel headquarters, and are

also divided between rubber and non-rubber areas. The first two columns depict the

results from the raw data, without any controls. As predicted, the NPFL is associated

with more positive changes in crop area than LURD, and the largest declines are

observed among villages closest to LURD’s headquarters. Villages that contain rubber

suffer larger declines in crop area than non-rubber villages among both groups.

These raw data, however, may be biased if there are systematic differences in


Theory Civilian Security

Member Selection LURD ≈ LPC ≥ NPFL

Ethnic Composition LPC > LURD ≈ NPFL

Territorial Competition Stable Areas > Unstable Areas

Commander Incentives NPFL > LURD ≈ LPC

Table 7.1: Competing predictions of rebel behavior in Liberia.

Rebel Group Near Rebel HQ Far from HQ No Rubber Rubber

NPFL A B C D

LURD E F G H

Table 7.2: Village-level security outcomes in rebel territory.

Raw Data Matched SampleNPFL LURD NPFL LURD

0-60km 19.8 -55.0 25.7 -82.860-120km 29.6 -39.8 34.3 -54.8

120-180km 48.0 1.3 60.8 14.9180+ km 76.4 -31.0 53.8 -24.6

Rubber -11.4 -34.4 -10.2 -32.2No Rubber 44.6 -28.9 17.1 -44.5All Villages 39.4 -28.4 42.8 -12.2

Table 7.3: Average percent change in crop area among villages, according to distancefrom rebel headquarters or presence of rubber.


-0.9

7 -

-0.5

0-0

.50

- 0

.00

0.0

0 -

0.5

0 0

.50

- 37

.5

NP

FL

198

6 v.

1991

LU

RD

200

1 v.

2003

-0.9

9 -

-0.5

0-0

.50

- 0

.00

0.0

0 -

0.5

0 0

.50

- 12

.0

LUR

D T

errit

ory

Fig

ure

7.3:

Per

cent

chan

gein

crop

area

atvilla

gele

vel

inN

PF

Lan

dL

UR

Dte

rrit

orie

s.


village characteristics between NPFL and LURD territories, or between rubber and

non-rubber villages. Ideally, villages would be randomly assigned into these cate-

gories. Since randomization is not possible, matching provides a possible alternative.

The goal is to compare villages at various distances from the NPFL’s headquarters to

similar villages in LURD territory. Villages in rubber areas are matched with similar

villages in non-rubber areas. By removing from the sample villages without suitable

matches, the bias is reduced (Ho et al 2007).

The villages are matched based on the village area,22 the percentage of crop area

in t0,23 and the number of battles within 30km of the village since t0, derived from the

ACLED dataset (Raleigh et al 2005). These matching covariates capture important

geographic and economic features of the villages, as well as the village’s exposure

to violence. Villages were matched using the Matching library in R, with a caliper

of 0.25 standard deviations (Sekhon 2007). Balance statistics are provided in the

chapter’s appendix and reveal that the matched samples are not entirely balanced

due to Liberia’s geographic characteristics. The results, therefore, must be treated

with caution.

The last two columns in Table 7.3 lists the results from matched samples of vil-

lages, and are largely consistent with the raw data. Again, the NPFL demonstrates

greater positive changes in crop area than LURD, and villages closest to LURD head-

quarters suffer the largest declines. Figure 7.4 plots these results with 95% confidence

intervals. The key feature of the figure is column (3), which shows a significant posi-

tive effect of NPFL rule on village crop area compared to LURD.

The effects of rubber are depicted in Figure 7.8. In NPFL territory, rubber villages

experience larger declines than non-rubber villages. In LURD territory, contrary to

the theory’s predictions, matched villages in rubber areas show lower declines than

non-rubber villages. Rubber villages in LURD territory show similar declines in both

the raw data (-34.4%) and in the matched sample (-32.2%). The changing effect

comes entirely from the non-rubber villages, which decline an estimated 28.9% in the

22Village area is measured in pixels, based on the Thiessen polygon. As a result, remote villagestend to have larger village areas.

23t0 is defined as the date of the initial satellite image, January 1986 for the NPFL and February2001 for LURD.


raw data and 44.5% in the matched sample.

Matching provides some insight but has two significant limitations in this context.

First, matching only considers a few covariates and overlooks other factors that might

affect civilian security and changes in crop area. For example, monitoring might be

more effective near a major road, or civilians might be more likely to abandon their

farms if they are near an international border or humanitarian aid operations. Second,

and more importantly, matching fails to account for the strong spatial dependence

among villages. If one village is attacked, people in neighboring villages are more

likely to flee. To account for these influences, a spatial model is necessary.

7.4.1 Spatial Model

A simultaneous autoregressive model (SAR) accounts for spatial dependence by using

a regression on values from other areas in the vicinity of village i (Bivand et al 2008).

The spatial error model takes the following form:

Y = Xβ + λWξ + ε

where W is a matrix of spatial dependence and ξ is the spatial component of the

errors (Ward and Gleditsch 2008). To account for spatial dependence, I construct a

neighbor matrix in which village i is considered to neighbor all villages within 15km.24

The influence of each neighbor on village i is assumed to be equal, but I relax this

assumption with alternative spatial weights in the robustness tests.

For the analysis, the dependent variable is the log ratio of crop area in t1 to

crop area in t0 for each village, multiplied by 1,000 to scale the coefficients. The

key independent variable is now the distance to rebel headquarters (in km). To

account for how the effectiveness of control changes over distance, I model this variable

as a polynomial. I also include the distance to the rebels’ second largest bases—

Buchanan for NPFL and Tubmanburg for LURD—although neither of these bases

24The 15km cutoff ensures that all villages are connected to at least one neighbor, a requirementof spatial models, and is substantively significant because the distance corresponds to a half-day’swalk across difficult terrain.


Raw

Data

(1) NP

FL Territory

Distance from

HQ

(km)

% Change in Crop Area

0-6060-120

120-180180+

-300 -100 0 100 200 300

(2) LUR

D Territory

Distance from

HQ

(km)


0-6060-120

120-180180+

-300 -100 0 100 200 300

(3) NPFL v. LU

RD

Distance from

HQ

(km)

% Difference in Crop Area

0-6060-120

120-180180+

-150 -50 0 50 100 150

Matched Sam

ple

Distance from

HQ

(km)


0-6060-120

120-180180+

-300 -100 0 100 200 300

Distance from

HQ

(km)


0-6060-120

120-180180+

-300 -100 0 100 200 300

Distance from

HQ

(km)


0-6060-120

120-180180+

-150 -50 0 50 100 150

Figu

re7.4:

Chan

gesin

croparea

accordin

gto

distan

ceto

rebel

head

quarters.


Raw

Dat

a

Reb

el T

errit

ory


Combined

NPFL

LURD

-100-50050100

Mat

ched

Sam

ple

Reb

el T

errit

ory

% Difference in Crop AreaCombined

NPFL

LURD

-100-50050100

Fig

ure

7.5:

Eff

ect

ofru

bb

eron

chan

ges

incr

opar

ea.


was fully established at the time the satellite images were taken. To account for the

effects of lootable resources, villages are coded for whether they contain diamonds

or rubber. The information on diamond locations comes from DIADATA (Gilmore

et al 2005), and the location of rubber plantations was derived from the satellite

images. A non-lootable resource, timber, is also included in the models as a placebo

test. The locations of both active and potential timber areas are coded based on

information from Conservation International. Figure 7.6 depicts the distribution of

these resources.

To examine the effects of ethnic social networks on rebel behavior, I code villages

that are predominately ethnic Gio or Mandingo. These villages are most likely to

share social and family ties with NPFL and LURD members, respectively. The coding

is based on linguistic characteristics of village names,25, the 2008 Liberian census, and

survey data from Lofa County (Fearon et al 2009), where all Mandingo villages are

located.

Distance to the capital city Monrovia and distance to the nearest land border are

included to control for the cost of displacement among civilians. Civilians might be

more inclined to abandon their farmland if they can easily access the refugee camps of

neighboring countries or reach the capital city, the only location where humanitarian

aid organizations operated during the war. Distance to a major road is included to

control for ease of access. Locations near major roads might be easier for rebel groups

to monitor—or to loot. This variable is coded from a road network dataset compiled

by the Liberian Institute of Statistics and Geo-information Services.

Battles within 30km of village i since t0, the date of the first image, measures

the effect of instability on crop area and is coded from the ACLED data. Figure

7.7 depicts the locations of battles during January 1990 to January 1991 and during

February 2001 to March 2003. For the LURD rebels, a dummy variable indicates

villages that were already under LURD control in February 2001, when the first

satellite images were taken, coded based on U.N. security council reports as well as

field interviews with LURD commanders.

The number of households (logged) in a village was derived from a 2005 United

25Nearly all Gio villages end in -ay, while Mandingo village names usually end in -du or -dou.


X

X

X

XXX XXXX XX X XXXXX XXXXXXXX X XXX

X X

Buchanan

Monrovia

Tubmanburg

Voinjama

Gbarnga

X Diamonds

Plantations

TimberRegions

Source: DIADATA, Conservation International, satellite imagery analysis.

Figure 7.6: Natural resources in Liberia.


Nations study that provides rough estimates of the number of housing structures in a

given location. By counting structures, both destroyed and intact, the data are robust

to the effects of insecurity-related population movement. Finally, the percent crop

area at t0 and a measure of village area (in pixels, logged) are included to account

for unique village characteristics not related to the conflict. To limit the effects of

measurement error caused by noise in the satellite images, I exclude villages with 20

pixels or less of crop land in either image (each pixel measures 30m by 30m). These

villages are essentially non-agricultural locations.26 Table 7.4 presents descriptive

statistics of the variables.

7.4.2 Results

The same model was run for both NPFL and LURD areas (except for the LURD

2001 and Gio variables), and Table 7.5 presents the results side-by-side. Villages

closer to the NPFL headquarters at Gbarnga experience lower declines in crop area.

The effect is large and substantive and does not vary much when controls are added.

Holding constant other factors, if a farmer 100km away relocates to the the NPFL

headquarters, his crop area is expected to increase 236% in the basic model (Model

1) and 259% when controls are added (Model 2). An opposite and significant effect

is observed in LURD territory, although the results are more sensitive to model spec-

ification. A farmer relocating to LURD headquarters from 100km away experiences

a 70% decline in crop area in the basic model (Model 3) and a catastrophic 97%

decline in crop area in the model with controls. Proximity to LURD’s secondary

headquarters in Tubmanburg also corresponds to significant declines in crop area, as

does proximity to a major road in LURD areas. No significant effect was observed

for NPFL’s secondary headquarters in Buchanan.

In terms of lootable resources, no significant effect is observed for diamonds, but

diamond mining remained small-scale throughout the conflict. The presence of rub-

ber has a strong negative effect in NPFL territory and a significant, although smaller,

26This restriction excludes 11% of villages in the NPFL sample and 38% of villages in LURDterritory. Running the regressions without this restriction does not substantively affect results, butdoes introduce spatial autocorrelation caused by the measurement error.


1989 - 1991 2001 - 2003

[

!!!!!

!!!!

!!!

!! !!

!!

!!

!!!!!!

!

!

! !!!!

! !! !!

!! !

!

!

!

!

Monrovia

Buchanan

Tubmanburg

Voinjama

Gbarnga

[!!!!!!!!!!!!!!!!!

!!!!

!!!!

!!!!!!

!!!

!!!

!!!!!! !!!

!!!

!!!!!!

!!!

!!

!

!

!!

!

Monrovia

Tubmanburg

Voinjama

Gbarnga

Buchanan

! 1

! 5

! 10

Source: ACLED

Figure 7.7: Number and location of battles.

NPFL LURDVariable Min Max Mean St.Dev. Min Max Mean St.Dev.DV -2449 3055 44.42 659.8 -2932 2053 -366.2 710.0Dist. Rebel HQ 0.45 246.9 112.3 50.0 3.5 257.7 190.2 50.8Dist. 2nd Base 3.9 294.9 132.0 68.9 3.3 220.2 66.1 43.5Diamonds 0 1 5.4e-3 7.3e-2 0 1 7.7e-3 8.8e-2Rubber 0 1 8.4e-2 0.28 0 1 0.10 0.30Timber 0 1 3.0e-2 0.17 0 1 1.1e-2 0.10Gio 0 1 0.02 0.15 - - - -Mandingo 0 1 1.2e-2 0.11 0 1 3.9e-3 6.2e-2LURD 2001 - - - - 0 1 1.2e-2 0.11Dist. Capital 14.2 290.5 141.8 70.3 14.9 271.7 89.9 52.6Dist. Border 0.10 172.6 73.3 49.5 0.10 122.8 62.5 33.8Dist. Road 0.0 50.3 7.9 8.1 0.0 39.4 7.4 6.6Battles within 30km 0 21 2.1 3.4 0 11 3.5 3.1Village Households 1 5241 63.2 174.9 1 2051 62.2 128.9% Crop Area t0 2.6e-3 0.23 4.3e-2 3.0e-2 3.6e-3 0.28 3.7e-2 2.5e-2Village Area 7.4e5 3.8e8 1.5e7 2.5e7 1.1e6 3.3e8 1.5e7 2.6e7

Table 7.4: Descriptive statistics.


NPFL LURD1986 v. 1991 2001 v. 2003

Model 1 Model 2 Model 3 Model 4Dist. to Rebel Headquarters -11.2** -12.8*** 16.0* 57.0**

(3.76) (3.35) (7.37) (19.6)Dist. Rebel HQ2 0.026* 0.033* -0.042* -0.205***

(0.015) (0.013) (0.022) (0.044)Rubber -286*** -289*** -134* -93.0*

(30.6) (30.9) (54.0) (56.0)Diamonds -68.7 94.4

(78.8) (133)Timber 52.5 191

(61.0) (216)Gio 15.2 -

(50.7) -Mandingo -35.7 130

(65.2) (427)LURD 2001 - 175

- (271)Dist. to Rebel Secondary Base 2.04 19.7**

(3.81) (7.58)Dist. 2nd Base2 0.006 -0.047

(0.010) (0.100)Distance to Capital -4.37 -52.9***

(3.03) (11.4)Dist. Capital2 -0.001 -0.161

(0.010) (0.104)Distance to Border 5.10 24.0*

(3.11) (13.3)Dist. Border2 -0.036* -0.139*

(0.020) (0.0.075)Distance to Road 2.35 22.1***

(2.74) (6.22)Dist. Road2 -0.249* -0.961**

(0.108) (0.298)Battles within 30km since t0 -3.20 -4.29 -2.38 -2.55

(2.80) (2.83) (7.20) (7.23)Village Households (log) 4.13 -0.933

(4.69) (10.4)% Crop Area t0 (log) -839*** -839*** -835*** -804***

(12.3) (12.5) (29.7) (30.1)Village Polygon Area (log) -76.6*** -79.2*** -138*** -138***

(8.97) (9.24) (19.4) (19.6)(Intercept) -773** -516 -2655*** -3061

(270) (502) (721) (3478)N 4430 4430 1293 1293λ 0.924 0.888 0.900 0.967LR Test Value 1794.8 1059.2 497.1 253.7λ p-value <2.22e-16 <2.22e-16 <2.22e-16 <2.22e-16Moran’s I Std. Deviate 0.1805 0.8851 3.059 0.6885Moran’s p-value 0.4284 0.1880 0.0011 0.2456

Note: Dependent variable is log ratio of crop area multiplied by 1,000.∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.10

Table 7.5: Spatial model of changes in crop area for NPFL (1986 v. 1991) and LURD(2001 v. 2003).


negative effect in LURD territory. In NPFL territory, being in a rubber area corre-

sponds to a 25% decline in crop area in either specification; in LURD territory the

decline is 12.5% in Model 3 and 9% in Model 4. The effects of ethnic links to villages

and nearby battles are not significant in the model, although the coefficients are in

the predicted directions. The high value of λ indicates that spatial autocorrelation

exists between the observations, but a Moran’s I test on the residuals reveals that

the spatial weights reduced the autocorrelation to the point where it is no longer

significant in Models 1, 2, and 4; spatial autocorrelation remains a concern in Model

3.

7.4.3 Robustness

Spatial models depend on assumptions regarding the structure of the spatial depen-

dence. Examining the results under a variety of spatial weights and specifications

provides some insight into the robustness of the results. In addition to the 15km

neighbor matrix, I constructed matrices that weight neighbors according to the in-

verse distance between them and the inverse distance squared. These weights imply

that villages farther apart have less influence on each other. Each of these weights,

including the basic neighbor matrix, was also extended to the 30km range.

The effects of proximity to rebel bases and rubber areas in NPFL territory re-

main as predicted, and their statistical significance increases under these alternative

specifications (results not shown). The effect of battles in NPFL territory remains

negative and becomes statistically significant, but the increased significance could be

explained by the spatial autocorrelation that remains in the model under these al-

ternative spatial weights. In LURD territory, distance to rebel headquarters remains

significant in the predicted direction for all weights matrices. Rubber and battles are

significant and negative in the basic 30km neighbor matrix, but lose their significance

in models using the distance-based weights.

Spatial regressions are sensitive to the specification of key variables (Anselin 2002).

In the model, I specified the distance variables as polynomials to account for non-

linear effects. To see if this specification is justified, a generalized additive model


(GAM), which estimates the functional form of key variables, is a useful tool. The

GAM plots, which can be found in the chapter’s appendix, indicate that the effects

of distance to rebel headquarters in both NPFL and LURD territory can be modeled

as a quadratic within 100km of the rebel headquarters; beyond this distance, model

specification may be problematic.

Running the spatial models on the subset of villages located within 100km of the

rebels’ headquarters adds confidence to the estimates. Table 7.6 lists the results,

which are broadly consistent with the original models.27 According to Model 1 a

farmer 100km away from NPFL headquarters would benefit from a 181% increase in

crop area, while Model 2 estimates a 245% increase. The results provided by Model

2 are nearly identical to the results using the full dataset, which estimate increases of

236-259%. Rubber in NPFL territory is estimated to correspond to a 22% decline in

crop area in either specification, which is very similar to the 25% decline estimated

in the full dataset. The results for LURD territory are also similar.28 In Model 3

a farmer relocating to LURD’s headquarters from 100km away would suffer a 51%

decline, while Model 4 estimates an 80% decline. This last result falls in the range of

70-97% estimated by the full dataset.

A major challenge of spatial analysis is the Modifiable Areal Unit problem (Hain-

ing 2003). Statistical results may be sensitive to how the variables are spatially

aggregated (Kalyvas 2008). To address this issue, I aggregate crop area at the level of

clans, the smallest administrative unit in the Liberian government, as an alternative

to the village-level analysis (not shown).29 Distance to rebel headquarters remains

statistically significant in the predicted direction for NPFL territory, as does the

presence of rubber plantations. In the LURD model, the coefficient for distance to

headquarters loses statistical significance but retains the predicted sign. The presence

27At first glance the results appear contradictory, but this is due to the large coefficient on thesquared distance.

28Only 100 villages fall within 100km of LURD’s headquarters at Voinjama; several variables hadto be excluded to prevent singularity in the model.

29There are 832 clans in Liberia. Village-level household figures were replaced by the numberof households in each clan, according to the 2008 census conducted by the Liberian government.Several variables, including distance to capital and distance to secondary base, had to be removedto prevent singularity.


NPFL 100km LURD 100km1986 v. 1991 2001 v. 2003

Model 1 Model 2 Model 3 Model 4Dist. to Rebel Headquarters 7.24 8.65 33.6* 61.3***

(6.86) (6.33) (14.5) (18.5)Dist. Rebel HQ2 -0.13** -0.18*** -0.264* -0.453***

(0.047) (0.048) (0.125) (0.132)Rubber -254*** -247*** - -

(33.3) (34.0) - -Diamonds -93.2 -

(113) -Timber 41.9 88.1

(144) (126)Gio -29.7 -

(59.6) -Mandingo - 634*

- (370)LURD 2001 - 72.9

- (214)Dist. to Rebel Secondary Base -12.5 -218***

(10.2) (56.2)Dist. 2nd Base2 0.046 0.661***

(0.037) (0.168)Distance to Capital -29.4** -

(9.48) -Dist. Capital2 0.080** -

(0.031) -Distance to Border 11.5* 71.2***

(5.17) (18.0)Dist. Border2 -0.106* -1.19***

(0.042) (0.274)Distance to Road 1.39 18.4

(3.95) (16.3)Dist. Road2 -0.225 -0.130

(0.157) (0.438)Battles within 30km since t0 8.27 7.28 -18.9 -11.7

(5.09) (5.10) (27.8) (31.9)Village Households (log) -0.243 -68.4*

(5.80) (32.5)% Crop Area t0 (log) -727*** -730*** -971*** -769***

(18.2) (18.6) (84.7) (114)Village Polygon Area (log) -29.8** -29.4* -117* -114*

(11.4) (11.7) (62.8) (68.5)(Intercept) -1727*** -1474 -3211** -13698**

(335) (1280) (1019) (4853)N 1854 1854 100 100λ 0.946 0.921 0.529 -0.361LR Test Value 822.1 588.71 10.537 1.7212λ p-value <2.22e-16 <2.22e-16 0.0011 0.1895Moran’s I Std. Deviate 2.7205 2.8604 0.6029 -0.1917Moran’s p-value 0.0033 0.0021 0.2733 0.576

Note: Dependent variable is log ratio of crop area multiplied by 1,000.∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.10

Table 7.6: Changes in crop area within 100km of rebel headquarters for NPFL (1986v. 1991) and LURD (2001 v. 2003).


of rubber is negative and statistically significant at the clan level in LURD territory.

7.5 Changes Over Time and the Endogeneity of

Rebel Bases

Additional information can be gained by examining changes in crop area over time

in the smaller study area depicted in Figure 7.2. SPOT images from 1993 and 1995

allow for a comparison of crop area in NPFL territory during 1991, 1993, and 1995

and within LPC territory during 1995.30 This framework provides a comparison of

security in NPFL versus LPC territory and addresses, to some extent, the endogenous

location of rebel bases. Instead of organizational challenges, security in rebel territory

might be driven by the strategic decision to site rebel bases in particular areas, or by

other omitted variables related to geography. The NPFL’s secondary headquarters

was located in Buchanan until the town fell to the LPC rebels in 1994. By measuring

the changes in crop area near Buchanan before and after this change in authority,

omitted variables related to the location of Buchanan are eliminated and the effect

of rebel behavior on local food production can be identified.

Table 7.7 depicts the results of the analysis. The first three columns apply to

NPFL territory while the final column describes the situation in LPC territory. The

geographically-compact nature of the study area and its demographic characteristics

allow several control variables to be dropped.31 The NPFL was in the process of

reorganizing its forces in January 1991, depicted in the first column. The effect of

distance to Buchanan, its secondary base, has the predicted sign but fails to reach

statistical significance. In January 1993, however, after the NPFL’s presence was

firmly established, there is a strong, positive effect on crop area near the Buchanan

base. This effect disappears in January 1995, in the third column, which is expected

30These observations could be combined into a panel model by adding a time component to thespatial weights matrix, but this specification would require additional assumptions that are difficultto justify on theoretical grounds.

31In addition, the model relaxes the restriction of villages with less than 20 pixels in both timeperiods to include villages that contain more than 20 pixels of crop land in either t0 or t1. Again,the results are not substantively affected when this restriction is dropped.

7.6. ASSESSING THE EVIDENCE 249

because the NPFL lost control of Buchanan to the LPC in 1994. Civilian abuse is

higher in the rubber areas in 1991 and 1995, but the coefficient switches signs in

1993. This temporary positive effect of rubber could reflect either modeling errors or

greater NPFL security in the plantations due to the 1992 agreement with Firestone

corporation, which led to a resumption of large-scale rubber extraction during this

time. Being closer to a main road is beneficial for civilians in NPFL territory during

1993 and 1995, which could imply better monitoring and control of rebel forces near

the highway.

The estimates for the LPC also support the theory’s predictions. Greater declines

in crop area occurred closer to the LPC’s secondary base in Buchanan.32 Greater

declines also occurred closer to the main roads in LPC territory, which could indicate

a greater opportunity for looting. Even though the LPC was ethnically homogeneous,

the group exhibits the characteristics of a predatory organization, as predicted. For

both groups, battles had a negative effect on civilians, but the coefficient only reaches

statistical significance in 1993. The changing levels of security in NPFL versus LPC

territory reveal that the results cannot be explained by omitted variables related to

the strategic selection of rebel bases or any geographical characteristics of the area.

Rather, group behavior explains the level of security provided to civilians.

7.6 Assessing the Evidence

The qualitative and quantitative evidence from Liberia’s war supports the theory

of commander incentives compared to existing theories. As predicted, the NPFL

rebels, with access to leader financing and credible promises, provided more security

than ULIMO, LPC, LURD, and MODEL (Hypotheses 1, 3, 4). Rubber, a lootable

resource, was associated with less security (Hypothesis 2). Security in NPFL territory

deteriorated in unstable and difficult-to-monitor areas, while monitoring had little

effect on troop behavior in the other groups (Hypotheses 5, 6). For the LURD and

LPC rebels, the most predatory behavior occurred in the vicinity of major bases

(Hypothesis 7).

32The rebels were headquartered in Greenville, Sinoe County.


NPFL NPFL NPFL LPC1991 1993 1995 1993v. 1995

Dist. to Rebel Secondary Base -7.94 -29.0*** -5.60 45.9*(7.69) (8.53) (18.6) (18.6)

Dist. 2nd Base2 0.100* 0.147* 0.181 -0.487*(0.059) (0.063) (0.125) (0.282)

Rubber -248** 237* -333*** -169(93.9) (107) (96.2) (181)

Timber 103 -92.4 257 -126(103) (118) (184) (172)

Distance to Road -11.4 -52.4*** -25.8* 34.1*(8.94) (8.76) (10.3) (19.4)

Dist. Road2 0.223 2.33*** 0.916* -0.976(0.447) (0.434) (0.520) (1.13)

Battles within 30km since prev. - -55.6** -13.4 -45.2- (17.7) (35.0) (63.2)

Village Households (log) 0.736 8.01 -1.74 -5.25(12.5) (12.4) (15.7) (19.8)

% Crop Area Previous (log) -536*** -898*** -896*** -480***(40.4) (35.9) (48.4) (51.2)

Village Polygon Area (log) -131*** -80.0** -87.5** 20.6(25.9) (26.1) (28.5) (41.6)

(Intercept) 64.4 -1205** -2414** -3212***(403) (446) (874) (733)

N 615 436 409 258λ 0.736 0.795 0.948 0.826LR Test Value 20.9 36.3 80.7 9.60λ p-value 4.9e-6 1.7e-9 <2.22e-16 1.9e-3Moran’s I Std. Deviate -0.836 0.420 1.66 1.32Moran’s p-value 0.798 0.337 0.0483 0.0929Note: Dependent variable is log ratio of crop area multiplied by 1,000.∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.10

Table 7.7: Change over time in a section of NPFL territory and spatial variationwithin a section of LPC territory.

7.6. ASSESSING THE EVIDENCE 251

These spatial patterns of civilian security, based on measurements of changing

crop area, cannot be explained by the endogenous choice of rebel bases or other ge-

ographic factors. While crop area benefitted from proximity to Buchanan during

NPFL rule, it suffered after the town fell to the LPC rebels, even though geographic

factors remained constant. The evidence poses a challenge to existing theories of

rebel behavior. Theories of ethnic composition and member selection cannot explain

the relative discipline of the ethnically diverse, resource-rich NPFL compared to the

ethnically-homogeneous, resource-poor LPC or MODEL (Competing Hypotheses 4

and 6). And theories of territorial competition cannot explain why even the most

stable areas under LURD control suffered more than 50% declines in crop area (Com-

peting Hypothesis 1). The next chapter summarizes the theoretical contributions of

the dissertation and offers some policy implications.


7.A Appendix

Area % Crop Area t0 Battles Within 30km0-60km

Mean NPFL 13144688 0.052973 1.1774Mean LURD 12999278 0.052602 1.1774Std. Bias 1.18 2.33 0K-S Statistic 0.17 0.21 0

60-120kmMean NPFL 14280491 0.03495 1.1765Mean LURD 14345221 0.03516 1.1765Std. Bias -0.28 -1.16 0K-S Statistic 0.17 0.069 0

120-180kmMean NPFL 12421513 0.03062 1.8365Mean LURD 12544698 0.030704 1.9231Std. Bias -0.63 -0.44 -5.27K-S Statistic 0.10 0.084 0.083

180+kmMean NPFL 9288780 0.034403 0.63125Mean LURD 9164558 0.035442 0.63125Std. Bias 2.07 -4.72 0K-S Statistic 0.05 0.16 0

Rubber NPFLMean Rubber 5565896 0.056161 2.1982Mean Non-Rubber 5528284 0.055912 2.1982Std. Bias 0.85 0.65 0K-S Statistic 0.08 0.015 0

Rubber LURDMean Rubber 6359800 0.038278 4.0838Mean Non-Rubber 6553879 0.03808 4.0838Std. Bias -4.12 0.94 0K-S Statistic 0.11 0.060 0

All VillagesMean NPFL 12881579 0.0374 1.7158Mean LURD 12666193 0.03762 1.7158Std. Bias 1.07 -0.87 0K-S Statistic 0.041 0.046 0

Table 7.8: Balance statistics for the matched samples.

7.A. APPENDIX 253

050

100

150

200

250

-2000200400600

DistanceHQ

s(DistanceHQ)

050

100

150

200

250

-800-4000200400

DistanceHQ

s(DistanceHQ)

NPFL

LURD

Fig

ure

7.8:

GA

Mplo

tsof

the

effec

tof

dis

tance

from

reb

elhea

dquar

ters

onch

ange

sin

crop

area

.

Chapter 8

Conclusion

Why do some groups provide security and promote local production, while others

loot and prey on the population? Why do some groups break into competing, violent

factions, while others remain unified?

In my attempt to answer these questions, I focus on the relationships that exist be-

tween rebel leaders and their troops, especially their top commanders. Rebel groups

must deal with the same principal-agent challenges that exist in any organization.

A leader sets a policy and offers incentives to subordinates, who then compare the

leader’s incentives with their best outside options. But during conflict, when rebel

leaders do not have access to externally-enforced contracts or credit markets, these

agency relationships have deep implications for group behavior. By exploring these

implications and testing the consequences both across conflicts and within Liberia,

this dissertation makes two broad theoretical and empirical contributions to the study

of rebellion. The theory also offers several policy implications for international inter-

ventions to end civil wars.

8.1 Contributions of the Dissertation

The dissertation’s two contributions relate to the effects of resources during conflict

and the role of external supporters of rebellion. First, the lack of externally-enforced

contracts means that different types of resources have different effects on the leader’s

254

8.1. CONTRIBUTIONS OF THE DISSERTATION 255

ability to control subordinates. When leader’s have direct access to financing, they

can create incentives for cooperation by offering cash payments. But when revenues

are collected by commanders or other subordinates—what I call lootable resources—

the leader’s ability to create incentives is weakened. Leaders can overcome these

problems to some extent with promises of future rewards. But without enforceable

contracts, the credibility of these promises often depends on the social bonds between

leaders and their top commanders. Across civil wars, I show that leader financing

and leader credibility are systematically related to less civilian abuse and a lower risk

of group factionalization, while lootable resources are associated with a substantial

increase in these types of dysfunctional behavior.

These dynamics also play out within civil wars. The effectiveness of spot pay-

ments depends on monitoring. Even when leaders can induce cooperation through

cash payments, they struggle to restrain their troops in unstable and remote areas.

When leaders do not have access to financing, they allow group members to fend for

themselves, even near rebel bases where predatory behavior is especially harmful to

the group. In Liberia, I show that civilian security, as measured through changes

in crop area, followed these spatial patterns. Civilians near the NPFL headquarters

were able to continue farming at pre-war levels, while civilians near the LURD head-

quarters suffered catastrophic declines in local food production. Also hard hit were

villages in contested areas or near lootable resources.

The different effects of leader financing versus lootable resources is a fundamental

departure from standard principal-agent theory, in which leaders can contract over

how revenues are shared. For example, the owner of a car dealership does not have

to worry that the salesmen will loot the vehicles or pocket the sales revenue. If the

salesmen steal or embezzle funds, the owner can call the police or take legal action.

These external enforcement mechanisms allow the owner to incentivize the salesman

through salaries and commissions. Increasing the value of the vehicles—switching

from Mazda to Maserati, for example—does not diminish the owner’s ability to in-

centivize the subordinates. For their part, the salesmen know that they can sue the

owner if the promised commissions do not materialize. When contracts are externally-

enforced, there is no difference between leader financing and lootable resources. This

256 CHAPTER 8. CONCLUSION

is not the case during conflict.

The second contribution of the dissertation is a theoretical framework for under-

standing the role of external actors in rebellion. External patrons provide weapons,

leader financing, and other types of support. This support often makes the difference

between groups that are able to challenge the state in a large-scale rebellion and those

that fade into obscurity. Because rebel leaders cannot access credit markets, external

patrons can influence the characteristics of the rebel group and the resources avail-

able to them. When patrons support rebellion to increase their bargaining leverage

in international affairs, they have incentives to support relatively low quality lead-

ers and provide relatively little leader financing. These motives are most common

among state patrons, and I show that state-sponsored rebel leaders have less access

to financing and that these rebels are more abusive and more prone to factionaliza-

tion than rebels without state support. By contrast, patrons with preferences that

align with those of the rebels, such as diaspora communities, tend to support more

trustworthy leaders and provide greater access to financial resources. These rebels

are more disciplined and less prone to factionalization.

Nearly all of the theory’s implications, stated in 12 hypotheses, find support in the

qualitative and quantitative analyses. The evidence supports Hypotheses 1-10, which

cover the effects of leader financing and credibility, lootable resources, monitoring,

and the influence of external patrons. The data do not provide convincing support

for the claims that higher quality leaders emerge in the absence of external support

(Hypotheses 11) or that groups with multiple patrons tend to be less abusive (Hypoth-

esis 12). Competing theories do not fare as well. Theories of member selection and

costly signaling are challenged by the increased disciplined that accompanies leader

financing and support from patrons with aligned preferences. Theories of territorial

competition and ethnic composition find some support, but also have limitations.

Theories of territorial competition, for example, offer little guidance for predicting

levels of civilian abuse across conflicts. Applied to Liberia’s war, the theories offer

some insight but fail to explain important variation in rebel behavior. Contrary to

theories of territorial competition, soldiers in uncontested parts of LURD territory

8.1. CONTRIBUTIONS OF THE DISSERTATION 257

preyed indiscriminately on civilians. And contrary to theories of ethnic composi-

tion, the ethnically-homogenous forces of LPC, MODEL, ULIMO-K, and ULIMO-J

were generally abusive, while the ethnically-diverse forces within NPFL behaved with

restraint in some areas. Ethnic composition did, however, seem to influence the be-

havior of MODEL’s forces in areas where they shared family and ethnic ties to local

communities.

Despite this evidence, the dissertation’s contributions are vulnerable to an omitted

variable bias. Perhaps rebel group behavior does not result from resources and the

influence of external patrons, but rather from some factor that remains unidentified.

This unidentified factor, if it is to undermine the theory, would need to simultaneously

cause group members to be disciplined and rebel leaders to be more successful at

attracting financial support from patrons with aligned preferences. If such a factor

exists, the correlation between leader financing and group behavior would have no

causal significance. Nor would the correlations among patron characteristics and

group behavior.

Although I cannot fully address this potential bias, I use several tools to add

confidence to the theory and diminish the possible influence of any omitted factor.

First, I use exogenous proxy variables for leader financing and lootable resources.

These proxies are based on pre-war geographic endowments that are not affected by

rebel group characteristics or the motives of external patrons. The effects of these

variables on rebel behavior can only flow through the group’s access to resources.

Consistent with the theory, I find that capital-intensive resources—a proxy for leader

financing—are associated with less abusive groups while low-capital resources are

associated with more abusiveness. Second, I use an in-depth analysis of Liberia’s

civil war to understand the selection of group members and leaders, and the causal

effect of resources on group behavior. Each of Liberia’s rebel groups emerged from a

pre-existing cadre of trained commanders and exiled politicians, who selected a leader

based on the potential for attracting external support. To be persuasive, an omitted

variable would have to not only cause differences in the inherent preferences among

these otherwise similar groups, but also explain why group members with different

preferences were deployed according to a spatial pattern that depends on distance to


rebel headquarters and the presence of lootable resources. Discovering such a variable

is unlikely, at least in Liberia.

8.2 Policy Implications

The theory has several policy implications for how international actors can design

interventions in civil wars, but also highlights the potential negative consequences of

interventions that are not carefully crafted to the particular situation.

Policy Implication 1. Sanctions on rebel sources of revenue should not be used

in isolation, but rather as an integrated element of a

broader intervention package.

The theory challenges policy makers to adopt a more nuanced understanding of

the role of resources in civil war. Not all resources have the same effects on rebel

behavior, and targeting one revenue stream might cause the rebel group to adapt in

ways that increase the group’s abusiveness and heighten the risks of factionalization.

For example, if a leader who depends on diaspora contributions finds his main source

of revenue suddenly interrupted, the leader might allow soldiers to loot as a way of

maintaining group membership. The same dynamic could apply to capital-intensive

natural resources or other assets controlled by the leader. Recalling the example

of Ituri, the FAPC rebels embarked on a weeks-long looting spree after their leader

transferred control of the Ariwara gold market to the government and ceased paying

salaries to the troops.

Sanctions against lootable resources are less likely to cause perverse effects since

these resources already contribute to predatory behavior and factionalization. But im-

plementing effective sanctions against lootable resources has proved difficult. Nearly

anyone with a shovel can collect alluvial gold, diamonds, or gemstones, and these pre-

cious stones are easily hidden during travels to areas where sanctions are not enforced.

For example, rebels and civilians in the Central African Republic avoid the Kimberley

process by selling their stones in neighboring Cameroon, Sudan, DR Congo, and Re-

public of Congo; larger traders simply export the stones via Bangui airport to cutting

facilities in Dubai and elsewhere (ICG 2010). Further, sanctions have no effect on

8.2. POLICY IMPLICATIONS 259

illicit goods such as narcotics, which are already the focus of international interdiction

efforts.

These challenges and potential adverse effects, however, do not imply that sanc-

tions should be abandoned. Rather, the financial leverage gained from sanctions

should be used in concert with a broader intervention package that includes a peace

agreement, demobilization benefits for the troops, and possibly the deployment of

peacekeepers. The theory shows that rebel leaders enforce discipline and cohesion

through spot payments, as well as promises of future reward. If the rebel leader knows

that the financing for spot payments will soon be sanctioned, she may be more willing

to agree to a peace deal in the short-term. In this situation, the post-war benefits

from the peace process would substitute for spot payments to the troops. Generous

demobilization packages can entice rebel members to abandon their arms, especially

if their access to payments from the leader or lootable resources is diminished. More

forceful interventions, such as the deployment of peacekeepers, may increase the costs

of continuing the war while also increasing the effectiveness of sanctions. Without

such a comprehensive package, the leader may respond to sanctions by loosening her

control in the organization and tolerating a higher level of looting and abuse.

These combined interventions are more likely to succeed when the rebel leader is

dependent on a revenue stream that can be effectively curtailed by sanctions, and

when the payout from demobilization exceeds the revenues earned by individual rebel

soldiers during the war. Such a strategy, like all negotiations during war, faces a

risk of failure and is open to manipulation. The 2004-5 demobilization program in

Liberia, for example, processed over 100,000 individuals for a war that involved fewer

than 20,000 combatants. The main challenge of this approach is inspiring the high

levels of international commitment and coordination required to implement it.

Policy Implication 2. International actors can influence rebel behavior by pro-

viding post-conflict rewards to commanders, but should

be sensitive to potential adverse effects of these policies.

When commanders expect to receive post-conflict rewards, they commit less civil-

ian abuse and are less likely to factionalize. Peace negotiations are especially difficult

for rebel groups. If commanders lose faith in the leader’s promises of rewards, they


might break from the group and jeopardize the peace process. Rebel leaders, how-

ever, have an incentive to maximize their own payout and might decide to exclude

the commanders from post-conflict rewards, even if this strategy increases the risk

of factionalization. International actors can shape the behavior of rebel groups and

increase the chances of a successful peace agreement by pressuring rebel leaders to

provide government positions to top commanders, or by creating programs to provide

commanders with other rewards.

International actors already implement programs that provide payments and vo-

cational training to demobilized soldiers. Also common are programs to integrate

rebel members into the state’s armed forces. These programs, however, generally

focus on low-level fighters and rarely recognize the importance of high-level rebel

commanders in the peace process. Top commanders are unlikely to be satisfied by

the same demobilization package given to their subordinates. The effectiveness of

these programs is further limited by their lack of conditionality. The rewards do not

depend on the commander’s actions during the war: disciplined commanders cannot

expect more rewards than abusive commanders. While these programs might increase

the chances for peace, they do not influence rebel behavior during the war. Aware

of these limitations, the international community has begun to use indictments from

the International Criminal Court and other forms of justice as a deterrent to abusive

behavior during war. The effectiveness of these programs, however, has yet to be

demonstrated.

Liberia provides a good example of the danger posed by top commanders, even

when demobilization programs are in place. After being excluded from offices in

the transitional government, commanders in both LURD and MODEL attempted

to overthrow their respective leaders. The swift decision by the United Nations to

support the rebel leaders as signatories of the peace agreement prevented a return to

war, but did nothing to assuage the frustrations of the commanders. The commanders

remained a threat to Liberia’s stability until a new program was created that paid a

large monthly stipend to 16 members from each faction and organized a ceremony to

acknowledge the commanders’ important role in the peace process.

These interventions can produce negative consequences. Civilian support for the


new post-war government might be undermined by the inclusion of abusive comman-

ders who terrorized the population. From the rebels’ perspective, these programs

might encourage the groups to inflate the number of top commanders in their ranks,

in an attempt to attract a bigger payout. International actors might be confronted by

hundreds of “senior commanders”, making more difficult the task of identifying the

important players who should be rewarded. In the most extreme cases, the benefits

provided to top commanders could diminish the rebel leader’s incentive to accept the

peace agreement. If the division of spoils between leader and commander are zero-

sum, any additional benefits provided to the rebel commander are subtracted from

the leader’s payout. International actors should take steps to ensure that comman-

ders receive enough rewards to lay down their arms, but these interventions should

be carefully tailored to the specific conflict to minimize the adverse effects.

Policy Implication 3. Humanitarian organizations should negotiate directly

with the rebel leader for access to rebel territory. If the

leader is unable to provide sufficient security, the organi-

zations should consider suspending operations.

It is an open secret that humanitarian organizations negotiate with rebel groups

for access to conflict areas, and that rebel groups often benefit from the aid delivered.1

By introducing resources, humanitarian aid has the potential to intensify the conflict

even while providing support to civilians in need. The aftermath of the Rwandan

genocide revealed the potential adverse effects of humanitarian aid, as Hutu militants

used the relief supplies to regroup their armed forces and solidify their control over

the civilian population. Other examples of humanitarian aid’s potential for fueling

war include Somalia and Cambodia.

Adverse effects are most likely to occur when humanitarian aid becomes a lootable

resource that is accessible to commanders without leader oversight. In these situa-

tions, local commanders use humanitarian aid to further their own short-term in-

terests, even at the expense of group goals. Commanders, for example, may have

an incentive to exacerbate the humanitarian situation in their areas in the hopes of

1The United Nations, for example, has published a manual and a set of guidelines for humani-tarian organizations that negotiate with armed groups. See McHugh and Bessler (2006).


attracting more aid. Or commanders may arbitrarily tax or loot the aid supplies,

depending on their immediate needs. By negotiating with these commanders, hu-

manitarian organizations increase the incentives for civilian abuse and the risks of

group factionalization.

Of course, direct contact with the rebel leader also carries the risk of adverse

effects. Rebel leaders, like commanders, might have an incentive to exacerbate the

humanitarian situation to increase the flow of aid or gain political leverage. I argue,

however, that rebel leaders are more likely to have an incentive to provide security in

their areas—and this claim is supported by the empirical evidence. By negotiating

directly with the rebel leader, humanitarian organizations strengthen the incentives

for security and also increase the leader’s ability to maintain discipline within the

group. Ideally, humanitarian organizations can gain access to the conflict areas with-

out providing material support to the rebel leader; the political benefits to the leader

should be enough incentive to facilitate humanitarian relief efforts.

For some rebel groups, however, the leader’s assurances are not enough to guaran-

tee security. Even with the leader’s permission, aid organizations might have to strike

additional deals with local commanders. In these situations, the leader is not able to

exert effective control over group members and humanitarian aid becomes a lootable

resource. Whenever local commanders have independent discretion over the activities

of humanitarian organizations, the organizations should consider suspending their op-

erations in the conflict areas. The decision to end humanitarian operations is never

an easy one, and it raises difficult ethical questions regarding the civilians left be-

hind. But the aftermath of the Rwandan genocide and other humanitarian debacles

have created a growing consensus around the first rule of humanitarian intervention:

“First, do no harm” (e.g. Anderson 1999). When aid becomes another lootable re-

source that fuels civilian abuse and increases the risk of group factionalization, harm

is being done.

Policy Implication 4. Effective control over arms and ammunition is the most

direct way of preventing rebellion.

The attention surrounding natural resources in civil wars often obscures the basic

fact that arms and ammunition are the critical resources necessary for rebellion.


Indeed, diamonds are not capable of killing anyone—but the people with guns who

control the mines definitely are. Without military supplies, no rebel group is capable

of challenging a modern state or seizing control of natural resources.

The Liberia case study reveals how rebellion hinges on access to weapons. The

commanders who later formed the NPFL languished in refugee camps for more than

two years before they found a leader who could secure access to weapons. The LURD

rebels were limited to small-scale hit and run attacks during 1999-2000 because they

had to rely on captured weapons and small amounts of ammunition purchased from

Guinean soldiers. It took three years for Thomas Yaya Nimely to form the MODEL

rebels because he was unable to find a source for arms until the government of Cote

d’Ivoire agreed to provide support.

Controlling the arms trade may seem impossible, but, as Chapter 3 argues, in-

surgents benefit most from large-scale shipments that can only come from states and

large-scale arms dealers. Targeting these actors is more effective and more feasible

than trying to eliminate the diffuse networks of small-scale arms smugglers. Con-

certed efforts to block large-scale arms transfers can prevent, or reduce, conflict. The

interdiction by the U.N. of a large arms transfer to Charles Taylor in mid-2003, for

example, is widely credited with forcing Taylor to step down as president and end-

ing Liberia’s war. Nevertheless, without sufficient international commitment, arms

embargoes are often ineffective.

Perhaps a better approach is to intensify efforts to regulate ammunition. Unlike

weapons, which can be used for decades, ammunition must be constantly replenished.

According to an in-depth study by the advocacy group Small Arms Survey (SAS

2006), there are three steps that international actors can take to reduce insurgents’

access to ammunition. First, the international community should develop a system

for marking ammunition and keeping records of transfers. This tracking system will

increase the ability of regulators to identify points of loss or diversion, and also allow

researchers in the conflict area to identify the origin of ammunition, paving the way

for sanctions. Second, the international community should promote safe and secure

storage of ammunition, particularly in transitional and conflict-prone countries. These

storage facilities could become a standard element of military assistance packages from


countries such as the United States, France, and Britain. Third, the international

community should ensure the rapid destruction of the substantial stocks of surplus

ammunition. These stores of ammunition are not only a public safety hazard due to

the risk of explosion, but also increase the opportunities for large-scale diversion of

surplus ammunition to insurgent groups.

Policy Implication 5. State-sponsorship of rebellion should be discouraged

through tough international sanctions and other punitive

measures.

State-sponsored rebel groups tend to be more abusive towards civilians and more

prone to factionalization. Although some states support rebellion to achieve ethnic

or even humanitarian objectives, most states use rebel groups as a means of achieving

foreign policy objectives that do not relate to the well-being of civilians in the conflict

areas. The international community currently adopts ad hoc policies when confronted

by evidence of state support for rebellion. For example, after uncovering evidence that

Eritrea was providing support to Somali militants, the United Nations and other

international actors imposed an arms embargo as well as travel restrictions and asset

freezes on Eritrean officials. But not all states are punished for their support of

rebellion. Rwanda, for example, has escaped sanctions despite overwhelming evidence

of its support for various insurgencies in DR Congo.

Ideally, the international community would create a standardized mechanism for

applying sanctions to any state that supports rebellion without international approval.

This mechanism would discourage states from using rebellion to gain political lever-

age, but still allow support for rebels whose goals are supported by the international

community. An effective sanctions regime, however, is difficult to implement under

the current United Nations system. Vetoes would prevent sanctions against powerful

states or regimes that are friendly to permanent members of the security council such

as China, Russia, and the United States. Nevertheless, tougher actions against state

sponsors will discourage rebellion, decrease average levels of abuse, and reduce the

number of factions, making conflicts easier to resolve.

8.3. LOOKING FORWARD 265

8.3 Looking Forward

In many parts of the world today, people live in areas controlled by armed groups or

in weak states where the risk of rebellion is ever-present. For the foreseeable future,

civil war will be a common feature of the world system, with profound humanitarian

consequences. Some rebel groups, despite significant challenges, maintain disciplined

forces and provide security to civilians. This security allows people, for the most

part, to remain in their villages and produce food and other essentials. Other groups

prey on the population, contributing to widespread displacement, malnutrition, and

war-related disease. This dissertation attempts to explain the order—albeit a violent

order—that sometimes emerges out of civil war. By connecting natural resources

and the motives of external patrons to the organizational challenges faced by rebel

groups, the theory uncovers the logic behind rebel behavior. Hopefully this theory will

inspire debate and lead to further progress in the rapidly developing scholarship on

conflict. Armed with this understanding, international actors will be better equipped

to mediate successful peace agreements and mitigate the tragic humanitarian costs of

war.

Appendix A

Fieldwork in Liberia

This appendix provides additional information on the research that I conducted in

Liberia, including my background in the country, randomized fieldwork in villages,

and interviews with key actors in the conflict. My first experiences in Liberia were

guided by a group of Liberian friends, some of whom were recently returned exiles

while others had remained in Liberia throughout the war. This social network pro-

vided a unique insight and entree into Liberian society and politics. Their perspectives

differed significantly from the opinions that I heard from expatriate aid workers, as

well as the international media’s portrayal of the Liberian war. By following their

lead, I learned how to abide by cultural norms, whether I was interviewing a village

elder or just asking directions from a passer-by. Perhaps more importantly, their

street-smarts and family networks taught me how to locate key informants in a coun-

try that was still finding its footing after fourteen years of war.

In addition to my group of friends, I developed a wider network while conducting

surveys for NGOs, covering the 2005 elections as a freelance journalist, and managing

research projects for my advisors. I also made a documentary film, Sliding Liberia.

These assignments allowed me to travel widely, visiting places and speaking with

people that I would not have encountered within the narrow confines of the research

question. At one point I spent two weeks traveling on foot through the most remote

corners of Lofa County to implement a series of behavioral experiments. Far from the

nearest road, these villages provided insight into how civilians survived during the

266

267

war, as well as the complicated logistics required to transport supplies, ammunition,

and soldiers to the front lines.

The first phase of research for this project began in mid-2007 and focused on

understanding the conflict economy of Liberia. The goal was to collect detailed in-

formation on how civilians and rebels maintained access to food during the war, how

rebels taxed and/or looted from civilians, and how this behavior varied across the

country. The unit of analysis was the village or town, and I interviewed elders, mar-

ket sellers, farmers, and local combatants to understand the local histories in each

location. There is a risk of bias in this type of research, as peoples’ exposure to

violence—or their involvement in perpetrating violence—shapes their memories and

their portrayal of the situation. A market woman who was looted and raped by a

soldier, for example, would understandably have a negative view of that particular

rebel group, even if rape was relatively rare at the time.

To guard against these biases, I spoke to four types of informants in each village.

First, I spoke with village elders. In Liberian society, elders are responsible for settling

personal disputes in the community, and they tend to have a broad knowledge of

how different events affected local families. Also, these elders were often the liaison

between the rebel group and the local population, so they can provide information

about taxation during the war. Second, I interviewed people who remained in the

community when the village was controlled by each rebel group, and thus have direct

experience with the situation during that time. Third, I interviewed people who fled

when the town was controlled by each rebel group and asked them why they fled, even

as others remained. Fourth, I sought out market sellers and farmers who engaged in

productive activities (or not) during each rebel regime.

I conducted this research in locations that were either randomly selected or sig-

nificant during the war. The random sample was comprised of ten villages in north-

western Liberia, stratified between Lofa, Grand Cape Mount, and Bomi Counties.

This region was, at various points, controlled by the NPFL, ULIMO, and LURD

rebels. Selecting villages in this area strengthened my ability to identify differences

among the rebels’ behavior while minimizing the substantial time, effort, and expense

268 APPENDIX A. FIELDWORK IN LIBERIA

required to reach such remote locations. To complement this random sample, I gath-

ered information in 20 locations of significance during the war. Some of the locations

were easily chosen, such as the locations of rebel headquarters, while other, more ob-

scure locations, were identified as the project progressed. For example, I visited the

village of Hansen’s Farm, with a population of less than 100 people, after residents

of Gbarnga informed me that the village fell inside the “buffer zone” created by the

ECOMOG peacekeepers in late-2003. Despite its small size, interviews in Hansen’s

Farm revealed valuable information on the business activities of LURD commanders

after the ceasefire. My research in four locations—Buchanan, Greenville, Harper, and

Pleebo—occurred while I was affiliated with other projects, which prevented me from

conducting formal interviews. Nevertheless, I spoke with locals informally in these

places to understand the situation during the war. Figure A.1 depicts the locations

where fieldwork was conducted; Table A.1 lists the name of each location.

The second phase of the project involved locating and interviewing rebel leaders,

top commanders, and other important actors in Liberia’s war. My network in Liberia,

especially a local journalist who will remain anonymous, helped me to locate an initial

group of former commanders from each rebel group who provided interviews. These

commanders then helped me locate other members of their organization. Working

through these social networks creates a risk of bias, especially among the more faction-

prone groups. The ULIMO rebels, for example, were divided between the ethnic

Krahn and ethnic Mandingo forces. If I only spoke with Krahn members of ULIMO,

I would have heard only side side of the events leading to the group’s split. To

overcome this bias, I asked every informant to recreate the organizational chart of

the rebel group. Through this repeated process, I was able to identify the identities

of the top leadership and command structure throughout the war, as well as the

ethnicity, location, and fate of each of these high-level members. I then set out to

interview all of the survivors identified on the charts.

In addition to the rebel leaders and commanders, I interviewed several actors who

were in certain locations at key times or played an economic role in the conflict.

For example, informant C05 was a low-ranking fighter in ULIMO but had first-hand

experience with the diamond trade in ULIMO territory. Informant C40 was a local

269

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

ii

i

ii

!

!

i

i

i

!!

i

i!

!

25

15

21

25

27

28

29

31

23

6

12

10

1

3

4

8

7

9

1113 1416 17 18

19 2022

24

26

30

Note: Asterisks indicate randomly-chosen locations.

Figure A.1: Map of field sites in Liberia.


Map Number Name County Random1 Mendekoma Lofa Yes2 Foya Lofa No3 Kolahun Lofa No4 Sammie-Ta Lofa Yes5 Voinjama Lofa No6 Kpotomai Lofa Yes7 Gpalakpalzu Lofa Yes8 Zuwulor Lofa Yes9 Zorzor Lofa No

10 Zolowo Lofa No11 Bo Waterside Grand Cape Mount No12 Lofa Bridge Grand Cape Mount No13 Bubalor Grand Cape Mount Yes14 Bola Bomi Yes15 Tubmanburg Bomi No16 Robertsport Grand Cape Mount No17 Madina Grand Cape Mount Yes18 Vorkor Bomi Yes19 Klay Bomi No20 Folibli Bomi Yes21 Monrovia Montserrado No22 15 Gate Montserrado No23 Weala Margibi No24 Hansen’s Farm Bong No25 Gbargna Bong No26 Ganta Nimba No27 Buchanan Grand Bassa No28 Zwedru Grand Gedeh No29 Greenville Sinoe No30 Pleebo Maryland No31 Harper Maryland No

Table A.1: List of field sites in Liberia.

271

staff member of MSF who negotiated with NPFL to maintain humanitarian access

to the group’s territory. Informant C45 was a French national who owned a logging

company that operated during NPFL rule but was forced to evacuate after ULIMO’s

arrival. Many of these interviews were conducted in Monrovia, but quite a few in-

volved tracking informants to distant towns or small farms. Ultimately, I conducted

formal interviews with 75% of all surviving leaders and top commanders, and had

informal conversations with quite a few of the 25% remaining, although they declined

to be interviewed. Of those I was not able to interview, nearly all provided testimony

to the Truth and Reconciliation Commission or the Special Court of Sierra Leone.

My interviews began with an initial introductory meeting, which was a key element

of the process. Since most of my informants were referred to me by their former

colleagues, I was often introduced to them in person. When this was not possible, I

reached them by mobile phones. I then informed them of who I was, the nature of

the project, and how I found them. I made it clear that I would not ask them any

questions regarding violence that they or members of their group committed. Rather,

I was interested in the day-to-day challenges faced by the informant during the war,

and how the informant overcame those challenges. This disclaimer was significant for

two reasons. First, the Truth and Reconciliation Commission was active in Liberia

during much of the research project, and all of my informants were concerned about

indictments. Second, shifting the focus to daily challenges allows the informants to

tell their story, without getting distracted by the larger narrative or politics of the

conflict. Many of the informants appreciated the opportunity to share their history

without being judged through the lens of civilian abuse. Finally, I told the informants

that I could not pay them for an interview, but was happy to purchase food and drinks

during the interview, which often lasted up to four hours.

This first meeting hopefully resulted in a time for a second meeting, at which point

the interview was conducted. The interview locations were selected by the informants,

and most occurred in local bars or restaurants, near the informants’ homes, during

times when no other patrons were present. The interviews were semi-structured and

traced the informants’ personal histories, including where they were born, their ed-

ucation and military training, and how they initially became involved in the war. I


asked them about their ranks in the rebel organization, how they were promoted, and

how they supplied food to their troops. The interviews also covered the informants’

relationships with the rebel leader and other commanders, as well as key events in the

groups’ histories, such as factionalization and military setbacks. Importantly, I asked

each informant to provide background information on every other high-level member

in the group, as part of the process of creating the organizational chart. By asking this

information of every informant, I repeatedly verified and cross-checked the informa-

tion provided by other informants. After each interview I compiled this information

into a database that was cross-referenced for each person mentioned and each key

event, allowing me to quickly search for inconsistencies in the information. If I found

significant discrepancies, I would telephone the informant to ask for clarification. In

some cases, a second (or third or fourth) interview was scheduled.

The success of these interviews depended to a large extent on my personal compo-

sure. It was vital that I made clear to the informants within the first few minutes of

the introduction that I was not going to “judge” them for what they did during the

war. Rather, my goal was to understand their goals and the challenges they faced.

The key to communicating this message was my sincerity; I believed and meant what

I said. My background in Liberia helped build relationships with the informants in

three ways. First, I spoke with the informants in Liberian English, but made sure

to avoid certain colloquial expressions that could be considered condescending. Sec-

ond, I traveled alone through the country on an inexpensive motorbike. This mode

of transport signaled my familiarity with the country and made me less threatening

than the aid workers and diplomats who travel in white SUVs with drivers. Third, I

ate and drank with the informants in local bars and restaurants in their home com-

munities. Although it was occasionally difficult to keep up with my informants when

it came to drinking beer at 9am, I gave it my best shot.

Interviewing civilians and former combatants in a post-war setting can be chal-

lenging and at times frustrating. For example, the publication of the Truth and

Reconciliation Commission report in July 2009 caused several informants to change

their mind and decline to be interviewed, even though I had been in repeated contact

with them for months. Despite these setbacks, by selecting informants from a variety

273

of backgrounds, aided by comprehensive organizational charts, minimizes the risk of

selection bias. Building a relationship with the informant and conducting interviews

in private spaces chosen by the informant created a comfort level that allowed even

difficult topics to be explored with some degree of candor. Finally, all information

was cross-checked and verified. Through these methods, I believe I have created an

accurate, representative picture of Liberia’s rebel organizations and the plight of the

civilians who suffered through fourteen years of war.

Appendix B

Interviews

B.1 Civilian Security, Local Production, Rebel Gov-

ernance

ID Description Faction(s) Location Date

A01 Refugees MODEL Peace Town, Cote d’Ivoire 6 Aug 2003

A02 Civilians LURD Voinjama-Macenta Road 15 Oct 2005

B01 Shop owner - Voinjama, Lofa County 11 July 2007

B02 Soldiers (5) LURD Voinjama, Lofa County 11 July 2007

B03 Female trader LURD Voinjama, Lofa County 12 July 2007

B04 Civilian LURD Voinjama, Lofa County 12 July 2007


B06 Female trader NPFL Voinjama, Lofa County 12 July 2007


B08 Female trader NPFL, ULIMO Voinjama, Lofa County 13 July 2007

B09 Female trader NPFL Voinjama, Lofa County 14 July 2007

B10 Commander LURD Voinjama, Lofa County 14 July 2007

B11 Civilian NPFL, ULIMO Kpotomai, Lofa County 15 July 2007

B12 Elder NPFL, ULIMO Kpotomai, Lofa County 15 July 2007

B13 Civilian NPFL, ULIMO, LURD Sammie-Ta, Lofa County 15 July 2007

B14 Elder NPFL, ULIMO, LURD Sammie-Ta, Lofa County 15 July 2007

B15 Female trader NPFL, ULIMO, LURD Sammie-Ta, Lofa County 15 July 2007

B16 Female trader NPFL, ULIMO Kolahun, Lofa County 16 July 2007

B17 Female trader NPFL, ULIMO, LURD Kolahun, Lofa County 16 July 2007

274

B.1. CIVILIAN SECURITY, LOCAL PRODUCTION, REBEL GOVERNANCE275

B18 Elder NPFL, ULIMO Kolahun, Lofa County 16 July 2007

B19 Male trader NPFL, ULIMO, LURD Foya, Lofa County 16 July 2007

B20 Civilians NPFL, ULIMO, LURD Mendekoma, Lofa Cty 16 July 2007

B21 Elders NPFL, ULIMO, LURD Gpalakpalzu, Lofa Cty 17 July 2007

B22 Elders NPFL, ULIMO, LURD Zuwulor, Lofa County 17 July 2007

B23 Female trader NPFL, ULIMO Zorzor, Lofa County 18 July 2007

B24 Female trader NPFL Zorzor, Lofa County 20 July 2007

B25 Male trader NPFL, ULIMO, LURD Zolowo, Lofa County 21 July 2007

B26 Civilian NPFL, ULIMO, LURD Madina, Cape Mt Cty 27 July 2007

B27 Civilian LURD Robertsport, Cape Mt 27 July 2007

B28 Civilian NPFL, ULIMO, LURD Robertsport, Cape Mt 27 July 2007

B29 Civilian NPFL, ULIMO, LURD Gbargna, Bong County 6 Aug 2007

B30 Male trader NPFL Gbarnga, Bong County 6 Aug 2007

B31 Female trader NPFL Gbarnga, Bong County 7 Aug 2007

B32 Female shop owner LURD Gbarnga, Bong County 7 Aug 2007

B33 Male shop owner LURD Gbarnga, Bong County 7 Aug 2007

B34 Civilian NPFL, ULIMO Gbarnga, Bong County 8 Aug 2007

B35 Male shop owner NPFL, ULIMO Gbarnga, Bong County 8 Aug 2007

B36 Civilian LURD Hansen’s Farm, Bong Cty 9 Aug 2007

B37 Elder NPFL, ULIMO, LURD Hansen’s Farm, Bong Cty 9 Aug 2007

B38 Civilian NPFL, ULIMO, LURD Hansen’s Farm, Bong Cty 9 Aug 2007

B39 Elder LURD Hansen’s Farm, Bong Cty 9 Aug 2007

B40 Civilian NPFL, ULIMO Weala, Margibi County 9 Aug 2007

B41 Male trader NPFL Gbarnga, Bong County 15 Aug 2007

B42 Gov’t official NPFL, ULIMO, LURD Robertsport, Cape Mt 20 Aug 2007

B43 Soldier NPFL Monrovia 21 Aug 2007

B44 Soldier LURD Tubmanburg, Bomi Cty 22 Aug 2007

B45 Civilians NPFL, ULIMO, LURD Tubmanburg, Bomi Cty 22 Aug 2007

B46 Elders NPFL, ULIMO, LURD Bola, Bomi County 23 Aug 2007

B47 Civilians LURD Lofa Bridge, Cape Mt 23 Aug 2007

B48 Civilian NPFL, ULIMO, LURD Lofa Bridge, Cape Mt 23 Aug 2007

B49 Civilian LURD Lofa Bridge, Cape Mt 23 Aug 2007

B50 Female trader NPFL, ULIMO Tubmanburg, Bomi Cty 24 Aug 2007

B51 Elders NPFL, ULIMO, LURD Vorkor, Bomi County 24 Aug 2007

B52 Elders NPFL, ULIMO, LURD Klay, Bomi County 24 Aug 2007

B53 Elders NPFL, ULIMO, LURD Follibli, Bomi Cty 24 Aug 2007

B54 Elders NPFL, ULIMO, LURD Bo Waterside, Cape Mt 26 Aug 2007

276 APPENDIX B. INTERVIEWS

B55 Civilian NPFL, ULIMO, LURD Bubalor, Cape Mt 26 Aug 2007

B56 Farmers NPFL, ULIMO, LURD Gbei Town, Bomi Cty 11 Mar 2010

B.2 Rebel Organization, External Patrons, Com-

mander Incentives


C01 Soldier, Commander ULIMO, LURD Monrovia 30 Apr 2009

C02 Senior Commander ULIMO/J, LURD Monrovia 4 May 2009

- - - (via telephone) 8 May 2009

- - - - 20 Jun 2009

- - - - 13 Aug 2009

C03 Sr. Political Official ULIMO/J, LURD Monrovia 7 May 2009

C04 Sr. Political Official ULIMO Monrovia 8 May 2009

C05 Soldier ULIMO/J Monrovia 8 May 2009

C06 Training Commander ULIMO/J Monrovia 11 May 2009

C07 Sr. Political Official ULIMO, LPC, LURD Monrovia 12 May 2009

C08 Sr. Political Official NPFL, NPFL-CRC Monrovia 13 May 2009

C09 Commander ULIMO/J, LURD Monrovia 14 May 2009

C10 Commander LURD Monrovia 18 May 2009


- - - - 19 May 2009

C12 Commander ULIMO, LURD Monrovia 18 May 2009

- - - - 19 May 2009

C13 Senior Commander ULIMO Monrovia 18 May 2009

- - - - 20 May 2009

C14 Female Commander LURD Monrovia 25 May 2009

C15 Senior Commander NPFL Monrovia 27 May 2009


C17 Commander INPFL, NPFL Monrovia 30 May 2009

C18 Commander NPFL, ULIMO/K, LURD Monrovia 1 Jun 2009

C19 Commander INPFL, NPFL Monrovia 1 Jun 2009

C20 Commander MODEL Monrovia 5 Jun 2009

C21 Soldier ATU Monrovia 6 Jun 2009

C22 Commander ULIMO/J, MODEL Zwedru 10 Jun 2009

C23 Commander MODEL Zwedru 10 Jun 2009

B.3. MISCELLANEOUS INTERVIEWS 277

C24 Senior Commander NPFL, AFL Monrovia 15 Jun 2009

C25 Commander MODEL Monrovia 17 Jun 2009

C26 Senior Commander NPFL 15 Gate 18 Jun 2009

C27 Commander NPFL 15 Gate 18 Jun 2009

C28 Commander ULIMO/K, LURD Monrovia 19 Jun 2009

C29 Sr. Political Official MODEL Monrovia 30 Jun 2009

C30 Commander LURD Monrovia 2 Jul 2009

C31 Commander MODEL Monrovia 3 Jul 2009

C32 Sr. Political Official MODEL Monrovia 4 Jul 2009

C33 Senior Commander MODEL Monrovia 4 Jul 2009

C34 Commander ULIMO/J, LURD Monrovia 8 Jul 2009

C35 Commander LPC, MODEL Monrovia 8 Jul 2009

C36 Commander MODEL Monrovia 10 Jul 2009

C37 Commander ULIMO/K Monrovia 14 Jul 2009

C38 Commander NPFL Monrovia 14 Jul 2009

C39 Sr. Political Official NPFL Monrovia 14 Jul 2009

C40 MSF local staff NPFL Monrovia 3 Aug 2009

C41 Commander ULIMO/J Monrovia 4 Aug 2009

C42 Commander LPC Monrovia 13 Aug 2009

- - - (via telephone) 14 Aug 2009

C43 Commander ULIMO/K Philadelphia (tel) 1 Sep 2009

C44 Commander ULIMO/K Philadelphia (tel) 23 Sep 2009

C45 Owner of Logging Firm NPFL, ULIMO Monrovia 28 Jan 2010

C46 Sr. Political Official LURD Monrovia 12 Feb 2010

C47 Sr. Political Official LURD Monrovia (tel) 11 Mar 2010

C48 Senior Commander ULIMO/J, MODEL Accra (tel) 14 Mar 2010

B.3 Miscellaneous Interviews


I25 Civil Society Leader FAPC Ariwara, Ituri, DR Congo 2 Feb 2009

I26 Gold Trader FAPC Ariwara, Ituri, DR Congo 2 Feb 2009

I27 Elders FAPC Kumudu, Ituri, DR Congo 2 Feb 2009

N07 Commander CNDP Bunagana, North Kivu, DR Congo 17 Nov 2007

N08 Political Official CNDP Bunagana, North Kivu, DR Congo 17 Nov 2007

S12 Soldier LTTE Kilinochchi, Sri Lanka 23 Jul 2004

S18 Soldier LTTE Jaffna, Sri Lanka 27 Jul 2004

278 APPENDIX B. INTERVIEWS

S35 Director, Orphanage for LTTE Batticaloa, Sri Lanka 26 Aug 2004

former child soldiers

B.4 TRC Testimonies


TRC0A7 Alhaji G.V. Kromah ULIMO/K Monrovia 11 Aug 2008

TRC0D1 Prince Y. Johnson NPFL, INPFL Monrovia 26 Aug 2008

TRC0FF Moses Z. Blah NPFL Monrovia 15 Sep 2008

TRC0FB Roland Duo NPFL Monrovia 11 Dec 2008

TRC159 Edward T. Zaymay NPFL Monrovia 4 Feb 2009

TRC160 George E.S. Boley LPC Monrovia 5 Feb 2009

TRC163 George E.S. Boley LPC Monrovia 5 Feb 2009

Bibliography

[1] Adelman, Howard. 2003. “The Use and Abuse of Refugees in Zaire.” In StephenJohn Stedman and Fred Tanner, eds. Refugee Manipulation: War, Politics, andthe Abuse of Human Suffering (Washington, D.C.: Brookings Institution Press).

[2] Akerlof, George A. and Rachel E. Kranton. 2005. “Identity and the Economicsof Organization.” Journal of Economic Perspectives 19, No. 1 (Winter), pp.9-32.

[3] Alao, Abiodun, John Mackinlay, and ’Funmi Olonisakin. 1999. Peacekeepers,Politicians, and Warlords: The Liberian Peace Process (Tokyo: United NationsUniversity Press).

[4] AMNH (American Museum of Natural History). 1997. “The Nature of Dia-monds.” Avaiable http://www.amnh.org/exhibitions/diamonds/. Accessed13 May 2010.

[5] Anderson, Mary B. 1999. Do No Harm: How Aid Can Support Peace—Or War(Boulder: Lynne Rienner).

[6] Anselin, Luc. 2002. “Under the hood: Issues in the specification and interpre-tation of spatial regression models.” Agricultural Economics 27, pp. 247-267.

[7] Arop Madut-Arop. 2006. Sudan’s Painful Road to Peace (USA: Booksurge).

[8] Azam, Jean-Paul. 2002. “Looting and Conflict between Ethno-Regional Groups:Lessons for State Formation in Africa.” Journal of Conflict Resolution 46, No.1 (February), pp. 131-153.

[9] Azam, Jean-Paul. 2006. “On thugs and heroes: Why warlords victimize theirown civilians.” Economics of Governance 7 (2006), pp. 53-73.

[10] Ballentine, Karen and Heiko Nitzschke. 2005. “Introduction.” In Karen Ballen-tine and Heiko Nitzschke, eds. Profiting from Peace: Managing the ResourceDimensions of Civil War (Boulder: Lynne Rienner).

279

280 BIBLIOGRAPHY

[11] Bapat, Navin A. 2011. “Understanding State Sponsorship of Militant Groups.”British Journal of Political Science (Forthcoming).

[12] Bates, Robert H. 1983. “Modernization, Ethnic Competition, and the Ratio-nality of Politics in Contemporary Africa.” In Donald Rothchild and Victor A.Olunsorola, eds. State versus Ethnic Claims: African Policy Dilemmas (Boul-der: Westview Press).

[13] BBC. 2001. “Warlord on trial in Croatia” (20 July).

[14] BBC. 2002. “Colombia’s rebel kidnappers” (7 January).

[15] BBC. 2003. “Palestinian Authority funds go to militants” (7 November).

[16] BBC. 2010. “Ethiopian famine aid ‘spent on weapons”’ (3 March).

[17] Bendor, J., A. Glazer, and T. Hammond. 2001. “Theories of Delegation.” An-nual Review of Political Science 4, pp. 235-69.

[18] Berman, Eli. 2009. Radical, Religious, and Violent: The New Economics ofTerrorism (Cambridge: MIT Press).

[19] Bicchieri, Cristina. 1990. “Norms of Cooperation.” Ethics 100, No. 4 (July),pp. 838-861.

[20] Bivand, Roger S., Edzer J. Pebesma, and Virgilio Gomez-Rubio. 2008. AppliedSpatial Data Analysis with R (New York: Springer).

[21] Blattman, Christopher and Jeannie Annan. 2010. “On the nature and causes ofLRA abduction: What the abductees say.” In Tim Allen and Koen Vlassenroot,eds. The Lord’s Resistance Army: Myth and Reality (London: Zed Books).

[22] Bob, Clifford. 2005. The Marketing of Rebellion: Insurgents, Media, and Inter-national Activism (New York: Cambridge University Press).

[23] Brabazon, James. 2003. “Liberia: Liberians United for Reconciliation andDemocracy (LURD).” Briefing Paper No. 1 (February 2003). Armed Non StateActors Project, Royal Institute of International Affairs.

[24] Bracamonte, Jose Angel Moroni and David E. Spencer. 1995. Strategy andTactics of the Salvadoran FMLN Guerrillas: Last Battle of the Cold War,Blueprint for Future Conflicts (Westport: Praeger Publishers).

[25] Browning, Christopher R. 1992. Ordinary Men: Reserve Battalion 101 and theFinal Solution in Poland (New York: HarperCollins).

BIBLIOGRAPHY 281

[26] Byman, Daniel, Peter Chalk, Bruce Hoffman, William Rosenau, and DavidBrannan. 2001. Trends in Outside Support for Insurgent Movements (SantaMonica: RAND).

[27] CAL (Center for Army Leadership). 2004. U.S. Army Leadership Manual (NewYork: McGraw-Hill).

[28] Capie, David. 2004. “Armed groups, weapons availability and misuse: Anoverview of the issues and options for action.” Centre for humanitarian dia-logue, Briefing Paper (25 May 2004).

[29] Carlton, Charles. 1998. “Civilians.” In John Kenyon and Jane Ohlmeyer, eds.The Civil Wars: A Military History of England, Scotland, and Ireland, 1638-1660 (New York: Oxford University Press).

[30] Chouvy, Pierre-Arnaud. 2010. Opium: Uncovering the Politics of the Poppy(Cambridge: Harvard University Press).

[31] CI (Conservation International). 2006. “Forest Cover Mapping and Change De-tection using Moderate-Resolution Satellite Imagery (Landsat, ASTER andMODIS).” Center for Applied Biodiversity Science.

[32] CIA (Central Intelligence Agency). 1985. “Ethiopia: Political and Security Im-pact of the Drought.” Intelligence Assessment No. 2432. Released by Freedomof Information Act, August 2002.

[33] Cohen, Dara K. 2010. Explaining Sexual Violence During Civil War. Ph.D.dissertation. Department of Political Science, Stanford University.

[34] Collier, David and James Mahoney. 1996. “Insights and Pitfalls: Selection Biasin Qualitative Research.” World Politics 49, No. 1, pp. 56-91.

[35] Collier, Paul and Anke Hoeffler. 2004. “Greed and grievance in civil war.” OxfordEconomic Papers 56, pp. 563-595.

[36] Coulter, Chris. 2009. Bush Wives and Girl Soldiers: Women’s Lives throughWar and Peace in Sierra Leone (Ithaca: Cornell University Press).

[37] Cunningham, David E. 2006. “Veto Players and Civil War Duration.” AmericanJournal of Political Science 50, No. 4 (October), pp. 875-892.

[38] Cunningham, David E., Kristian Skrede Gleditsch, and Idean Salehyan. 2009.“It Takes Two: A Dyadic Analysis of Civil War Duration and Outcome.” Jour-nal of Conflict Resolution 53, No. 4 (August 2009), pp. 570-597.

282 BIBLIOGRAPHY

[39] Davis, Anthony. 2003. “Philippine security threatened by small arms prolifera-tion.” Jane’s Intelligence Review (August 2003).

[40] Demetriou, Spyros. 2002. “Politics from the Barrel of a Gun: Small Arms Pro-liferation and Conflict in the Republic of Georgia (1989-2001).” Small ArmsSurvey Occasional Paper, No. 6.

[41] Driscoll, Jesse Russell. 2009. Exiting Anarchy: Militia Politics After the Post-Soviet Wars. PhD Dissertation, Department of Political Science, Stanford Uni-versity.

[42] Dunn, D. Elwood and Svend E. Holsoe. 1985. Historical Dictionary of Liberia(Metuchen: The Scarecrow Press).

[43] Dyer, Gwynne. 2005. War: The Lethal Custom (New York: Carroll & Graf).

[44] Economist. 2009. “Flying anything to anybody.” (18 December 2008).

[45] Ellingsen, Tore and Magnus Johannesson. 2008. “Pride and Prejudice: TheHuman Side of Incentive Theory.” American Economic Review 98, No. 3 (June),pp. 990-1008.

[46] Ellis, Stephen. 1999. The Mask of Anarchy: The Destruction of Liberia and theReligious Dimensions of an African Civil War (New York: New York UniversityPress).

[47] Elster, Jon. 1989. “Social Norms and Economic Theory.” The Journal of Eco-nomic Perspectives 3, No. 4 (Autumn), pp. 99-117.

[48] FAS (Foreign Area Studies). Liberia: A Country Study (Washington, D.C.:The American University).

[49] Fearon, James D. 1995. “Rationalist Explanations for War.” International Or-ganization 49, No. 3 (Summer), pp. 379-414.

[50] Fearon, James D. 1999. “Why Ethnic Politics and ’Pork’ Tend to Go Together.”Mimeo.

[51] Fearon, James D. 2008a. “Economic Development, Insurgency, and Civil War.”In Elhanan Helpman, ed. Institutions and Economic Performance (Cambridge:Harvard University Press).

[52] Fearon, James D. 2008b. “The Rise of Emergency Relief Aid.” In Michael Bar-nett and Thomas C. Weiss, eds. Humanitarianism in Question: Politics, Power,Ethics (Cambridge: Cambridge University Press).

BIBLIOGRAPHY 283

[53] Fearon, James D. and David D. Laitin. 1996. ”Explaining Interethnic Coop-eration.” The American Political Science Review 90, No. 4 (December), pp.715-735.

[54] Fearon, James D. and David D. Laitin. 2003. ”Ethnicity, Insurgency, and CivilWar.” American Political Science Review 97, No. 1 (February 2003), pp. 75-90.

[55] Fearon, James D. and David Laitin. 2007. “Civil War Termination.” Paperpresented at the Annual Meeting of the American Political Science Association,Chicago, IL.

[56] Foody, Giles M. 2002. “Status of land cover classification accuracy assessment.”Remote Sensing of the Environment 80, pp. 185-201.

[57] Frey, Bruno S. and Reto Jegen. 2001. “Motivation Crowding Theory.” Journalof Economic Surveys 15, No. 5 (December), pp. 589-611.

[58] Gates, Scott. 2002. “Recruitment and Allegiance: The Microfoundations ofRebellion.” Journal of Conflict Resolution 46, No. 1 (February), pp. 111-130.

[59] Ghobarah, Hazem Adam, Paul Huth, and Bruce Russett. 2003. “Civil Wars Killand Maim People—Long After the Shooting Stops.” American Political ScienceReview 97, No. 2 (May), pp. 189-202.

[60] Giles, Lionel. 1910. “Introduction.” In Sun Tzu, The Art of War (Project Guten-berg E-Text Edition).

[61] Gilmore, Elisabeth, Nils Peter Gleditsch, and Jan Ketil Rod. 2005. “ConflictDiamonds: A New Dataset.” Conflict Management and Peace Science 22, pp.257-272.

[62] Gleditsch, Nils Petter; Peter Wallensteen, Mikael Eriksson, Margareta Sollen-berg, and Havard Strand. 2002. “Armed Conflict 19462001: A New Dataset.”Journal of Peace Research 39, No. 5, pp. 615637.

[63] Gneezy, Uri and Aldo Rustichini. 2000. “Pay Enough or Don’t Pay At All.”Quarterly Journal of Economics 115, No. 3 (August), pp. 791-810.

[64] Greene, Graham. 1936. Journey Without Maps (USA: Doubleday).

[65] Guevara, Che. [1961] 1998. Guerilla Warfare (Lincoln: University of NebraskaPress).

284 BIBLIOGRAPHY

[66] Habyarimana, James, Macartan Humphreys, Daniel N. Posner, and Jeremy M.Weinstein. 2009. Coethnicity: Diversity and the Dilemmas of Collective Action(New York: Russel Sage Foundation).

[67] Haining, Robert. 2003. Spatial Data Analysis: Theory and Practice (Cam-bridge: Cambridge University Press).

[68] Hechter, Michael. 1987. Principles of Group Solidarity (Los Angeles: Universityof California Press).

[69] Hess, George R. and Jeff M. Bay. 1997. “Generating confidence intervals forcomposition-based landscape indexes.” Landscape Ecology 12, pp. 309-320.

[70] Ho, Daniel E., Kosuke Imai, Gary King, and Elizabeth A. Stuart. 2007. “Match-ing as Nonparametric Preprocessing for Reducing Model Dependence in Para-metric Causal Inference.” Political Analysis 15, No. 3 (Summer), pp. 199-236.

[71] Hovil, Lucy and Eric Werker. 2005. “Portrait of a Failed Rebellion: An Accountof Rational, Sub-Optimal Violence in Western Uganda.” Rationality and Society17, No. 1 (2005), pp. 5-34.

[72] HRW (Human Rights Watch). 1998. “War Without Quarter: Colombia andInternational Humanitarian Law” (October).

[73] HRW (Human Rights Watch). 2001. “Crisis of Impunity: The Role of Pakistan,Russia, and Iran in Fueling the Civil War in Afghanistan” Vol. 13, No. 3C (July1).

[74] HRW (Human Rights Watch). 2003a. “Trapped Between Two Wars: ViolenceAgainst Civilians in Western Cote d’Ivoire” Vol. 15, No. 14A (August).

[75] HRW (Human Rights Watch). 2003b. “Weapons Sanctions, Military Supplies,and Human Suffering: Illegal Arms Flows to Liberia and the June-July 2003Shelling of Monrovia.” Briefing Paper (3 November 2003).

[76] HRW (Human Rights Watch). 2005. The Curse of Gold (1 June).

[77] HRW (Human Rights Watch). 2008. “Trapped and Mistreated: LTTE AbusesAgainst Civilians in the Vanni” (December).

[78] Huband, Mark. 1998. The Liberian Civil War (Portland: Frank Cass Publish-ers).

BIBLIOGRAPHY 285

[79] Humphreys, Macartan and Jeremy M. Weinstein. 2006. ”Handling and Man-handling Civilians in Civil War.” American Political Science Review 100, No.3 (August), pp. 429-447.

[80] Iannacone, Laurence R. 1992. “Sacrifice and Stigma: Reducing Free-riding inCults, Communes, and Other Collectives.” The Journal of Political Economy100, No. 2 (April), pp. 271-91.

[81] ICG (International Crisis Group). 1998. “Congo at War: A Briefing on theInternal and External Players in the Central African Conflict.” ICG CongoReport No. 2 (17 November).

[82] ICG (International Crisis Group). 2002a. “Liberia: The Key to Ending RegionalInstability.” Africa Report No. 43 (24 April).

[83] ICG (International Crisis Group). 2002b. “Liberia: Unravelling.” Africa Brief-ing Paper (19 August).

[84] ICG (International Crisis Group). 2007. “Sri Lanka’s Muslims: Caught in theCrossfire.” Asia Report No. 134 (29 May).

[85] ICG (International Crisis Group). 2009. “Liberia: Uneven Progress in SecuritySector Reform.” Africa Report No. 148 (13 January).

[86] ICG (International Crisis Group). 2010. “Dangerous Little Stones: Diamondsin the Central African Republic.” Africa Report No. 167 (16 December).

[87] IRIN (U.N. Integrated Regional Information Network). 2003a. “Liberia: LURDrebels promise to return stolen relief vehicles” (22 August).

[88] IRIN (U.N. Integrated Regional Information Network). 2003b. “Liberia: Reliefworkers return to rebel-held port of Harper” (16 September).

[89] IRIN (U.N. Integrated Regional Information Network). 2004a. “Liberia: Bryantto intervene in rebel rift” (21 January).

[90] IRIN (U.N. Integrated Regional Information Network). 2004b. “Liberia: Envi-ronmentalists Report Illegal Logging” (12 February).

[91] IRIN (U.N. Integrated Regional Information Network). 2004c. “Liberia: UNpeacekeepers arrive, but rebel gunmen stay on” (27 February).

[92] IRIN (U.N. Integrated Regional Information Network). 2004d. “Bitter LURDfighters say their leaders betrayed them” (19 April).

286 BIBLIOGRAPHY

[93] Jones, Gareth R. 1986. “Socialization Tactics, Self-Efficacy, and Newcomers’Adjustments to Organizations.” The Academy of Management Journal 29, No.2 (June), pp. 262-269.

[94] Kalyvas, Stathis N. 2001. “‘New’ and ‘Old’ Civil Wars: A Valid Distinction?”World Politics 54, No. 1, pp. 99-118.

[95] Kalyvas, Stathis N. 2006. The Logic of Violence in Civil War (Cambridge:Cambridge University Press).

[96] Kaplan, Oliver. 2010. Civilian Autonomy in Civil Wars. PhD Dissertation, De-partment of Political Science, Stanford University.

[97] Kaplan, Robert. 1994. ”The Coming Anarchy.” The Atlantic Monthly (Febru-ary).

[98] Kohn, George Childs, Ed. 2007. Dictionary of Wars, Third Edition (New York:Checkmark Books).

[99] Konneh, Augustine. 1996. Religion, Commerce, and the Integration of theMandingo in Liberia (Lanham: University Press of America).

[100] Kreps, David M. 1990. “Corporate culture and economic theory.” In James E.Alt and Kenneth A. Shepsle, eds. Perspectives on Positive Political Economy(New York: Cambridge University Press).

[101] Lacina, Bethany and Nils Petter Gleditsch. 2005. “Monitoring Trends in GlobalCombat: A New Dataset of Battle Deaths.” European Journal of Population21, No. 2-3 (June), pp. 145-166.

[102] Laqueur, Walter. 2004. Guerrilla Warfare: A Historical and Critical Study (NewBrunswick: Transaction Publishers).

[103] Le Billon, Philippe. 2001. “The political ecology of war: natural resources andarmed conflicts.” Political Geography 20 (2001), pp. 561-584.

[104] Lee, Gary Yia. 2005. “The Hmong Rebellion in Laos.” Unpublishedmanuscript. Available http://members.ozemail.com.au/~yeulee/Topical/

Hmong%20rebellion%20in%20Laos.html. Accessed 7 February 2011.

[105] Levitt, Jeremy I. 2005. The Evolution of Deadly Conflict in Liberia (Durham:Carolina Academic Press).

BIBLIOGRAPHY 287

[106] Lin, D.Y. and L.J. Wei. 1989. “The Robust Inference for the Cox ProportionalHazards Model.” Journal of the American Statistical Association 84, No. 408(December), pp. 1074-1078.

[107] Lischer, Sarah Kenyon. 2003. “Collateral Damage: Humanitarian Assistance asa Cause of Conflict.” International Security 28, No. 1 (Summer), pp. 79-109.

[108] MacCoun, Robert J., Elizabeth Kier, and Aaron Belkin. 2006. “Does SocialCohesion Determine Motivation in Combat?: An Old Question with an OldAnswer.” Armed Forces & Society 32, No. 4 (July), pp. 646-654.

[109] Mao Tse-Tung. 1937. “Basic Tactics.” Available: http://www.marxists.

org/reference/archive/mao/selected-works/volume-6/mswv6_28.htm.Accessed 21 August 2010.

[110] MAR (Minorities at Risk Project). 2004. “Chronology of Afars in Dji-bouti.” Available http://www.unhcr.org/refworld/docid/469f3882c.html.Accessed 19 June 2011.

[111] Marshall, Monty G. and Ted Robert Gurr. 2005. Peace and Conflict 2005 (Col-lege Park: Center for International Development & Conflict Management).

[112] Mas-Colell, Andreu, Michael D. Whinston, and Jerry R. Green. 1995. Microe-conomic Theory (Oxford: Oxford University Press).

[113] Maugham, R.C.F. [1920] 1969. The Republic of Liberia (New York: NegroUniversities Press).

[114] McHugh, Gerard and Manuel Bessler. 2006. Guidelines on Humanitarian Nego-tiations with Armed Groups (New York: United Nations).

[115] McPherson, J.H.T. [1891] 2009. History of Liberia (USA: Feather Trail Press).

[116] Merlin. 2003. “SouthEast Liberia Trip Report”. Unpublished field report byMagnus Wolfe Murray, country manager of Merlin Liberia (17 October).

[117] Miller, Gary J. 1992. Managerial Dilemmas: The Political Economy of Hierar-chy (New York: Cambridge University Press).

[118] NPR (National Public Radio). 2008. “Army Documents Show Lower Recruit-ing Standards.” Morning Edition (17 April). Available http://www.npr.org/

templates/story/story.php?storyId=89702118. Accessed 26 July 2010.

[119] Ogunleye, Bayo. 1995. Behind Rebel Line: Anatomy of Charles Taylor’s HostageCamps (Enugu: Delta Publications).

288 BIBLIOGRAPHY

[120] Olson, Mancur. 1993. ”Dictatorship, Democracy, and Development.” AmericanPolitical Science Review 87, No. 3 (September), pp. 567-576.

[121] Pape, Robert A. 2003. “The Strategic Logic of Suicide Terrorism.” AmericanPolitical Science Review 97, No. 3 (August), pp. 343-361.

[122] Pham, John-Peter. 2004. Liberia: Portrait of a Failed State (New York: ReedPress).

[123] Raleigh, Cionadh and Havard Hegre. 2005. “Introducing ACLED: An ArmedConflict Location and Event Dataset.” Paper presented to the conference on“Disaggregating the Study of Civil War and Transnational Violence,” Universityof California Institute of Global Conflict and Cooperation, San Diego, CA, 7-8March.

[124] Reno, William. 1999. Warlord Politics and African States (Boulder: LynneRienner).

[125] Rice, Edward E. 1990. Wars of the Third Kind: Conflict in UnderdevelopedSocieties (Berkeley: University of California Press).

[126] Salehyan, Idean. 2008. “No Shelter Here: Rebel Sanctuaries and InternationalConflict.” The Journal of Politics 70, No. 1 (January), pp. 54-66.

[127] Salehyan, Idean. 2010. “The Delegation of War to Rebel Organizations.” Jour-nal of Conflict Resolution 54, No. 3 (June), pp. 493-515.

[128] Sarkees, Meredith Reid. 2000. “The Correlates of War Data on War: An Updateto 1997.” Conflict Management and Peace Science 18, No. 1 (February), pp.123-144.

[129] SAS (Small Arms Survey). 2006. Targeting Ammunition: A Primer (Geneva:Small Arms Survey).

[130] SAS (Small Arms Survey). 2011. “Crime and Victimization in Liberia: ReadingBetween the Lines.” Liberia Armed Violence Assessment, Issue Brief No. 2.

[131] Sawyer, Amos. 1992. The Emergence of Autocracy in Liberia (San Francisco:ICS Press).

[132] Schraeder, Peter J. 1993. “Ethnic Politics in Djibouti: From ’Eye of the Hurri-cane’ to ’Boiling Cauldron’.” African Affairs 92, No. 367 (April) pp. 203-221.

[133] Scott, Colin. 1995. “Humanitarian Action and Security in Liberia, 1989-1994.”Occasional Paper No. 20 (Providence: Institute for International Studies).

BIBLIOGRAPHY 289

[134] Sekhon, Jasjeet S. 2007. “Multivariate and Propensity Score Matching Softwarewith Automated Balance Optimization: The Matching package for R.” Journalof Statistical Software 42, No. 7 (June), pp. 1-52.

[135] Shapiro, Jacob N. 2007. The Terrorist’s Challenge: Security, Efficiency, Control.PhD dissertation, Department of Political Science, Stanford University.

[136] Shick, Tom W. 1971. “A Quantitative Analysis of Liberian Colonization from1820 to 1843 with Special Reference to Mortality” Journal of African History12, No. 1, pp. 45-59.

[137] Snyder, Richard. 2001. “Scaling Down: The Subnational ComparativeMethod.” Studies in Comparative International Development 36, No. 1 (Spring),pp. 93-110.

[138] Solomon, Christiana. 2006. “The role of women in economic transformation:market women in Sierra Leone.” Conflict, Security & Development 6, No. 3(October), pp. 411-423.

[139] South, Ashley. 2008. Ethnic Politics in Burma (New York: Routledge).

[140] Spence, Michael. 1973. “Job Market Signaling.” The Quarterly Journal of Eco-nomics 18, No. 3 (August), pp. 355-374.

[141] Staniland, Paul S. 2010. Explaining Cohesion, Fragmentation, and Control inInsurgent Groups. PhD Dissertation, Department of Political Science, Mas-sachusetts Institute of Technology.

[142] Stanton, Jessica. 2009. Strategies of Violence and Restraint in Civil War. PhDDissertation, Department of Political Science, Columbia University.

[143] Stedman, Stephen John. 2002. “Policy Implications.” In Stephen John Stedman,Donald Rothchild, and Elizabeth M. Cousens, eds. Ending Civil Wars: TheImplementation of Peace Agreements (Boulder: Lynne Rienner).

[144] Stehman, Stephen V. and Raymond L. Czaplewski. 1998. “Design and Analysisfor Thematic Map Accuracy Assessment: Fundamental Principles.” RemoteSensing of the Environment 64, pp. 331-344.

[145] Stouffer, Samuel A., Arthur A. Lumsdaine, Marion H. Lumsdaine, Robin M.Williams, Jr., M. Brewster Smith, Irving L. Janis, Shirley A. Star, and LeonardS. Cottrell, Jr. 1949. The American Soldier: Combat and Its Aftermath (Prince-ton: Princeton University Press).

290 BIBLIOGRAPHY

[146] Therneau, Terry M. and Patricia M. Grambsch. 2000. Modeling Survival Data:Extending the Cox Model (New York: Springer-Verlag).

[147] Times of London. 1850. “Liberia” (10 October).

[148] Times of London. 1892. “Liberia.” (n.d.)

[149] Times of London. 1931. “The Liberian Report: Slavery Charge Proved” (11January).

[150] TRC (Truth and Reconciliation Commission, Republic of Liberia). 2009. “FinalReport” (Monrovia: TRC).

[151] Turner, Thomas. 2002. Angola’s Role in the Congo War. In John F. Clark, ed.The African Stakes of the Congo War (New York: Palgrave Macmillan).

[152] Turner, Thomas. 2007. Congo Wars: Conflict, Myth, Reality (New York: ZedBooks).

[153] UNHCHR (UN High Commission for Human Rights). 2010. “Report of theMapping Exercise documenting the most serious violations of human rights andinternational humanitarian law committed within the territory of the Demo-cratic Republic of the Congo between March 1993 and June 2003” (August).

[154] UNSC (United Nations Security Council). 1994a. “Sixth Progress Report ofthe Secretary-General on the United Nations Observer Mission in Liberia”(S/1994/1006).

[155] UNSC (United Nations Security Council). 1994b. “Seventh Progress Reportof the Secretary-General on the United Nations Observer Mission in Liberia”(S/1994/1167).

[156] UNSC (United Nations Security Council). 1995a. “Eighth Progress Report ofthe Secretary-General on the United Nations Observer Mission in Liberia”(S/1995/9).

[157] UNSC (United Nations Security Council). 1995b. “Eleventh Progress Reportof the Secretary-General on the United Nations Observer Mission in Liberia”(S/1995/473).

[158] UNSC (United Nations Security Council). 1996a. “Fifteenth Progress Reportof the Secretary-General on the United Nations Observer Mission in Liberia”(S/1996/47).

BIBLIOGRAPHY 291

[159] UNSC (United Nations Security Council). 1996b. “Seventeenth Progress Reportof the Secretary-General on the United Nations Observer Mission in Liberia”(S/1996/362).

[160] UNSC (United Nations Security Council. 1997. “Twenty-Third Progress Reportof the Secretary-General on the United Nations Observer Mission in Liberia”(S/1997/478).

[161] UNSC (United Nations Security Council). 2000a. “Final Report of the Moni-toring Mechanism on Angola Sanctions.” (S/2000/1225).

[162] UNSC (United Nations Security Council). 2000b. “Report of the Panel of Ex-perts appointed pursuant to Security Council resolution 1306 (2000), paragraph19, in relation to Sierra Leone.” (S/2000/1195).

[163] UNSC (United Nations Security Council. 2001. “Report of the Panel of Expertspursuant to Security Council resolution 1343 (2001), paragraph 19, concerningLiberia” (S/2001/1015).

[164] UNSC (United Nations Security Council). 2003a. “Report of the Panel of Ex-perts appointed pursuant to Security Council resolution 1458 (2003), concerningLiberia” (S/2003/498).

[165] UNSC (United Nations Security Council). 2003b. “Report of the Panel ofExperts on Somalia pursuant to Security Council resolution 1474 (2003)”(S/2003/1035).

[166] UNSC (United Nations Security Council). 2005a. “Report of the Group of Ex-perts on the Democratic Republic of Congo.” (S/2005/30).

[167] UNSC (United Nations Security Council. 2005b. “Sixth progress report of theSecretary-General on the United Nations Mission in Liberia” (S/2005/177).

[168] UNSC (United Nations Security Council). 2006. “Letter dated 26 January2006 from the Chairman of the Security Council Committee established pur-suant to resolution 1533 (2004) concerning the Democratic Republic of Congo.”(S/2006/53).

[169] UNSC (United Nations Security Council). 2008. “Final report of the Group ofExperts on the Democratic Republic of Congo.” (S/2008/773).

[170] UNSC (United Nations Security Council). 2010. “Report of the Monitor-ing Group on Somalia pursuant to Security Council resolution 1853 (2008).”(S/2010/91).

292 BIBLIOGRAPHY

[171] UNODS (United Nations Office on Drugs and Crime). 2007. “AfghanistanOpium Survey 2007” (September).

[172] UNODS (United Nations Office on Drugs and Crime). 2009. “Colombia CocaCultivation Survey” (June).

[173] Utas, Mats. 2005. “Victimcy, Girlfriending, Soldiering: Tactic Agency in aYoung Woman’s Social Navigation of the Liberian War Zone.” Anthropologi-cal Quarterly 78, No. 2, pp. 403-430.

[174] USDS (United States Department of State). 1996. “Testimony by William H.Twaddell, Assistant Secretary of State for African Affairs, Hearing on LiberiaBefore the House International Relations Committee” (26 June).

[175] Valentino, Benjamin A. 2004. Final Solutions: Mass Killing and Genocide inthe 20th Century (Ithaca: Cornell University Press).

[176] Valentino, Benjamin, Paul Huth, and Dylan Balch-Lindsay. 2004. “‘Drainingthe Sea’: Mass Killing and Guerrilla Warfare.” International Organization 58(Spring), pp. 375-407.

[177] Van Maanen, John and Edgar H. Schein. 1977. “Toward a Theory of Organiza-tional Socialization.” Mimeo (Cambridge: Massachusetts Institute of Technol-ogy).

[178] Vlassenroot, Koen and Timothy Raeymaekers. 2004. “The Politics of Rebellionand Intervention in Ituri: The Emergence of a New Political Complex?” AfricanAffairs 103, No. 412, pp. 385-412.

[179] Vo Nguyen Giap. 1970. The Military Art of People’s War (New York: MonthlyReview Press).

[180] Ward, Michael D. and Kristian Skrede Gleditsch. 2008. Spatial Regression Mod-els (Los Angeles: Sage).

[181] Washington Post. 2011. “Vang Pao, anti-communist guerrilla fighter and leaderof Hmong refugees, dies at 81” (8 January).

[182] Weber, Max. 1947. The Theory of Social and Economic Organization, Trans-lated by Talcott Parsons (New York: The Free Press).

[183] Weinstein, Jeremy M. 2005. “Resources and the Information Problem in RebelRecruitment.” Journal of Conflict Resolution 49, No. 4 (August), pp. 598-624.

BIBLIOGRAPHY 293

[184] Weinstein, Jeremy M. 2007. Inside Rebellion: The Politics of Insurgent Violence(Cambridge: Cambridge University Press).

[185] Williamson, Oliver E. 1983. ”Organization Form, Residual Claimants, and Cor-porate Control.” Journal of Law and Economics 26, No. 2 (June), pp. 351-366.

[186] Wilson, Clay. 2007. “Improvised Explosive Devices (IEDs) in Iraq andAfghanistan: Effects and Countermeasures.” Congressional Research Service,Library of Congress (RS22330).

[187] Wong, Leonard, Thomas A. Kolditz, Raymond A. Millen, Terrence M. Potter.2003. “Why They Fight: Combat Motivation in the Iraq War.” Strategic StudiesInstitute, U.S. Army War College (June).

[188] Wood, Elisabeth Jean. 2006. “Variation in Sexual Violence during War.” Poli-tics & Society 34, No. 3 (September), pp. 307-341.

[189] Wooldridge, Jeffrey M. 2010. Econometric Analysis of Cross Section and PanelData, 2nd ed. (Cambridge: The MIT Press).

VIOLENT ORDER: REBEL ORGANIZATION AND LIBERIA’S CIVIL …cd347ss0802/Lidow_DissFinal... ·...

Documents

Transcript of VIOLENT ORDER: REBEL ORGANIZATION AND LIBERIA’S CIVIL …cd347ss0802/Lidow_DissFinal... ·...