Inequality, Grievances, and Civil War

This book argues that political and economic inequalities following group
lines generate grievances that can motivate civil war. The theoretical approach
highlights ethnonationalism and how linkages between group identities and
inequalities spur mobilization and resort to violence. Although contemporary
research on civil war has largely dismissed grievances as irrelevant, empha-
sizing instead the role of opportunities, the authors show that many alleged
nonresults for grievances stem from atheoretical measures, typically based on
individualist data. The authors develop new indicators of political and eco-
nomic exclusion at the group level and demonstrate how these exert strong
effects on the risk of civil war. They provide new analyses of the effects of
transnational ethnic links and the duration of civil wars and extended case
discussions illustrating causal mechanisms.

Lars-Erik Cederman is Professor of International Conflict Research at the

Center for Comparative and International Studies at ETH Zürich. He is the
author of Emergent Actors in World Politics (1997), the editor of Construct-
ing Europe’s Identity (2001), and the co-editor of New Systems Theories of
World Politics (2010). He is the winner of the 2012 and 2002 American
Political Science Association’s Heinz I. Eulau Awards, the 2000 Furniss Book
Award, and the Horace H. Rackham Dissertation Award from the University
of Michigan.

Kristian Skrede Gleditsch is Professor in the Department of Government,

University of Essex, and a Research Associate at the Peace Research Institute
Oslo (PRIO). He is the author of All International Politics Is Local (2002)
and Spatial Regression Models (2008) as well as numerous journal articles.
He is the winner of the 2012 American Political Science Association’s Heinz
I. Eulau Award, the 2011 International Association for Conflict Management
outstanding article in the field award, the 2007 Karl Deutsch Award from the
International Studies Association, and the 2000 American Political Science
Association’s Helen Dwight Reid Award.

Halvard Buhaug is Research Professor at the Peace Research Institute Oslo

and Professor of Political Science at the Norwegian University of Science
and Technology. He received the 2006 Royal Norwegian Society of Sci-
ences and Letters award for excellent research by young scholars in human
sciences. Recent academic publications include articles in the Journal of
Conflict Resolution, Global Environmental Change, International Security,
International Organization, and Proceedings of the National Academy of
Sciences of the USA.
Cambridge Studies in Contentious Politics

Editors
Mark Beissinger Princeton University
Jack A. Goldstone George Mason University
Michael Hanagan Vassar College
Doug McAdam Stanford University and Center for Advanced Study in the
Behavioral Sciences
Sarah A. Soule Stanford University
Suzanne Staggenborg University of Pittsburgh
Sidney Tarrow Cornell University
Charles Tilly (d. 2008) Columbia University
Elisabeth J. Wood Yale University
Deborah Yashar Princeton University

Ronald Aminzade et al., Silence and Voice in the Study of Contentious

Politics
Javier Auyero, Routine Politics and Violence in Argentina: The Gray
Zone of State Power
W. Lance Bennett and Alexandra Segerberg, The Logic of Connective
Action: Digital Media and the Personalization of Contentious
Politics
Clifford Bob, The Marketing of Rebellion: Insurgents, Media, and
International Activism
Charles Brockett, Political Movements and Violence in Central America
Valerie Bunce and Sharon Wolchik, Defeating Authoritarian Leaders in
Postcommunist Countries
Lars-Erik Cederman, Kristian Skrede Gleditsch, and Halvard Buhaug,
Inequality, Grievances, and Civil War
Christian Davenport, Media Bias, Perspective, and State Repression
Gerald F. Davis, Doug McAdam, W. Richard Scott, and Mayer N. Zald,
Social Movements and Organization Theory
Donatella della Porta, Clandestine Political Violence
Todd A. Eisenstadt, Politics, Identity, and Mexico’s Indigenous Rights
Movements
Daniel Q. Gillion, The Political Power of Protest: Minority Activism
and Shifts in Public Policy
Jack A. Goldstone, editor, States, Parties, and Social Movements
Tamara Kay, NAFTA and the Politics of Labor Transnationalism

Continued after the Index

Inequality, Grievances, and Civil War

LARS-ERIK CEDERMAN
Center for Comparative and International Studies,
ETH Zürich

KRISTIAN SKREDE GLEDITSCH

University of Essex and Peace Research Institute Oslo

HALVARD BUHAUG
Peace Research Institute Oslo
32 Avenue of the Americas, New York, ny 10013-2473, usa

Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of education,
learning and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9781107603042
© Lars-Erik Cederman, Kristian Skrede Gleditsch, and Halvard Buhaug 2013
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2013

Printed in the United States of America

A catalog record for this publication is available from the British Library.
Library of Congress Cataloging in Publication Data
Cederman, Lars-Erik, 1963–
Inequality, grievances, and civil war / Lars-Erik Cederman, Center for Comparative and
International Studies, ETH Zürich; Kristian Skrede Gleditsch, University of Essex and Peace
Research Institute Oslo; Halvard Buhaug, Peace Research Institute Oslo.
pages cm. – (Cambridge studies in contentious politics)
Includes bibliographical references and index.
isbn 978-1-107-01742-9 (hardback) – isbn 978-1-107-60304-2 (pbk.)
1. Civil war – Social aspects. 2. Equality. 3. Human rights. I. Gleditsch, Kristian Skrede,
1971– II. Buhaug, Halvard, 1972– III. Title.
jc491.c374 2013
303.6ʹ4–dc23 2013001656
isbn 978-1-107-01742-9 Hardback
isbn 978-1-107-60304-2 Paperback
Cambridge University Press has no responsibility for the persistence or accuracy of urls for
external or third-party Internet Web sites referred to in this publication and does not guarantee
that any content on such Web sites is, or will remain, accurate or appropriate.
Contents

Tables page viii

Figures ix
Preface xi

1. Introduction 1

part i: theories and concepts

2. Inequality and Grievances in the Civil War Literature 11
3. From Horizontal Inequality to Civil War via Grievances 30

part ii: empirical analysis of civil war onset

4. Political Exclusion and Civil War 57
5. Economic Inequality and Civil War 93
6. Transborder Ethnic Kin and Civil War 119
7. Country-Level Inequalities and Civil War 143

part iii: beyond civil war onset

8. The Duration and Outcome of Civil Wars 171
9. Conclusions for Theory and Policy 205

References 227
Index 251

vii
Tables

4.1. Comparing excluded and included groups’ conflict

propensities page 71
4.2. Ethnic groups’ conflict propensity by power-access
category 72
4.3. Political horizontal inequality and group-level onset of civil
war, 1946–2009 73
5.1. Economic horizontal inequality and group-level onset of
civil war, 1991–2009 108
6.1. TEK and group-level onset of civil war, 1946–2009 132
7.1. Mean sample values by conflict outcome, 1991–2009 154
7.2. Logit model of civil war onset, 1991–2009 155
7.3. Multinomial logit models of civil war onset, 1991–2009 157
7.4. High horizontal inequality in 1999 and civil war onset,
2000–9 161
7.5. High horizontal inequality in 2009 without conflict,
1991–2009 164
7.6. Summary of findings 165
8.1. Approaches to civil war dynamics/duration 175
8.2. Cox proportional hazard regression estimates of
termination, 1946–2009 194
8.3. Multinomial logit by outcome type (to conflict
continuation), 1946–2009 197
8.4. Cox proportional hazard regression of conflict
termination, conflict level, 1946–2009 203
9.1. A summary of the confirmed hypotheses of this book 206
9.2. Top-ten countries with largest excluded and discriminated
shares of the population in 2009 221

viii
Figures

3.1. The causal path from inequalities to onset through

grievances page 36
3.2. The causal path from horizontal inequalities to grievances 37
3.3. The causal path from grievances to civil war onset 44
3.4. Grievance-based mechanism as spurious explanation 51
3.5. Grievances bypassed through reverse causation 52
4.1. The polity model with included and excluded groups 59
4.2. Comparing the effect of power access status on civil war
onset 75
4.3. Comparing the effect on civil war onset for EPR categories 76
4.4. The effect of status reversals on the probability of civil war
onset 77
4.5. Conflict propensity of included and excluded groups as a
function of their relative size 78
4.6. Conflict propensity as a function of the number of the
group’s previous conflict onsets 79
5.1. G-Econ cells for Yugoslavia overlaid with GeoEPR group
polygons for Slovenes and Albanians 103
5.2. Result of spatial wealth estimation for groups in
Yugoslavia 104
5.3. Result of spatial wealth estimation for groups in the Sudan 105
5.4. Result of spatial wealth estimation for groups in Myanmar 106
5.5. Conflict probability for groups less wealthy than the
country average 109
5.6. Conflict probability for groups wealthier than the country
average 110
5.7. Comparing the effect of relative poverty on the conflict
probability of included and excluded ethnic groups 111
ix
x Figures

5.8. First-order difference between excluded and included

groups 112
6.1. The ethnonationalist triad with an excluded TEK group 124
6.2. The ethnonationalist triad with an included TEK group 124
6.3. The probability of conflict as a function of relative TEK
size 133
6.4. Conflict probability varying with size of included and
excluded TEK groups 134
6.5. Contour plot showing the effect on conflict probability of
relative group and TEK sizes 136
7.1. Vertical versus horizontal political inequality in 2000 152
7.2. Vertical versus horizontal economic inequality in 2000 153
7.3. Estimated risk of ethnic governmental conflict as a
function of extent of political discrimination (black line)
with 90% confidence interval (gray area) 159
7.4. Estimated risk of ethnic territorial conflict as a function of
extent of economic inequality (black line) with 90%
confidence interval (gray area) 160
7.5. ROC plot of out-of-sample prediction, 2000–9 162
7.6. Scatter plot of individual out-of-sample predictions,
2000–9 163
8.1. Sequential fighting in civil wars 183
8.2. Outcome probabilities by time, organizations with link to
an excluded group 199
8.3. Outcome probabilities by time, organizations with link to
an excluded group and support from a transnational
constituency included in another state 200
8.4. Outcome probabilities by time, organizations with no link
to an excluded ethnic group 201
9.1. Average democracy levels by world region 214
9.2. Ethnically excluded population by world region 216
9.3. Ethnically discriminated population by world region 217
9.4. Share of excluded population by country in 2009 219
9.5. Share of discriminated population by country in 2009 220
Preface

Apart from acquiring real estate, there are few things that make you more
indebted than writing collaborative monographs. Indeed, this book is the prod-
uct not only of our own efforts, but also those of our colleagues and students
who made major contributions in terms of conceptual development, data col-
lection, software development, and analysis. In this very respect, the book falls
somewhere between a co-authored book and a co-edited volume.
More than anything else, this book is the fruit of several years of intense
research collaboration involving a tight network of conflict researchers in
Europe. These activities started as an informal network that we labeled Geo-
graphic Research On War Network (GROWnet), which originally included
colleagues at the Peace Research Institute Oslo (PRIO) and the Norwegian Uni-
versity of Science and Technology in Norway, ETH Zürich and University of
Geneva in Switzerland, and the University of Essex in the United Kingdom, and
was subsequently extended to collaborators at Uppsala University in Sweden.
Within this context, we have published a large amount of co-authored research,
with an author list featuring every possible combination of the three authors of
this book and beyond. Some of the research paving the way for the book was
funded by an ECRP grant on “Disaggregating Civil War” (06ECRPFP004)
awarded by the European Science Foundation, 2007–10, through individual
grants from the national research councils, that is, the Research Council of
Norway (182399), Swiss National Science Foundation (105511–116795), and
UK Economic and Social Research Council (RES-062–23-0259). Gleditsch also
acknowledges support from the Research Council of Norway (180441/V10).
As of the summer of 2012, the network received new funding from the Euro-
pean Union as a COST Action (IS 1102), which will allow it to expand its
activities to many more partners in Europe under the heading of the European
Network of Conflict Research (ENCoRe).

xi
xii Preface

Sidney Tarrow played an instrumental role at a pivotal moment by sug-

gesting that we publish our book project with Cambridge University Press,
where Lewis Bateman has supported our work with both patience and pro-
fessionalism. All in all, we held three book workshops at ETH Zürich on
10 November 2011, at Yale on 24 February 2012, and at PRIO on 10 March
2012. We received excellent feedback from the participants at those workshops,
including Keith Darden, Hanne Fjelde, Scott Gates, Jack Goldstone, Håvard
Hegre, Helge Holtermann, Simon Hug, Stathis Kalyvas, Janet Lewis, Nicholas
Sambanis, Frances Stewart, Andres Vargas, Nils Weidmann, and Tore Wig.
We also benefited from comments from participants at seminar presentations
at Koç University on 16 April 2012 and the London School of Economics and
Political Science on 9 May 2012, including Belgin San Akça, Ali Çarkoğlu,
Emre Hatipoğlu, Bill Kissane, Omar McDoom, Michael Mousseau, and
Jameson Lee Ungerer. In addition, Carles Boix, Michael Hechter, Erik
Melander, and Camber Warren commented on parts of the manuscript.
Current and past PhD students and researchers of the International Con-
flict Research Groups at ETH Zürich have selflessly sacrificed large chunks
of their valuable research time to provide public goods in the shape of data
projects and analysis. Nils Weidmann and Christa Deiwiks volunteered their
skills in terms of data management and computational and spatial analysis.
Julian Wucherpfennig and Philipp Hunziker played an instrumental role in
project management and econometric analysis. Luc Girardin, with the help of
Sebastian Schutte and Philipp Hunziker, created the computational infrastruc-
ture that made the project possible, including the data portal GROWup . Manuel
Vogt, Nils-Christian Bormann, and Seraina Rüegger managed the coding and
updating of large data structures, all related to the Ethnic Power Relations
(EPR) data set. Indeed, this book project revolves around this data resource,
which originates from research collaboration involving Andreas Wimmer and
Brian Min, both then at UCLA. We are grateful to a long list of experts
who offered their advice regarding, the coding of countless country cases;
see Chapter 4, footnote 7, and http://www.icr.ethz.ch/data. In particular, we
would like to acknowledge the expert input by James Scarritt and Manuel Vogt,
who helped us update the coding of Sub-Saharan Africa. Our data collection
efforts were generously supported by the National Center of Competence in
Research, “Challenges to democracy in the 21st century,” and funded by the
Swiss National Science Foundation (SNSF) and the Swiss Network for Inter-
national Studies (SNIS).
At the University of Essex, we have benefited from comments and sug-
gestions from current and former colleagues Tobias Böhmelt, Han Dorussen,
Ismene Gizelis, Steve Pickering, and Ulrich Pilster. Nils Metternich and Andrea
Ruggeri served as research officers on our ESF project and have become val-
ued collaborators who have influenced the research reported in this book in
numerous ways.
Preface xiii

Research underpinning this book has also benefited greatly from brown-
bag seminars and informal discussions with colleagues at PRIO under the
auspices of the Centre for the Study of Civil War (CSCW). Being associated
with the GROWnet, Scott Gates, Nils Petter Gleditsch, Johan Dittrich Hall-
berg, Håvard Hegre, Helge Holtermann, and Gudrun Østby have followed the
development of this project and generously offered insightful and constructive
comments and suggestions along the way. Likewise, we have benefited tremen-
dously from input on GIS, data, and statistical methods from Jan Ketil Rød,
Håvard Strand, and Andreas Forø Tollefsen. External CSCW affiliates such
as Sabine Carey, David Cunningham, Kathleen Cunningham, Magnus Öberg,
Idean Salehyan, and Gerald Schneider have also provided valuable comments.
Finally, we would like to thank Andrew John Feltham for his excellent help
managing our activities through CSCW.
It should be stressed that a lot of crucial steps were taken in connection
with collaborative publications that preceded this book. Andreas Wimmer
and Brian Min co-authored an article that inspired Chapter 4 (see Cederman,
Wimmer, and Min 2010). Nils Weidmann co-authored the article that consti-
tutes the initial basis of Chapter 5 (see Cederman, Weidmann, and Gleditsch
2011). Chapter 6 draws heavily on an article co-authored with Idean Sale-
hyan and Julian Wucherpfennig (see Cederman, Gleditsch, Salehyan, and
Wucherpfennig 2013), Chapter 7 builds on some of our other joint work
(see Buhaug, Cederman, and Gleditsch Forthcoming), and Chapter 8 relies
on material from articles that we co-authored with David Cunningham and
Idean Salehyan (Cunningham, Gleditsch, and Salehyan 2009) as well as Julian
Wucherpfennig and Nils Metternich (Wucherpfennig, Metternich, Cederman,
and Gleditsch 2012). We are also grateful for the publishers’ permission to use
material from these publications.
Last but not least, we would like to acknowledge the continuous support
from our partners and families who provided both intellectual inspiration and
a healthy counterweight to the “workaholic” tendencies of academic life. With-
out their boundless patience and generous willingness to bear the burden of
often-absent spouses and fathers, this book could not have been written.
1

Introduction

Do grievances cause civil war? The desperate struggles of discriminated and

stateless peoples around the world suggest that the answer to this question
must be affirmative, as illustrated by such cases as the Palestinians in the West
Bank and Gaza, the Fur in the Sudan, the Tamils in Sri Lanka, the Karen in
Myanmar, and the Kurds in Turkey and elsewhere. Indeed, the upheavals in
North Africa and the Middle East in 2011 demonstrate that it is difficult to
sustain regimes that exclude large parts of the population from political power
along ethnic or nonethnic lines.
In stark contrast to these observations, however, much of the contempo-
rary literature on civil war takes a very different view. Regarding explanations
rooted in political and economic grievances with suspicion, leading scholars
of civil war typically give short shrift to grievance-based accounts on the basis
of results indicating that ethnic diversity and unequal individual wealth distri-
butions have no statistically distinguishable relationship to internal conflict. In
particular, ethnic grievances as triggers of civil wars receive little support in
this literature, despite their having attracted substantial attention in qualitative
studies. Arguing that grievances are the product, rather than a cause, of vio-
lence, or otherwise so omnipresent that they cannot account for civil conflict,
these authors question the sincerity of political entrepreneurs’ appeals to ethnic
nationalism, dismissing them as the opportunistic and self-serving arguments
of warlords, thugs, and criminals.
In his best-selling book The Bottom Billion, Collier (2007, p. 18) expresses
these doubts explicitly:
So what causes civil war? Rebel movements themselves justify their actions in terms
of a catalogue of grievances: repression, exploitation, exclusion. Politically motivated
academics have piled in with their own hobbyhorses, which usually cast rebels as heroes.
I have come to distrust this discourse of grievances as self-serving.

1
2 Inequality, Grievances, and Civil War

Along similar lines, Mueller (2000, p. 92) interprets grievances as an oppor-

tunistic cover for greedy and even criminal activities:
What passed for “ethnic warfare” in Bosnia and Croatia seems then to have been some-
thing far more banal: the creation of communities of criminal violence and predation.
In the end, the wars resembled the movie images of the American Wild West or of
gangland Chicago, and they often had less to do with nationalism than with criminal
opportunism and sadistic cruelty, very commonly enhanced with liquor.

If ethnic conflict is fundamentally banal then it could potentially take place in

any society, as illustrated by British soccer hooligans and motorcycle gangs in
Denmark: “Under the right conditions, thugs can rise to a dominant role, others
can lend a hand or withdraw into terrified isolation or studied indifference, and
any place can degenerate into a Bosnia or a Rwanda” (Mueller 2000, p. 68).
From a less radical vantage point, most scholars who rely on quantitative
evidence insist that ethnic frustrations fail to explain internal conflict. In a
review of political-economy approaches to “ethnic mobilization and ethnic
violence,” Fearon (2006, pp. 857–8) comes to precisely this conclusion:
Cross-national statistical studies find surprisingly few differences between the determi-
nants of civil war onset in general, versus “ethnic” civil wars in particular. Once one
controls for per capita income, neither civil wars nor ethnic civil wars are significantly
more frequent in more ethnically diverse countries; nor are they more likely when there
is an ethnic majority and a large ethnic minority.

In fact, this skeptical attitude as regards grievances as causes of civil war

and ethnic conflict is so pronounced among rationalist scholars that Horowitz
(2002, p. 547) characterizes it as “antipathy to antipathy.”
Challenging these nonfindings, we argue that they, to a large extent, result
from inappropriate theoretical assumptions and problematic empirical opera-
tionalizations. To be sure, measuring grievances is more easily said than done.
As mental states that can easily be misinterpreted, such phenomena are notori-
ously difficult to pin down objectively. We wholeheartedly agree with Blattman
and Miguel’s (2010, p. 18) diagnosis:
At present, the economic motivations for conflict are better theorized than psychological
or sociological factors. Individual preferences in existing models typically include only
material rewards and punishments. One key implication is that we have not derived the
falsifiable predictions that distinguish between material and non-material theoretical
accounts. Yet, the greater degree of existing theory on economic factors does not imply
that researchers should discard non-economic explanations of conflict.

Much of the contemporary research on civil war aggregates the analysis to the
country level. While expressing similar doubts about the relevance of grievances
as explanations, a new wave of scholarship has shifted the attention away from
entire countries to micro-level studies of specific processes of civil violence. In
seminal contributions to the civil war literature, Kalyvas (2003; 2006) casts
doubt on the validity of “master cleavages” as the key to such conflicts. Instead,
Introduction 3

he argues in favor of a much more disaggregated perspective that highlights

mundane micro-level motivations, such as local feuds and the settling of petty
grudges that have little to do with the warring parties’ broader ideological
justifications. Although Kalyvas remains open to the possibility of ideological
preferences influencing the outbreak of conflict, and some micro-level studies
leave room for the role of political grievances (e.g., Gates 2002; Wood 2003;
Weinstein 2007), most authors contributing to this literature deviate from such
explanations (e.g., Fearon 2006; Blattman and Miguel 2010).
Our empirical strategy differs from both the country-level and micro-level
literatures by focusing on interactions between the state and actors at inter-
mediate levels of aggregation, such as ethnic groups and rebel organizations
(Cederman and Gleditsch 2009). This approach allows us to go well beyond
conventional country-level studies in terms of empirical detail, while at the
same time allowing us to maintain the broad horizons of a truly global com-
parison. Following in the footsteps of Gurr’s (1993a; 2000b) and Horowitz’s
(1985) classical contributions to the literature on ethnic conflict, this is a book
about the impact of political grievances on civil war in general and the conflict-
fueling role of ethnic nationalism in particular. As illustrated by the revolts in
Tunisia, Egypt, and Libya erupting in the spring of 2011, anger with political
exclusion without reference to ethnic cleavages may lead to violence, but eth-
nonationalist civil wars remain arguably the most important, and also most
misunderstood, class of grievance-related violence.
As a way to overcome the formidable obstacles associated with the opera-
tionalization and measurement of grievances, our strategy is to take one step
back in order to detect structural situations that can be safely assumed to
cause frustrations in the first place. In particular, we postulate that political
and economic inequalities afflicting entire ethnic groups, rather than merely
individuals, are especially likely to fuel resentment and justify attempts to
fight perceived injustice. According to the terminology introduced by Stew-
art (2008b), such asymmetries can be labeled “horizontal inequalities” since
they concern established groups, as opposed to “vertical inequalities” among
isolated individuals and households. Part of the reason why grievance-based
arguments have found little support in the quantitative literature is that such
studies have typically measured vertical inequality, while ignoring its horizontal
counterpart.
Since the most powerful counterarguments to grievance-based explanations
have been backed up by quantitative indicators, this book centers on the devel-
opment of better measures of inequalities that are directly linked to group-level
theories of civil war. Furthermore, the problems of measuring political and
economic inequalities call for new data. There is an important research tra-
dition that relies on well-established data sources, such as the Minorities at
Risk (MAR) data set. Nevertheless, these and similar data impose restrictions
that make them less suitable to study the types of processes that we focus
on in this book. Therefore, we build on a new version of the Ethnic Power
4 Inequality, Grievances, and Civil War

Relations data set that we refer to as EPR-ETH, which traces ethnic groups’
access to state power (see Chapter 4). Our analysis also makes frequent use of
Geographic Information Systems (GIS), which facilitate measurement of sub-
national properties and configurations. In particular, we rely on a geo-coded
extension to EPR-ETH (GeoEPR), which provides detailed information about
the EPR groups’ settlement areas (see Chapter 5). Furthermore, the EPR-ETH
data include a new extension that covers transnational ethnic kin (see Chapter
6). Finally, the book also employs a mapping that codes links between the
EPR-ETH groups and conflict data coded for rebel organizations. All these
data resources can be accessed through our data portal GROWup (Geographic
Research On War: unified platform).1
If properly reconceptualized as group-level claims resulting from macro-
level processes, such as nationalism and state formation, rather than as fixed
ethno-demographic configurations or apolitical collections of individual char-
acteristics, grievances can be systematically linked to political violence through
actor-specific mechanisms. Taking this step from “factors to actors” enables us
to postulate and evaluate a number of specific hypotheses concerning conflict
parties’ behavior under varying ethno-political configurations. Our argument
does not, in any way, exclude possible effects of alternative causal mechanisms
that are not related to inequalities or grievances. Rather than setting up false
dichotomies that pit “greed” or “opportunities” against “grievances,” our aim
is to show that the latter category strongly influences the probability of civil
conflict even while controlling for the former. Thus, the goal is to resurrect and
refine a specific class of grievance-based explanations rather than debunking
the alternatives.
The second part of our book analyzes the causes of civil war outbreak. In
Chapter 4, we find that ethnic groups that are excluded from governmental
influence are more likely to experience conflict than those that enjoy secure
access to executive power. Recent loss of power or outright discrimination,
rather than mere exclusion, tends to increase the risk of conflict even further.
Our results indicate that political horizontal inequality often triggers civil
violence.
Moreover, the findings of Chapter 5 demonstrate that group-level economic
inequality can also lead to violent conflict, especially for groups with wealth lev-
els below the national average, as exemplified by the Chechens in Russia and the
Albanians in Yugoslavia. However, the evidence is much more mixed for groups
that are wealthier than the average group. Furthermore, our analysis shows that
the conflict propensity of disadvantaged and to some extent advantaged groups
appears to hinge on political exclusion in the sense that economic horizontal
inequality only matters where there is also political horizontal inequality.
In Chapter 6, we show that this group-level perspective also allows us to
capture transborder processes involving ethnonationalist kin. Despite recent
advances in this area, a central puzzle remains unresolved: namely that ethnic

1 The system is available at http://growup.ethz.ch.

Introduction 5

groups that at least in theory could count on support from large transborder
ethnic kin (TEK) groups have often remained surprisingly peaceful, such as
the stranded Russian populations in the “near abroad.” Postulating a curvilin-
ear, conflict-inducing effect of the TEK group’s relative size compared to the
incumbent, state-controlling group, we find that the risk of conflict increases
within the middle range of the size spectrum. Moreover, our results suggest
that the net effect, compared with situations without transnational links, is
conflict-dampening for large TEK groups that enjoy access to executive power
in their countries, as illustrated by the lack of conflict in many post-Soviet
states. In contrast, our model shows that excluded TEK groups, such as the
Kurdish minorities in Turkey, Iran, and Iraq before 2003, tend to increase the
risk of civil war.
The group-level analyses are complemented by findings aggregated up to
the country level in Chapter 7, including different measures of inequality and
conflict. The aggregated analysis makes it possible to compare ethnic to noneth-
nic conflicts and the risk of conflict with excluded groups to the risk of civil
war in countries without ethnic cleavages. Moreover, we contrast explanations
based on horizontal inequalities to those that feature vertical inequalities. The
results suggest that our group-level findings can be readily generalized from
group relations to the country level, while also demonstrating the advantages
of replacing conventional ethno-demographic indicators with measures that
are more sensitive to the underlying political logic of ethnonationalist conflict
and within-country variation.
Part III goes beyond the traditional focus on conflict onset. In Chapter
8, we examine the duration and outcome of conflicts, which allows us to
consider the actual characteristics of organizations involved in conflicts and
their relationship to ethnic groups, and to take seriously the possibility that
ethnic groups may not be unitary actors and that rebel organizations can
have a complex relationship to ethnic constituencies. We demonstrate that
ethnonationalist exclusion influences not only the initial emergence of conflict
but also its duration. Again, the argument is that decisions by states to exclude
groups from power, rather than the mere existence of ethnic cleavages, are
what make a difference for explaining patterns of violence.
Finally, Chapter 9 concludes the book. There we summarize our findings and
explore trends affecting entire world regions. This analysis allows us to draw
general conclusions for theory and policy, especially with respect to inclusion
of ethnic groups through power sharing.
Having summarized our findings, it may be useful to anticipate the reasons
why we come to such different conclusions compared with the dominant view.
In fact, there are several reasons why both grievances and inequalities have
been downplayed in contemporary scholarship on civil wars. To address these
shortcomings, this book offers the following improvements:

r Intermediate disaggregation rather than just individual or country-level anal-

ysis. Most systematic studies of civil wars tend to be general but heavily
6 Inequality, Grievances, and Civil War

aggregated, and thus lacking in empirical precision, or disaggregated but

focused on single countries, and thus lacking in generality. Focusing on con-
flict processes either at the country level or at the level of individuals or
villages, much of the contemporary literature overlooks intermediate levels
of aggregation where collective grievances are directly relevant, and where
inequalities are most easily detected. To fill this gap, we explore explicit
relational configurations at the mesolevel of aggregation, namely the group
and organization levels, before aggregating our findings back to the country
level.
r Motivational rather than merely cognitive mechanisms. Because of the strong
influence of narrowly construed rationalistic and cognitive theorizing, many
popular explanations tend to overlook the role of grievances. Rather than
privileging cognitive processes at the expense of emotions, our book postu-
lates explicit causal mechanisms that show how collectively felt grievances
result from structural inequalities and may produce violent conflict under
specific conditions.
r Ethnonationalism rather than merely ethnicity as a conflict cause. The con-
ventional literature on “ethnic conflict” tends to debate the role of ethnicity
in itself, as a demographic or individualist property that can be extricated
from its political context. Such reasoning loses sight of the state, which plays
a central role as a prize and an autonomous actor in nationalist conflicts.
Our theoretical framework centers on the political function of ethnicity,
especially where inequality creates tensions between ethnic groups that can
be exploited for ethnonationalist mobilization.
r Theoretically relevant data and measures, rather than the standard toolbox.
Scarcity and fragmentation of appropriate data on the relevant actors have
also made it difficult to measure grievances, even indirectly. Indices such
as the ethnic fractionalization index and the Gini coefficient of inequality
are inherently individualist and therefore offer a poor operationalization of
horizontal inequalities. Based on new data and new methods, we develop
alternatives to these conventional measures by analyzing group-level mech-
anisms explicitly.

The issue of grievances and violence may seem merely academic. Yet, very
much as successful medical treatments hinge on proper diagnosis, conven-
tional methods of conflict resolution depend critically on how the causes of
conflict are analyzed. Indeed, it comes as no surprise that those who dismiss
ethnonationalist claims as being both ubiquitous and irrelevant for conflict
tend to be skeptical about power sharing as a method for settling conflicts and
prevent renewed violence. Instead, these scholars focus on ways to prop up
weak governments and to help them improve their counter-insurgency cam-
paigns. Referring to Fearon and Laitin’s (2003) influential study in an article in
the New York Times, Bass (2006, p. 2) draws similar policy inferences: “The
Fearon-Laitin thesis suggests that the debate over the future of fragile countries
Introduction 7

should turn from questions of ethnic demography to the need for good govern-
ment, economic development and adequate policing.”
In view of our findings, however, without attention to ethnonationalist
grievances, such policies are likely to be ineffective, and in some cases pos-
sibly even counterproductive, especially in the long run. In very hard-to-solve
nationality conflicts, interventions based exclusively on “all sticks and no car-
rot” policies that do not address the underlying sources of grievances will typ-
ically fail (Petersen 2011). Instead, the best way to break the cycle of violence
driven by political exclusion and economic inequality is to involve groups that
have been marginalized by giving them a real stake in their country’s future.
Indeed, some of the most intractable and damaging conflict processes in the
contemporary world, such as the Israeli–Palestinian civil war, are to a large
extent rooted in political and economic injustice. It is very unlikely that they
will ever be resolved by shoring up the coercive capacity of the state alone unless
the claims of marginalized populations are taken seriously. In the concluding
chapter of this book, we will return to these important policy implications. For
now, we turn to the task of preparing the conceptual ground for our empirical
investigations.
part i

THEORIES AND CONCEPTS

T he first part of the book contextualizes our theoretical arguments with

respect to the civil war literature. After reviewing this literature in Chap-
ter 2, we introduce our causal mechanisms in Chapter 3.

9
2

Inequality and Grievances in the Civil War Literature

This chapter surveys the literature on the link from inequality and grievances to
civil war violence. As will become clear, the perceived validity of such factors
has remained controversial in conflict research. After considering the classical
literature, as well as some of the more recent writings in this area, we turn
to a critique of the literatures that inform our own theory-building efforts in
Chapter 3.

Classical Contributions
Given the obvious relevance of power and wealth distributions as potential
sources of conflict, it is not surprising that inequality plays a central role in
classical theories of civil war and revolution.1 In an influential article, Davies
(1962) argued that revolutions were motivated by frustrations resulting from
an evolving gap between individuals’ aspirations and their actual economic sta-
tus. Also adopting an explicitly psychological perspective, Ted Robert Gurr’s
(1970) well-known theory of relative deprivation characterizes various types
of collective violence as reactions to frustrations stemming from unfulfilled
aspirations, usually related to material well-being (see review in Brush 1996).
Such a perspective differs radically from earlier sociological theories of mob
behavior that explained collective violence as a societal pathology (e.g., Le Bon
1913). Instead, relative deprivation theorists argue that individuals’ widespread
discontent with their social situation triggers conflict, especially where mod-
ernization fuels a “revolution of rising expectations” (Davies 1962).
Although related indirectly to inequality through this psychological mecha-
nism, relative deprivation theory does not explicitly focus on interpersonal or
intergroup wealth comparisons (Hogg and Abrams 1988; Regan and Norton

1 This section draws on Cederman, Weidmann, and Gleditsch (2011).

11
12 Inequality, Grievances, and Civil War

2005; cf. Gurr and Duvall 1973). Yet, other theories adopt a structural perspec-
tive by explicitly linking various types of inequality to structural imbalances
in society, such as uneven income or land distributions (Russett 1964; Muller
1985; Muller and Seligson 1987). Partly inspired by Marxist principles, the lit-
erature on peasant rebellions explains violent collective action as a response to
unequal wealth allocation (Moore 1966; Scott 1976). Frustrated with their lot,
the peasant masses and other underprivileged groups are expected to take up
arms as a way to seize power and redistribute wealth in their favor. Huntington
(1968, p. 375) sums up the core logic of the argument:

Where conditions of landownership are equitable and provide a viable living for the
peasant, revolution is unlikely. Where they are inequitable and where the peasant lives
in poverty and suffering, revolution is likely, if not inevitable. . . . No group is more con-
servative than a landowning peasantry and none is more revolutionary than a peasantry
that owns too little land or pays too high a rental.

Relative deprivation theory remains perhaps the most prominent explana-

tion linking grievances to conflict, but this perspective has a very mixed
record in terms of empirical evidence (Oberschall 1978; Brush 1996). Early
on, Gurr’s frustration-aggression thesis attracted criticism from Snyder and
Tilly (1972), who argued that opportunity-based mobilization, rather than
grievances, causes internal conflict and revolutions. Contending that all soci-
eties contain a number of aggrieved and frustrated individuals, they did not
think “there is any general connection between collective violence and hardship
such that an observer could predict one from the other” (p. 520; see also Tilly
1978; Skocpol 1979). By this interpretation, widespread but diffuse anger does
not suffice to spark upheavals. What is needed is a minimum level of resources
and organization. Tilly (1973, p. 10) explains that

revolutions and collective violence tend to flow directly out of a population’s cen-
tral political processes, instead of expressing diffuse strains and discontents within the
population; . . . that the specific claims and counterclaims being made on the existing
government by various mobilized groups are more important than the general satisfac-
tion or discontent of those groups, and that claims for established places within the
structure of power are crucial.

Along similar lines, a series of studies challenges earlier results pertaining

to income inequality, which was usually seen to be closely connected to the
notion of relative deprivation (see, e.g., Weede 1987). By the end of the 1980s,
the debate about the effect of inequality and grievances on internal conflict
remained unresolved, with virtually all possible causal connections – includ-
ing negative, positive, curvilinear, or none – being represented in the literature
(Lichbach 1989).
The mixed empirical record of the literature did not prevent major theoreti-
cal progress being made toward articulating the relationship between inequality
and conflict. Much of this thinking stems from early work that broadens the
Inequality and Grievances in the Civil War Literature 13

narrowly materialist view of Marxist conflict theory to encompass theories that

stress ethnic and cultural aspects of political competition. Among these early
contributions, we find Gellner’s (1964) pioneering characterization of moder-
nity based on culturally and impersonally mediated communication as opposed
to premodern structural exchanges. Also criticizing both strictly materialist
explanations of mobilization processes, Michael Hechter (1978) introduced
the notion of “cultural division of labor” that accounts for the emergence of
grievances in cases where cultural differences coincide with occupational exclu-
sion. While this theoretical scheme does not deny the relevance of economic
forces, it accords the cultural dimension strong influence over political mobi-
lization, which becomes especially marked when it draws on both dimensions.
Based on this group-level logic, Hechter (1975) studied cases of “internal colo-
nialism” that features the exploitation of subordinate peripheries by dominant
ethnic groups, as illustrated by the relationship between the “Celtic fringe” and
the English heartland. As we will see, this group-based concept of inequality
anticipates Stewart’s notion of horizontal inequalities to which we will return
in Chapter 3.
Although often introduced as a competitor to Hechter’s notion of cultural
division of labor, a related explanation of ethnic conflict interprets it as a
result of group competition over scarce resources. Early examples of this per-
spective can be found in Deutsch (1961), Barth (1969), Hannan (1979), and
more recently in Olzak (2006). In these writings, competition and conflict
ensue when an invading group encroaches on an established group’s ecologi-
cal niche (Hechter and Okamoto 2001, p. 196). As opposed to modernization
theory and Marxist perspectives that expect ethnic differences to erode away
with economic development, both approaches to labor markets and intergroup
stratification along these lines see ethnic dominance and competition as inher-
ent parts of the modernization process. Yet, by retaining an explicit link to
production processes and resource competition, these theoretical contributions
do not take the full leap to viewing power as an end in itself and say less about
the role of political institutions in conflict processes. We therefore turn to
Donald Horowitz’s work, which explicitly advances the study of ethnic con-
flict in both these directions.
In what still remains a classical contribution with contemporary relevance,
Horowitz’s (1985) book Ethnic Groups in Conflict represents the most compre-
hensive study of ethnic conflict ever undertaken. Generally compatible with the
aforementioned group-based interpretation, his book develops sophisticated
theoretical arguments linking inequality and grievances to conflict through
causal mechanisms that feature cognitive comparisons among ethnic groups.
Building directly on Tajfel’s social psychology (see Tajfel and Turner 1979),
Horowitz’s seminal work introduces an explicit group perspective that over-
comes the individualism of Gurr’s relative deprivation theory. According to
his account, ethnicity serves as a metaphorical family concept that generates
solidarity and intense emotions of belonging. Importantly, his theory thus fills
14 Inequality, Grievances, and Civil War

important gaps in the causal derivations of purely structural theories. Instead of

merely relying on the abstract logic of ecological competition and production
forces, Horowitz uses crucial notions such as collective self-esteem and group
worth that help explain why ethnic conflicts often feature intense emotional
outbursts and sometimes even hatred.
As observed above, Horowitz analyzes positional power struggles that can-
not be reduced to material motivations. Refusing to succumb to economic
determinism, he explains how representatives of ethnic groups seek political
influence for its own sake either by excluding other groups from influence or
by seeking inclusion of their own group. This sense of “group entitlement”
manifests itself not just in terms of resource distribution, but also in terms of
cabinet seats and various symbolically charged state policies in the areas of
language, religion, and culture.
Despite its general relevance, Horowitz’s book primarily covers conflicts in
Asia, Africa, and the Caribbean. Possibly because of this geographic focus,
nationalism is hardly mentioned, although many of the conflict mechanisms
discussed by Horowitz relate to ethnic competition for state power and could
thus easily have been labeled in nationalist terms. Furthermore, within these
specific Third World regions, the book limits itself to so-called unranked polit-
ical systems, thus leaving ranked polities aside. The latter are characterized
by strict and long-standing hierarchies typically imposed through conquest.
Again, this possibly unduly reduces the scope of his study. Finally, despite its
manifest erudition and wealth of empirical cases, Horowitz’s classical study
does not attempt to evaluate the effect of its postulated causal mechanisms
systematically beyond case studies (usefully and often brilliantly), identifying
major patterns of conflict and their causes.
The challenge of collecting systematic data on ethnic groups was taken up
by Gurr in the 1980s. While Horowitz (1985) relied on erudite references to
a wealth of empirical cases, Gurr’s new research program launched a major
data collection effort, the Minorities at Risk (MAR) data set, which quickly
became the standard source of systematic data on ethnic groups. Thanks to
its comprehensiveness, in terms of both geographic coverage and the number
of variables included, the new data resource has spawned an entire research
program that analyzes various aspect of ethnic conflict.
The first set of results was published in Gurr (1993a; 1993b). Modify-
ing and extending his original relative deprivation theory, Gurr used his
new data to study ethnic minorities’ reactions to state-imposed disadvantages
and discrimination. In reaction to the criticism of Tilly and other resource
mobilization theorists, Gurr now stressed the role of mobilization alongside
grievances as a necessary component in conflict processes. Indeed, the new
results showed that “protest and rebellion by communal groups are jointly
motivated by deep-seated grievances about group status and by the situation-
ally determined pursuit of political interests, as formulated by group leaders
and political entrepreneurs” (Gurr 1993b, pp. 166–7). The follow-up study
Inequality and Grievances in the Civil War Literature 15

Gurr (2000b) broadened the theoretical framework to include “an ethiology

of conflict” based on ethnocultural identity, collective incentives, capacities of
collective action, and opportunities for political action. However, as suggested
by the normatively inspired title of the data set itself, Gurr never wavered
from the initial focus on grievances. This book and the influential article in
Foreign Affairs (Gurr 2000b) conclude that ethnic conflict had actually started
to decline thanks to a new spirit of accommodation and compromise after the
Cold War.
There can be no doubt about the past and future influence of Gurr’s data
and analysis on conflict research. True, the exposition of his research results
can sometimes be hard to grasp because of the richness of the data and Gurr’s
proclivity for complex causal diagrams. However, the main reason why the
MAR-based research program has attracted criticism relates to the data source
itself. We will come back to this issue at the end of this chapter. Together, the
classical works by Hechter, Horowitz, and Gurr have laid the theoretical and
empirical foundations for contemporary studies of grievances in ethnic conflict
and civil war processes, including the present book.

Grievance-Skepticism in the Recent Literature

Although serving as major source of theoretical inspiration for students of eth-
nic conflict, these classical studies have not remained unchallenged. In the late
1990s, a team of economists at the World Bank led by Paul Collier started
publishing a series of influential papers on the causes of civil war. Position-
ing themselves explicitly against grievance-based theories in political science
and sociology, while targeting the relative deprivation theory in particular,
these authors followed in the footsteps of earlier critics of relative deprivation
(although typically without referring to these explicitly). As a response to the
difficulties of measuring grievances directly, their approach usually relies on
quantitative indicators that attempt to capture ethno-demographic configura-
tions at the country level, for example through ethnic fractionalization or the
population share represented by the largest ethnic group. Below, we will discuss
the limitations of this approach.
In a much-cited article, Collier and Hoeffler (2004) claim that civil wars are
caused by “greed” rather than “grievance.” According to this “labor market”
approach to civil wars, wars are more likely where potential rebels have espe-
cially low opportunity costs of fighting and natural resources invite warlords
and their followers to enrich themselves by looting and rent-seeking. Still criti-
cizing grievances as causes of civil wars, more recent writings by Collier and his
associates weaken the stress on greed-driven arguments in favor of a broader
notion of “opportunities” (Collier, Hoeffler, and Rohner 2009).
In a particularly influential article, Fearon and Laitin (2003) also argue that
ethnic grievances are unrelated to political violence. Instead, they propose a
theory of insurgency that hinges on state weakness, especially in peripheral
16 Inequality, Grievances, and Civil War

areas with rough terrain. In contrast to Collier and Hoeffler’s predominantly

individualist and apolitical interpretation, Fearon and Laitin adopt a theoretical
perspective that focuses on the role of political institutions. Yet, by adopting
Collier and Hoeffler’s opposition between grievances and other factors, Fearon
and Laitin’s theoretical approach pits frustrations and inequalities along ethnic
lines against their own favored explanations, which highlight insurgency as an
unemotional technology that relies on the state’s weakness in peripheral areas
and on the corrupting influence of oil production. Because such insurgencies
tend to involve relatively small groups of fighters, the outbreak of civil war
should be unrelated to social grievances in the larger population. Arguing that
there is “little evidence that one can predict where a civil war will break out
by looking for where ethnic or other broad political grievances are strongest”
(p. 75), Fearon and Laitin warn policy makers and scholars against inferring

that ethnic diversity is the root cause of civil conflict when they observe insurgents in
a poor country who mobilize fighters along ethnic lines. Instead, the civil wars of the
period have structural roots, in the combination of a simple, robust military technology
and decolonization, which created an international system numerically dominated by
fragile states with limited administrative control of their peripheries (p. 88).2

Yet, already one year later, in Fearon’s (2004b) much-cited study of civil war
duration, ethnic grievances found their way back into the analysis, albeit in a
particular context. Drawing on Weiner’s (1978) classical concept of the “sons
of the soil,” his analysis of the duration of civil war shows that territorially
concentrated and peripheral ethnic groups tend to react violently to what they
perceive as incursions into their region by ethnically distinct migrants com-
ing from other parts of the country. In most cases these typically originate
from the dominant ethnic groups, therefore triggering governmental interven-
tions on their behalf. Since such constellations are notoriously difficult to settle
according to Fearon, they tend to last particularly long. Oddly enough, how-
ever, the article says little about how to reconcile the surprising finding that
a particular type of ethnic grievance matters a great deal for how long civil
wars last, whereas grievances are found to have no relevance whatsoever to the
outbreak of conflict, as argued by Fearon and Laitin (2003). We will elaborate
on this issue in Chapter 8, which explains the duration of internal conflict
primarily in terms of the exclusion of ethnic groups while suggesting that the
sons-of-the-soil category is a special case of ethnically excluded groups.
More recently, Fearon and Laitin (2011) devoted an entire article to sons-
of-the-soil conflicts, this time also considering the issues of grievances more
broadly as well as conflict onset. Attempting to account for the seeming dis-
crepancy between their more elaborate account of sons-of-the-soil dynamics

2 Apart from the work of Gurr and his colleagues that we referred to earlier, other quantitative
studies have found evidence of ethnic grievances mattering for the onset of civil conflict; see
Ellingsen (2000), Sambanis (2001), and Goldstone et al. (2010).
Inequality and Grievances in the Civil War Literature 17

that clearly features ethnic grievances and their original grievance-skeptical

position, the authors reiterate the point made in Fearon and Laitin (2003) –
and by Tilly and others decades earlier – that conflict outbreaks may be unre-
lated to broad social grievances because of the ease with which small bands
of insurgence can trigger large-scale violence. For the first time, however, the
analysis leaves open the possibility that ethnic grievances may actually be at
work. Although not entirely exogenous, the appearance of migrant populations
competing over the same local resources as the settled population constitutes
a shock to the latter, which raises the level of its discontent. Still reluctant to
back down from their general skepticism as regards the role of grievance in
civil war, Fearon and Laitin (2011, p. 209) now admit that “the cases we have
discussed suggest that this increase in grievance can produce civil war.”
The highly aggregated civil war literature, most prominently represented by
Collier and Hoeffler (2004) and Fearon and Laitin (2003), has come in for
sustained criticism by a new wave of scholarship that favors a radical disag-
gregation of the analysis down to the microlevel in order to tighten the logic of
causal inference (Sambanis 2004a; Kalyvas 2007; 2008b; Tarrow 2007). How-
ever, the substantive assumptions justifying these studies are usually not all that
different from those of the first wave that relied on quantitative, country-level
proxies. In particular, much of the research on the microdynamics of civil war
also tends to be skeptical of the role of grievances in conflict processes. This
tendency is evident in Kalyvas’ (2006) pioneering work. Questioning conven-
tional macro interpretations of civil war that hinge on conflict parties’ ideolog-
ical goals, Kalyvas argues in favor of a much more mundane perspective that
stresses private grudges and motivations. According to this view,
“political violence” is not always necessarily political identities and actions cannot be
reduced to decisions taken by the belligerent organizations, to the discourses produced
at the center, and to the ideologies derived from the war’s master cleavage (Kalyvas
2003, p. 487).

Similarly, recent research on the individual motivations for joining rebel move-
ments, such as Humphreys and Weinstein (2006), gives priority to material
incentives. In doing so, these writers build on earlier studies that applied
rational-choice analysis to attempts to overcome collective-action problems
through selective incentives (Popkin 1979; Lichbach 1995; see also references
in Blattman and Miguel 2010).
Of course, not all of the recent micro-based scholarship plays down
grievance-related explanations. Most prominently, Wood (2003) stresses the
central importance of moral outrage over governmental repression as a motiva-
tion fueling mobilization. In an effort to account for different types of violence
in civil wars, Weinstein (2007) proposes a theory that differentiates between
rebel organizations that recruit based on genuine grievances and those that
focus on materialist, short-term incentives. It is in the latter case that the vio-
lence can be expected to become indiscriminate. Furthermore, many of the case
18 Inequality, Grievances, and Civil War

studies in Collier and Sambanis’ (2005) edited volume that attempt to evalu-
ate the Collier and Hoeffler (2004) model reach conclusions linking political
and/or economic inequality directly to internal conflict (e.g., see Zinn 2005
on Nigeria; Ali, Elbadawi, and El-Batahani 2005 on the Sudan; Humphreys
and ag Mohamed 2005 on Mali; and Ross 2005 on Indonesia). Exploring the
micro-foundations of rebellion, Gates (2002) provides a formal framework
that allows him to integrate non-pecuniary awards and ethnicity in his account
of rebel recruitment. Likewise, Cuesta and Murshed (2010) introduce a game-
theoretic model that traces the relationship between microlevel motivations
based on greed and grievance. All in all, however, such work remains the
exception rather than the rule. Most of the literature on the micro-foundations
of conflict is characterized by a clear effort to punctuate the “myth” of ethnic
conflict by reinterpreting such conflicts as the result of opportunities seized by
self-interested actors.
Among the grievance skeptics, the theorists of “new wars” adopt perhaps
the most radical perspective. Dismissing the very notion of “ethnic conflict” as
a cover for the criminal activities of thugs and bullies, Mueller (2004) views
internal conflict as nothing but opportunistic predation that leaves no room
for higher motives relating to the righting of wrongs, or for more general
expressions of ideology or solidarity. Anticipating Kalyvas’ (2003) criticism of
theories that rely exclusively on macrocleavages, Mueller (2000, p. 63) stresses
the diversity of micromotives in civil war:

Some locals did join in the process, sometimes out of ethnic loyalty, sometimes to settle
old scores, most often, it seems, opportunistically to pursue profit in the chaos. In
many cases, the war conditions did bring out the worst in some people, and victims did
sometimes know their victimizers – though this is something that happens in most civil
wars, not just ethnic ones.

Explicitly denying that ethnic cleavages are more conflict-prone than other
divisions, Mueller views ethnicity as just one of many coordination devices
that could be exploited for collective mobilization:

Ethnicity proved essentially to be simply the characteristic around which the perpetra-
tors and the politicians who recruited and encouraged them happened to array them-
selves. It was important as an ordering device or principle, not as a crucial motivating
force. The same sort of dynamic could hold if the thugs’ organizational principle were
class or ideological allegiance or even handedness or loyalty to a specific soccer team
(p. 62).

Similarly skeptical of ethnicity’s inherent relevance in explanations of political

violence, other scholars have also questioned the raison d’être of the very
notion of ethnic conflict. For example, Charles King (2001, p. 168) agrees with
Mueller’s rendering of ethnic conflict as essentially so banal that it loses all
explanatory power:
Inequality and Grievances in the Civil War Literature 19

We cannot know for certain why large-scale violence, of whatever type, breaks out. Even
if we could, the factors involved would probably be disappointingly banal; clashing
economic interests, politicians’ attempts to oust opponents, lots of young men with
nothing to do and easy access to guns.

Subscribing to a similar interpretation, Gilley (2004, p. 1159) also argues that

it is virtually impossible to separate ethnic from nonethnic conflicts because
“[u]nlike ‘class conflict’ which can be proved or disproved by using pretty
stable measures of the people involved (income, education, occupation, etc.),
the same cannot be said of ethnicity.” In this view, ethnic identities are so fluid
and prone to manipulation that they cannot be meaningfully distinguished from
other types of attachments (cf. Chandra 2006).
Despite their diverse disciplinary origins, these critiques of grievance-based
approaches to conflict typically adhere to the same line of argumentation. In
particular, they make an important assumption that is usually left untested.
We refer to it as the ubiquity-of-grievances claim. It suggests that the outbreak
of internal conflict cannot be explained through grievances because these are
much too widespread to account for rare events such as large-scale political
violence. As we have seen above, this argument was originally introduced by
Tilly and other sociologists in the 1970s as a criticism of relative deprivation
(Oberschall 1978, p. 298; McAdam, McCarthy, and Zald 1988, p. 697; Piven
and Cloward 1991, p. 439). Yet, it keeps recurring as a leitmotif in the more
recent literature on civil war.
Explicitly advancing the ubiquity-of-grievances claim, Collier and Hoeffler
(2004) contend that since “grievances are sufficiently widespread to be common
across societies and time” (p. 589) and “all societies may have groups with
exaggerated grievances” (p. 564), their explanatory value remains insignificant.
Reviewing the findings of studies in the area, Laitin (2007, p. 25) confirms that
ethnic grievances are commonly felt and latent; the factors that make these grievances
vital and manifest differentiate the violent from the nonviolent cases. Ex ante measures
of grievance levels are not good predictors of the transformation of latent grievances
into manifest ones. And it is the factors that turn latent grievances into violent action
that should be considered as explanatory for that violence.3

Invoking the same logic in support of his wholesale rejection of grievance-

based arguments, Mueller (2000, p. 14) points out that “although expressions
of hatred between groups and peoples are exceedingly (and rather drearily)
common . . . violence among supposed haters is remarkably rare.” This obser-
vation has major implications for conflict:
Furthermore, although there may be plenty of hate, hostility, and animosity between var-
ious peoples, warfare does not seem to correlate well with these qualities. That is, it is not
notably found in the areas with the most intergroup hostility and cleavage (p. 15).

3 As we have already noted, Fearon and Laitin (2011, p. 209) are still committed to this argument
first advanced in Fearon and Laitin (2003).
20 Inequality, Grievances, and Civil War

The last part of Mueller’s explanation renders explicit a second assumption that
is often made together with the ubiquity-of-grievance claim, which we will refer
to as the irrelevance-of-grievances claim. According to this view, even if the
degree to which populations are aggrieved varies to some extent, this variation
is not strongly associated with the outbreak of civil wars, because once we con-
trol for the “root causes” of conflict, it becomes clear that opportunist leaders
exaggerate or invent their frustrations in order to pursue other, usually selfish
and materialist ends. Resembling the realists’ cynical view in international rela-
tions debates, this line of argumentation seeks to debunk liberal “myths” about
civil war that depict rebels and revolutionaries in a romantic light. Casting the
adherents of the “grievance” perspective as naive, the realist critics accuse them
of “listening too earnestly to the accounts of the combatants” (Laitin 2007,
p. 23): “When ethnic war breaks out, journalists congregate like ravens. They
inevitably ask combatants to tell narratives explaining the killing” (p. 23).
Whether appealing to either of the superpowers during the Cold War or to
“world opinion” in more recent years, such scholars claim that opportunistic
and self-interested rebel leaders will inevitably try to dress up their cause as a
way to attract support and resources.
The quantitative civil war literature typically operationalizes ethnic gri-
evances with aggregated demographic proxies, such as ethnic fractionalization
or individual-level inequality operationalized as the Gini coefficient (Fearon
and Laitin 2003; Collier and Hoeffler 2004), ethnic polarization (Montalvo
and Reynal-Querol 2005), or ethnic domination measured in terms of the size
of the largest group (Collier and Hoeffler 2004). These purported measures of
grievances are then compared with a battery of indicators that are meant to
capture explanations highlighting various aspects of self-interested motivations
and resource-based mechanisms. Summing up these and other studies, Laitin
(2007, p. 23) claims that quantitative data “undermine confidence in theories
purporting to show that national aspirations, differences, or demographies are
associated with communal violence or civil war.”

Weaknesses in the Grievance Skeptics’ Arguments

It is now time to question the criticisms advanced by the grievance skeptics. Our
argumentation focuses on three main weaknesses in recent conflict research:
namely, the bias in favor of individualist perspectives, the structuralist lack
of attention to explicit motivations, and the ethnic neutrality of the state in
explanations of civil war. We discuss each of these problems in turn.

Individualist Tendencies in the Conflict Literature

As we have argued in Chapter 1, conventional country-level indictors often
miss the target. Quick to dismiss grievances as explanatory factors, scholars
who rely on such measures often overestimate the theoretical appropriateness
of their proxies. In particular, the underlying assumptions supporting their
Inequality and Grievances in the Civil War Literature 21

methodological decisions converge on individualist thinking that is not nec-

essarily compatible with the “collectivist” theories that serve as the actual
targets of the criticism. For example, the commonly used ethnic fractionaliza-
tion index measures ethno-demographic diversity among individuals but does
not correspond to any well-known theory of ethnic conflict (Cederman and
Girardin 2007).4 More generally, several authors have pointed to the weakness
of aggregate proxies and the extent to which they fail to capture the theoretical
positions they are meant to operationalize (e.g., Posner 2004; Sambanis 2004a;
Cramer 2006; Chandra and Wilkinson 2008).
At the same time, the individualist ontology also pervades much of the more
recent literature on microdynamics. In particular, this orientation goes hand
in hand with a strong skepticism regarding group-level claims. Clearly, groups
do not act; only members of groups do. Therefore, there are good reasons
to treat essentialist claims about the unity of ethnic groups with considerable
caution. In a careful critique of the macroliterature on conflict, Kalyvas (2003,
p. 481) warns against reifying categories and ascribing them collective agency
by default:
The interchangeability of individuals that underlies the concept of group conflict and
violence is variable rather than constant. The locus of agency is as likely to be at the
bottom as at the top, so civilians cannot be treated as passive, manipulated, or invisible
actors; indeed, they often manipulate central actors to settle their own conflicts.

Individual decisions are just as likely to undercut the cohesion of a group as to

strengthen it. In many civil war settings, individuals have to make life-and-death
decisions by choosing sides between the government and the rebels. Targeting
essentialist assumptions of group unity common in the international relations
literature, Kalyvas (2008a, p. 1045) asserts that “ethnic defection” is much
more common than normally assumed, especially in the course of ongoing
conflict processes: “Even when ethnic divisions cause the eruption of civil war
in the first place, these identities do not always remain stable and fixed during
the conflict; if they do change, they may soften rather than only harden.”
We agree with much of this criticism targeting group-level analysis. Indeed,
there can be no doubt that the literature on ethnic conflict abounds with “ethni-
cist” interpretations that overlook internal rifts inside ethnic groups and impose
ethnic cleavages on an inchoate flux of identification, let alone any coherent
sense of collective agency. Ultimately, the crucial question is whether the dis-
connected score-settling at the micro-level adds up to any macro-level effects
that deviate in any substantial way from the inferences of macro-level theoriz-
ing. Indeed, the personal quarrels and village feuds at the individual level may

4 Arguing against the individualist foundations of the ethno-linguistic fractionalization index,

Fearon, Kasara, and Laitin (2007) propose an explanation of conflict between pairs of ethnic
groups that is based on multiplying their demographic sizes. Despite its formal interest, however,
this mechanism appears to be unrelated to any known and tested theory of conflict and has never
been further developed by the authors.
22 Inequality, Grievances, and Civil War

be nothing more than “noise” that does little to affect the logic of the main
conflict cleavages. Conducting his empirical analysis based on detailed archival
material, Kalyvas (2006) offers a wealth of historical evidence in support of his
postulated microdynamic regularities. Yet, in the absence of systematic sam-
pling of events across multiple countries and conflicts, it is difficult to generalize
beyond Kalyvas’ erudite references. Of course, Kalyvas is much too careful to
fall into the trap of extrapolating from single cases, rightly cautioning against
drawing sweeping conclusions about collective intentions, but other authors
contributing to the microliterature have certainly succumbed to this tempta-
tion (Kalyvas 2008b).5
As argued by the critics of group-level analysis, the issue of whether groups
can be said to meaningfully exist and engage in collective agency is ultimately
an empirical question that varies from case to case. Whether it makes sense
to generalize from observations of the civil wars in Greece or Afghanistan
depends very much on the scope of the comparison. However, some schol-
ars, in particular anthropologists, have taken their meta-theoretical quest to
constructivist extremes that deny the very usefulness of the notion of ethnic
groups (Jenkins 1997). As the most articulate proponent of this radical position,
Brubaker (2004, p. 4) elaborates on the problems associated with group-level
theorizing:
Groupness is a variable, not a constant; it cannot be presupposed. It varies not only
across putative groups, but within them; it may wax and wane over time, peaking
during exceptional – but unsustainable – moments of collective effervescence. Ethnicity
does not require such groupness. It works not only, or even especially, in and through
bounded groups, but in and through categories, schemas, encounters, identifications,
languages, stories, institutions, organizations, networks, and events.

Continuing his antiessentialist reasoning, Brubaker offers recommendations for

theory-building in the area of nationalism and ethnic politics:
Ethnicity, race, and nation should be conceptualized not as substances or things or
entities or organisms or collective individuals – as the imagery of discrete, concrete,
tangible, bounded, and enduring “groups” encourages us to do – but rather in relational,
processual, dynamic, eventful, and disaggregated terms (p. 11).

Taking a further step toward de-essentializing the theorization of contentious

politics, Brubaker arrives at the logical endpoint of his argumentation: not only
should “groups” be dropped from social-scientific theories, but also the very
notion of “identity” (Brubaker 2004).
In our view, such radical theoretical advice goes too far. Brubaker’s perspec-
tive is doubtlessly useful for case studies of fluid micro-level situations where
identities are acutely contested as well as macro analysis of the longue durée

5 In an explicit quantitative test focusing on the Israeli-Palestinian conflict, Miodownik, Bhavnani,

and Choi (2011) find empirical support for Kalyvas’ main theory about the intensity of violence
as a function of the distance from each side’s base.
Inequality and Grievances in the Civil War Literature 23

inspired by process-theorists such as Norbert Elias ([1939] 1982).6 Yet, this

does not imply that group-level assumptions should be abandoned for all types
of analysis. In particular, where the time spans are more limited and the ideolog-
ical “glue” of ethnonationalist politics generates considerable group cohesion,
the assumption of stable group identification makes a lot of sense.
For the situations studied in this book, group-based approaches are likely
to apply particularly well because the threat and escalation of conflict tend to
reinforce collective identities and group cohesion (Coser 1964). In his path-
breaking theory, Simmel (1955) argues that such pressures may even bring
groups into being where they did not exist in the past. While political violence
clearly accelerates identity formation, it is important to realize that such pro-
cesses are often set in motion well before the outbreak of large-scale political
violence. For this reason, Brubaker (2004, p. 19) may slightly overstate his
point when he calls for “sensitivity to the variable and contingent, waxing and
waning nature of groupness, and to the fact that high levels of groupness may
be more the result of conflict (especially violent conflict) than its underlying
cause” (see also Kalyvas 2008a). Indeed, gradually escalating, nonviolent con-
flict typically precedes violent phases and there is no reason to believe that
the former cannot also contribute to the crystallization of the identities and
boundaries of ethnic groups. Yet, rather than speculating about the sequence
in purely theoretical terms, we offer a systematic test of to what extent spe-
cific preconflict configurations involving ethnic groups can be used to explain
conflict duration in Chapter 8. Without ruling out the possibility of reverse
causation, we suspect that the strongly constructivist interpretations that doubt
whether ethnic groups and their grievances actually predate or preexist in con-
flict processes have been exaggerated, especially given the relative inertia of
ethnonationalist constellations (Calhoun 1997, p. 32). To see this point, one
does not have to subscribe to the opposite, radically essentialist claims of
Kaufmann (1996), who insists that conflict “locks in” ethnic identities to
such an extent that they can never change again. Ironically enough, hard-
liners advocating uncompromising and repressive state policies are only too
happy to lend an advertently, and sometimes even intentionally, helping hand
to entrepreneurs engaged in ethnonationalist mobilization against the incum-
bent government (see also Chapter 3).
Without accepting the unity of ethnic groups as an ontological certainty,
then, we adopt a cautiously pragmatic position. Our starting point is a Webe-
rian conception of ethnicity that views it as any subjectively experienced sense
of commonality based on the belief in common ancenstry and shared culture
(Weber 1978, pp. 385–98; see also Cederman 2013, p. 533). Ethnic groups,
then, are cultural communities based on a common belief in putative descent.
Rather than assuming that ethnic groups remain constant over long historical

6 For an argument on how to endogenize macro-level identities in international relations, see

Cederman (1997).
24 Inequality, Grievances, and Civil War

periods, we treat them as sufficiently stable to enable us to exogenize their

properties in the shorter time scales that are relevant to onset analysis. In Parts
II and III, we offer empirical support for this level of theorizing. If our pos-
tulates are borne out by empirical evidence at the macro level, we cannot be
certain of being right in all respects, but we certainly have good reasons to
doubt that micro-level dynamics are so powerful that they distort or dominate
higher-level conflict processes.

Opportunity-Driven Structuralism at the Expense of Motivation

It may seem paradoxical that conventional approaches to civil war suffer from
an excess of both structuralism and individualism. However, this apparent
paradox dissolves once one realizes that structural theories are very much
compatible with a focus on individual decision making as long as strong,
uniform assumptions are made regarding the main actors’ preferences. Indeed,
the individualist thrust present in recent scholarship on civil wars can be partly
traced back to economic theories of individual incentives and rentseeking. The
persistent unwillingness to accept grievance-based accounts of internal conflict
reflects a deep ontological belief in actors’ selfish and unsentimental pursuit
of their narrow material interests. This is what we have already referred to as
“antipathy to antipathy” (Horowitz 2002).
Nobody has brought this reasoning to the point more cogently than Mancur
Olson (1965), whose radically individualist thesis questions the viability of
group-level motivations of any kind, especially those that draw on ideology or
other nonmaterialist ideals. According to this view, rather than contributing to
a group’s agenda, most individuals will shun the risks of armed conflict while
hoping to get a free ride on the risk-taking of others (Tullock 1971). This crass
perspective has influenced several generations of conflict researchers, as we
have seen, starting most prominently with the school of resource mobilization
in the 1970s.
In principle, the focus on organizational and logistical obstacles to collective
action is clearly a sound one, but it has led to unreflective theorizing (see the
critique in Kalyvas and Kocher 2007), and to an almost total abandonment of
collective grievances as a possible cause of mobilization and conflict:

In the 1970s, resource mobilization and political process theorists stopped asking why
people felt frustrated enough to engage in collective protest rather than organize through
conventional political channels, and instead asked when and how they secured the
resources to combat their exclusion from those channels (Polletta and Jasper 2001,
p. 286).

Until recently, most of the social movement literature has been dominated
by notions of “opportunity structures” at the expense of an explicit analy-
sis of actors’ reasons for fighting and of the role of emotions in mobilization
processes, but grievances are often present in accounts of “framing” and “nar-
ratives” (Opp 2009, ch. 6).
Inequality and Grievances in the Civil War Literature 25

In many respects, the most recent waves of political-economy scholarship

resemble this originally sociological perspective. Keen to employ objective indi-
cators in their quantitative analysis, these researchers are less concerned with
the precise motivations of the actors in question. Instead, conflict is explained
as an outcome of particular “technologies” of combat that become effec-
tive under objective circumstances that relate to the availability of natural
resources or the power projection of the state or rebel organizations. Indeed,
the most prominent proponents of the greed-based interpretation of civil war
have retreated to a completely open-ended position that privileges “opportu-
nities” over “motives”: According to Collier, Hoeffler, and Rohner’s (2009)
“feasibility hypothesis,” it can be expected “that where a rebellion is feasible
it will occur” (p. 2).
However, the stress on “opportunities” is not the only reason why actors’
behavioral motivations have fallen by the wayside. Recent theories of ethnic
conflict, in particular, have opted for an explicitly cognitive viewpoint that
serves to inject a healthy dose of scientific thinking into a field that has tended
to generate more heat than light. Inspired by cognitive psychology and rational
choice theory, this quest to “demystify” and to “sanitize” cruel acts of political
violence offers a certain comfort to external observers by rationalizing them in
terms familiar to mainstream political science and neoclassical economics.
The last few decades of conflict research, especially in the field of interna-
tional relations, have been profoundly influenced by both cognitive psychology
and rational choice theories. In fact, theories drawn from these literatures were
engaged in a heated debate concerning the best way to conceptualize nuclear
deterrence during the Cold War (Achen and Snidal 1989; Jervis 1989), but
have more recently had a strong influence on explanations of ethnic conflict
in the post–Cold War period (e.g., Stein 2002). Such studies often rely on
Tajfelian Social Identity Theory as a source of inspiration, which leaves some
room for emotions in its stress on group members’ need for self-esteem (Tajfel
and Turner 1979). Yet, more recent developments in social psychology have
consolidated the dominance of cognitive interpretations at the expense of moti-
vational and emotion-based explanations of conflict (Hale 2008). Applications
of rational choice theory to ethnic politics have further reinforced the central-
ity of cognitive mechanisms in conflict research. Ethnic identities are seen as
convenient “focal points” that serve as coordination devices in mobilization
processes (Hardin 1995), as convenient boundary markers that support dis-
tributional claims (Posner 2005), or as cost-reducing arrangements that boost
interethnic trust (Fearon and Laitin 1996).
The downside of these efforts seems to be a strangely unemotional and
bloodless rendering of conflict processes that are totally divorced from the
hatred and resentment witnessed by countless qualitative accounts of conflict
(cf. Horowitz 1985).7 Again, there has been a widespread tendency to set up

7 For these reasons, one should be careful not to draw too far-reaching inferences from experi-
mental studies that have become popular in the conflict literature. Obviously, there is only so
26 Inequality, Grievances, and Civil War

a theoretical straw man, in this case associated with “ancient hatred” inter-
pretations of ethnic violence (Petersen 2002). Having dismissed any reference
to emotions in conflict analysis as belonging to the domain of sensational-
ist journalists and uninformed politicians, these mainstream analysts quickly
dump the emotional ballast in favor of less troublesome factors that fit the
well-tested tools of political analysis. But such heavy-handed criticism risks
throwing out the motivational baby with the primordialist bathwater. Emo-
tions are not inherently irrational and they matter hugely in politics, especially
in heated conflicts that have the potential to erupt into large-scale violence
(Petersen 2002; Mercer 2005). Ultimately, the behavioral effect of emotional
mechanisms remains an empirical issue and it is therefore a mistake to rule
them out by theoretical fiat.

Critique of the Ethnic Neutrality of the State in Explanations

of Civil War
Apart from individualism and insufficient attention to explicit motivations,
a third tendency in the contemporary conflict literature contributes to the
analytical neglect of grievances. Although civil wars take place in situations that
confront incumbent governments with political organizations that challenge the
government’s claim to sovereign rule, the state is often strangely absent from
explanations of how ethnicity matters for conflict. Some accounts accord the
state a much more active role in conflict processes, but treat it as an ethnically
neutral “arena” that does not have any particular stake in clashes involving
ethnic groups other than attempting to provide law and order more or less
effectively. The use of demographic proxies as measures of “ethnic grievances”
reinforces the apolitical rendering of ethnicity that fails to account for it as an
explicitly political relationship of power. Ethno-demographic measures, such
as fractionalization and polarization, are blind to politics and cannot therefore
capture the conflict-causing effect of politicized ethnicity. For example, Fearon
and Laitin’s (2003) approach to civil war accords the state a central role in
their main conflict-inducing mechanism. Yet, their notion of state weakness
is entirely orthogonal to ethnic politics. Instead, ethnicity enters their analysis
separately as a measure of societal diversity along ethnic lines.
In other cases, especially those characterized by state failure, the state is not
merely neutral, but even absent, as in Posen’s (1993) theory of the “emerging
security dilemma.” Drawing on the notion of anarchy from international rela-
tions theory, this well-known explanation of internal conflict stipulates that
state collapse opens the doors to interethnic fighting. Similarly, theories of
“new wars” that play down the extent to which civil war features actors with
explicitly political agendas tend to assume anarchic conditions (Mueller 2004;
for a trenchant critique, see Kalyvas 2001).

much an experimenter can do to emulate the social context and emotional stakes of real conflict
within the walls of a laboratory.
Inequality and Grievances in the Civil War Literature 27

We argue that these theoretical shortcomings stem from a tendency to high-

light ethnicity in ways that neglect the role of nationalism. At least since the
second half of the nineteenth century, most conflicts involving ethnic categories
concerned politicized ethnicity rather than ethnicity per se. Indeed, the key to
politics in the modern world hinges on the control of the state. In his sustained
criticism of “stateless” theories of ethnic conflict, Brass (1991, p. 250) reserves
a central place in conflict processes for the modern state:

The state is itself both a resource and a distributor of resources. It is not an abstraction,
but a set of repressive, allocative, and distributive institutions and decisionmaking
bodies. Its functioning is facilitated and the potential for violent conflict in the struggle
for control over it is reduced if the state operates behind a veil of legitimacy and if
contestants in the struggle for control operate according to widely accepted rules.

As long as the focus is mainly on short-term material incentives, such as oil and
diamonds, or more generally on individual opportunity costs, it is easy to lose
sight of the profound importance of securing access to executive control over
the state. Of course, in clientilist regimes, state power often entails material
wealth, but in the era of nationalism, controlling the state can also be an end
in itself (Horowitz 1985; Wimmer 2002). As we will argue in the next chapter,
those who are excluded from influence over governmental decision making,
and forced to live under undignified “alien rule,” are the most likely to rebel
against the status quo. Therefore, attempts to reduce ethnicity to apolitical
ethno-demographic measures, such as fractionalization or polarization, will
always fail to capture the arguably most important class of grievances, namely
those relating to uneven distribution of political power along ethnic lines.

Data Problems Afflicting the Recent Grievance Literature

Before introducing our own theoretical framework in the next chapter, we end
the current one with a critical discussion of the data-related challenges that
confront recent studies positing a link between grievances and internal con-
flict. If the grievance skeptics’ arguments succumb to theoretical pitfalls, the
empirical message of the grievance proponents has been muddled by serious
data problems, most of them related to the original design of the Minorities
at Risk project that we introduced earlier in this chapter. By focusing on the
notion of minorities “at risk,” that is, minorities that are already discriminated
against or otherwise disadvantaged, this data set breaks new ground as regards
political relevance, but unfortunately at the price of introducing potential selec-
tion bias. While the sample includes some “advantaged” minorities, it remains
incomplete since Gurr and his colleagues do not include many powerful groups,
some of which may decide to challenge the incumbent government (Cederman,
Wimmer, and Min 2010).
Of course, selection bias only applies in relation to specific research ques-
tions and is not an inherent property of data sets, as noted by Posner (2004)
28 Inequality, Grievances, and Civil War

and Hug (2013). To the extent that researchers focus on ethnic group charac-
teristics that are unrelated to grievances, such as whether they are religiously
or linguistically defined (Chai 2005) or Saideman’s (2001) explanation of why
some groups receive external support and others do not, the issue of selec-
tion bias do not necessarily pose problems (see the discussion in Hug 2013).
More importantly for our purposes, most MAR-based research that evaluates
the effect of discrimination and grievances at the country level escapes this
dilemma since it treats such variables as independent or explanatory factors
aggregated to the country level rather than outcomes for individual groups.
While aggregation to the country level may generate bias due to nonrandom
measurement error (Hug 2013), which may or may not be a problem with
respect to MAR’s sampling, the main conclusions of important studies such
as Regan and Norton (2005) and Goldstone et al. (2010) are not likely to be
affected by the known problems relating to selection bias.
In an early challenge to Collier’s greed-based perspective, Regan and Norton
(2005) base their analysis on discrimination aggregated to the country level.
In confirmation of Gurr’s updated relative deprivation perspective, they con-
clude that grievances play an important role in accounting for the outbreak of
different stages of internal conflict while finding no support for Collier’s rad-
ical interpretation. Because of its highly aggregated nature, the article cannot
offer a definitive adjudication between competing causal mechanisms, but its
theoretical framing is very insightful and has inspired the current book.
Framing their work, as we also do, in direct opposition to both Collier and
Hoeffler (2004) and Fearon and Laitin (2003), Goldstone et al. (2010) repre-
sents an impressive attempt to highlight the importance of discrimination as a
major predictor of a broad set of politically unstable situations, including civil
war. This study is especially innovative since it adopts an explicitly predictive
framework and focuses on out-of-sample evaluation. It is by now well known
that influential models in the quantitative civil war literature are guilty of con-
siderable over-fitting in-sample and therefore serve as particularly poor tools of
out-of-sample forecasting (Ward, Greenhill, and Bakke 2010). In their pioneer-
ing study, the team led by Goldstone found that a model that includes political,
institutional variables related to regime type, infant mortality, conflict-ridden
neighborhoods, and, for our purposes most importantly, state-led discrimina-
tion of ethnic groups, performs especially well. As regards the latter indicator,
Goldstone et al. report that
countries with high levels of state-led discrimination against at least one minority group,
according to the Minorities at Risk political or economic discrimination indicators,
faced roughly triple the relative odds of future civil war onsets than those without such
discrimination (p. 197).8

8 It should be noted that this central result anticipates our own analysis, regarding both the effect
of political discrimination and exclusion (see Chapters 4 and 7) and economic inequalities (see
Chapters 5 and 7).
Inequality and Grievances in the Civil War Literature 29

Again, the discrimination data are pulled from MAR without causing any
serious inferential problems relating to a skewed selection of cases since the
entire analysis is conducted for entire states. In contrast, scholars who have
been tempted to extract more detailed information from this data set about the
workings of conflict-inducing mechanisms at the level of ethnic groups are less
immune to the problem. This certainly applies to some of the conclusions in
Gurr’s own work, but it is probably fair to say that a majority of the MAR
literature continues to struggle with the built-in limitations of “minorities at
risk” as the main units of analysis. As we have seen, in these cases, the risks
to causal inference can be potentially quite serious as reported by Hug (2003;
2013) and Fearon (2003).
We suspect that these data-generated difficulties constitute one of the major
reasons why students of ethnic conflict and civil war have remained skeptical
as regards the validity of explanations prominently featuring inequalities and
grievances. Rather than merely using countries as units of analysis as a way
to protect our analysis from threats to causal inference, we prefer to evaluate
grievance-related explanations at the group level for theoretical reasons already
stated earlier in this chapter. Fortunately, the Ethnic Power Relations data set,
which we will introduce in Chapter 4, allows us to have it both ways.
After this methodological and empirical interlude, however, it is time to
return to the main theoretical task of making sense of inequality and grievances
in processes that generate internal conflict. This is an urgent task, because we
have criticized the empirical literature mostly on conceptual and theoretical
grounds, arguing that the main reason why the grievance-skeptics have not
been able (or willing) to find support for the explanations they criticize is
mainly related to their use of theoretically inappropriate indicators. As argued
in Cederman and Girardin (2007), we need to move our theorizing beyond
“off-the-shelf” measures that tap into various ethno-demographic aspects of
societies, such as ethnic fractionalization, since these fail to do justice to the
political logic of ethnonationalism and thus reduce grievance-based explana-
tions to theoretically irrelevant straw men.
3

From Horizontal Inequality to Civil War

via Grievances

What would it take to show that grievances actually do cause political violence?
In Chapter 2, we noted that it is difficult to measure grievances directly, at least
if the empirical scope encompasses a large number of cases. Therefore, we
opt for an indirect approach that identifies conditions under which grievances
are likely to emerge. In particular, we posit that inequalities among groups
correspond to such situations, and proceed by investigating whether they are
related to the outbreak of civil war in the empirical chapters to follow. In this
chapter, we propose a combination of theoretically grounded mechanisms that
together constitute a causal pathway that connects inequalities with violent
conflict through grievances. Without a direct test of the grievance mechanisms
themselves, it is all the more important to bolster their credibility by showing
that they can be derived from conceptually coherent and empirically validated
theories.
The main goal of this chapter is to construct a theoretical scheme that
explains how inequalities cause violent conflict. Before doing so, however, it is
appropriate to briefly introduce two key types of inequalities and their historical
origins.

Vertical and Horizontal Inequalities

Our first task is to conceptualize inequality as a structural, asymmetric condi-
tion governing social relations among actors. Given Charles Tilly’s role as a
prominent critic of grievance-based theorizing in general, and relative depri-
vation theory in particular, it may seem surprising to use one of his works as
an initial source of inspiration for our theorizing. Yet, Tilly’s (1999) provoca-
tive book on Durable Inequality is a useful conceptual starting point because
it helps overcome the problems afflicting the conventional conflict literature.

30
From Horizontal Inequality to Civil War via Grievances 31

More specifically, it addresses head-on the three main weaknesses that we

identified toward the end of the previous chapter.
First, Tilly’s approach to categorical inequalities distances itself explicitly
from individualist interpretations that view “inequality and poverty as out-
comes of individual-by-individual competition according to widely shared stan-
dards of merit, worthiness, or privilege” (Tilly 2006, p. 4). According to Tilly,
such a view “obscures the significance or organized distinctions and interac-
tions among members of different categories” (p. 4).
Second, this contribution deviates from Tilly’s earlier writings by focusing
much more consistently on the role of exploitation in social processes. This
perspective makes it straightforward to derive motivations stemming from
frustrations with such situations. In this sense, Tilly’s theory “builds a bridge
from Max Weber on social closure to Karl Marx on exploitation, and back”
(Tilly 1999, p. 7).
Third, and finally, Tilly’s book incorporates an explicitly institutionalist
perspective that can be characterized as an “organizational view of inequality-
generating mechanisms” (Tilly 1999, p. 8). Organizations, rather than sponta-
neous and “societal” competition, impose structure and reinforce asymmetric
distributions of power and wealth. As we will see, this aspect allows us to
appreciate the central role played by the state in conflict processes.
All in all, Tilly’s Durable Inequality advances a sweeping and general
approach to the emergence of inequalities. Most centrally, Tilly argues that
the most durable inequalities in social systems emerge along categorical dis-
tinctions, such as gender, ethnicity, or citizenship, rather than along more
gradual differences pertaining to individual attributes. Embedded in social
relations, especially within organizational structures, categorical inequalities
are produced by a set of mechanisms, most prominently by “exploitation”
and “opportunity hoarding,” and then further cemented by “emulation” and
“adaptation.” Echoing Marxist theory, Tilly (1999, p. 10) defines exploita-
tion as a situation where “powerful, connected people command resources
from which they draw significantly increased returns by coordinating the effort
of outsiders whom they exclude from the full value added by that effort.”
Similarly, opportunity hoarding corresponds to settings where “members of
a categorically bounded network acquire access to a resource that is valu-
able, renewable, subject to monopoly, supportive of network activities, and
enhanced by the network’s modus operandi” (Tilly 1999, p. 10).
More recently, Stewart (2008b) and her colleagues have developed the
notion of “horizontal inequalities,” which proposes a more narrowly applied
conceptualization of inequality than that implied by Tilly’s categorical inequal-
ities, while linking it specifically to conflict among identity groups. Contrasting
this concept to vertical or individual-level inequalities, Stewart (2008a, p. 3)
defines horizontal inequalities as “inequalities in economic, social or political
dimensions or cultural status between culturally defined groups.” If vertical
32 Inequality, Grievances, and Civil War

inequality compares individuals and households without sorting them into cat-
egories, horizontal inequality compares entire group identities to each other
without paying any attention to the internal diversity of each category.
As does Tilly’s approach to categorical inequalities, Stewart addresses the
three weaknesses afflicting the conflict literature that we highlighted in Chap-
ter 2. First, Stewart (2008a, p. 3) argues that, to a large extent, much of the
conventional quantitative research has failed to find evidence for a positive
effect of greater inequality on civil war because of its reliance on individualist,
rather than group-based, measures of income and power differences:
But the majority of internal conflicts are organized group conflicts – they are neither
exclusively nor primarily a matter of individuals committing acts of violence against
others. What is most often involved is group mobilization of people with particular
shared identities or goals to attack others in the name of the group (Stewart 2008a,
p. 11).

Second, Stewart (2008a, p. 18) stresses the central importance of grievances as

motivations for violence conflict:
strong political [horizontal inequalities] mean that leaders of groups feel politically
excluded and are thus more likely to lead opposition and possibly rebellion; while
socioeconomic inequalities mean that people as a whole have strong grievances on
ethnic lines and are thus likely to be more readily mobilized.

Third, her explicitly institutional focus pays special attention to the structure
of the state and the nature of its policies:
Whether democratic or not, governments can be accommodating and inclusive making
violent opposition less likely. In contrast, governments (especially nondemocratic ones)
may make no attempt to meet people’s demands and can react to opposition with harsh
repression, which may provoke a further violent reaction (Stewart 2008a, p. 20).

Following the lead of Horowitz and Gurr, Stewart (2008a) conceptualizes

horizontal inequality by considering political, economic, social, and cultural
dimensions. Political horizontal inequalities entail blocked or limited access to
central decision-making authority within the state. In this book, we refer to this
type of power asymmetry as political exclusion. The economic dimension taps
the distribution of income between groups. Social horizontal inequality mea-
sures primarily groups’ uneven social access, for example in terms of education
and societal status. Finally, the cultural aspect captures group-level inequalities
with respect to cultural policies and symbols, including traditional holidays
and religious rights.
In this book, we limit ourselves to an analysis of political and economic
horizontal inequalities because they are arguably the most important causes of
conflict, and certainly those that are the least difficult to measure systemati-
cally. Although these two types of inequality are often likely to be related, we
follow Stewart in treating the distribution of power and wealth as conceptually
From Horizontal Inequality to Civil War via Grievances 33

separate components. Of course, this does not mean that social and cultural
inequalities are unimportant in this respect. Quite the contrary, access to edu-
cation (Gellner 1983) and state policies regarding language and religious rights
(Langer 2005) are known to be powerful triggers of conflict. Moreover, politi-
cal and economic horizontal inequalities have major consequences for, and are
influenced by, cultural and social asymmetries at different societal levels. While
the former pertains mostly to elite-level influence, such as the share of cabinet
posts, the latter measures the economic wellbeing of an entire population.

The Origins of Horizontal Inequalities

Although our main task is to explain the effect of inequalities on internal
conflict via grievances, we need to place our causal mechanisms in their proper
macro-historical context. Where do horizontal inequalities come from? Group-
level inequalities among humans have existed since the emergence of agriculture
and sedentary habitation (Malesevic 2010) and may even be inherent to any
complex social system (Sidanius and Pratto 1999). Whereas most sociologi-
cal explanations of power differentials downplay or entirely ignore violence,
we agree with Malesevic (2010) that historically long-standing group-level
inequality to a large extent stems from warfare and domination. Since ancient
times, processes of conquest and colonization have been the most important
drivers of political and economic inequalities (Williams 2003, p. 59). In his
multi-volume work on social power, Mann (1986) offers the perhaps most
elaborate account of how warfare and domination are capable of transform-
ing patterns of inclusion and exclusion. Stressing the autonomous role played
by military power, Mann shows that political influence cannot be reduced to
forces of production. By turning Marx’s historical materialism on its head, this
approach to stratification clarifies how military superiority produced political
and economic dominance rather than the other way around.
Yet, the causal priority of geopolitical factors in the premodern world
has increasingly been undermined by popular sovereignty and mass politics.
Although war and conquest gave rise to major societal cleavages that are still
with us today as residues of previous colonization and dominance hierarchies,
nationalism fundamentally transformed the process of state formation and put
an end to empire building, with profound consequences for social differenti-
ation. In the new era of national self-determination and popular sovereignty
that followed after the American and French Revolutions in the late eighteenth
century, it become increasingly difficult and costly to conquer territory, let
alone to control it against the will of the local population (Gilpin 1981, p. 142;
Rosecrance 1986; Kaysen 1990).
While nationalism slowed down the power-driven stratification processes of
the premodern world, it generated new inequalities based on processes of mass
mobilization and cultural invention (Anderson 1991). In the modern world,
cultural, and especially ethnic, stratification increasingly influences access to
34 Inequality, Grievances, and Civil War

political power. Developments in communication technology and mass liter-

acy, together with doctrinal changes pertaining to national self-determination
and popular sovereignty, paved the way for this profound transformation of
political legitimacy (Calhoun 1993).
Focusing on the emergence of party politics in Europe from the nineteenth
century, Rokkan (see Flora 1999) offers the most comprehensive account of
how this process produced power differentials that promoted some regions as
centers of power and demoted others to the status of powerless peripheries.
According to Rokkan’s “freezing hypothesis,” these processes involved consid-
erable path dependence in terms of mobilization. Once mobilized as political
parties or social movements, cultural identities became organizationally locked
in, thus rendering the invention of new mobilization platforms difficult, if not
impossible.
In premodern times, human societies lacked the mass-based categorical
abstractions and means of communication to transform such macro cleavages
into mass-based political action. According to Gellner (1964), the nationalist
mode of politics centers on identification carried by a shared and standardized
high culture that replaced the direct interpersonal relations of smaller social
units, such as the village, in the premodern world. In the modern era, collective
identification with, and emotional attachment to, the abstract concept of the
nation made mass mobilization possible beyond local communities based on
direct contact.
This fundamental rewiring of social connectivity had major consequences
for geopolitics. As opposed to the “color-blindness” of traditional power pol-
itics, nationalist politics hinges on who is in and who is out. This way of
sorting people created new boundaries and differences, usually along ethnic
lines, which politicized distinctions that had previously been of no political
consequence. As argued by Tilly (1999, p. 172), in this very sense, nationalism
can be thought of as a generator of “categorical inequality” since

it asserts and creates paired and unequal categories, either (a) rival aspirants to nation-
hood or (b) members of the authentic nation versus others. It involves claims to prior
control over a state, hence to the exclusion of others from that priority. It authorizes
agents of the nation to subordinate, segregate, stigmatize, expel, or even exterminate
others in the nation’s name.

From this vantage point, the era of nationalism could thus be likened to the
game of musical chairs. Given that there are many more potential peoples that
could claim nationhood than there are viable governance units in the modern
state system (Gellner 1983, p. 2), fierce political competition concerns who
is going to secure a seat of power. However, when the “music” stops, many
nationalities, such as the Kurds and the Palestinians, will still be left without
a seat. In other cases, several competitors will have to share a slot, a situation
that often causes tensions and possibly even open conflict.
From Horizontal Inequality to Civil War via Grievances 35

Rather than tracing this development historically over centuries, however,

this book analyzes the consequences of nationalism comparatively in the con-
temporary world, mainly focusing on internal conflict during the post–World
War II period. By the end of World War II, the Western world had to a
large extent become ethnically homogenized, while war-weariness reduced the
attractiveness of exclusionary nationalism (Barkin and Cronin 1994; Mann
2005). Beyond this area, nationalist principles continued to spread to every
corner of the globe, but the diffusion process was highly uneven. The uneven
spread of nationalism delayed mass-level political mobilization in many parts
of the world, thus creating differences in terms of both economic and political
development that were often exploited by alien rulers (Gellner 1964). Viewed
as an important part of this macrohistorical transformation, the decoloniza-
tion process generated a large number of weak and ethnically heterogeneous
states, many of which soon experienced their own serious nationality problems,
rooted in patterns of political and economic exclusion along ethnonationalist
lines (Jackson 1990).
Of course, the decision to exclude and dominate ethnic groups does not
only depend on ideological commitments to abstract principles of nationalism.
It also has more prosaic reasons, such as the desire to control wealth accruing
from governmental control over raw materials and trade flows (Bates 1983).
In some cases, rulers may even prefer to fight low-intensity wars with excluded
groups at the periphery of the country rather than exposing themselves to the
risk of being stabbed in the back by power-sharing partners through military
coups (Roessler 2011). Moreover, besides nationalism, both political and eco-
nomic inequalities can result from global processes relating to technological
innovation and processes of globalization (see, e.g., Olzak 2006).

Conflict-Generating Mechanisms
Having conceptualized and traced the origins of vertical and horizontal inequal-
ities, we now turn to the crucial causal link between inequality and conflict.
The starting point of our account of ethnonationalist warfare builds on the
realization that ethnic groups find themselves in radically different situations
for various historical reasons. Whereas some ethnic groups came out on top
in the geopolitical game, others were conquered and colonized early on, and
therefore lost out in the competition for wealth and influence.
Our strategy is to explore the role of grievances indirectly by investigating
the empirical link between inequalities and civil war outbreak at the level of
groups. Of course, it is in principle possible to measure grievances directly (e.g.,
Petersen 2002) and to trace mobilization processes explicitly (e.g., Beissinger
2002), but such detailed analysis is beyond the scope of the current study. As
will become clear below, collecting and evaluating global, comparative data on
political and economic horizontal inequalities is a major challenge in its own
right. Thus, we make the leap from the explanans to the explanandum in two
36 Inequality, Grievances, and Civil War

figure 3.1. The causal path from inequalities to onset through grievances.

steps. The first step explains how structural asymmetries relating to political
and economic inequalities generate grievances. The second step shows how
grievances may trigger violent conflict, thus bridging the explanatory gap.
Building on Coleman’s (1990) classical meta-theoretical approach, Figure
3.1 outlines our research strategy graphically. In order to avoid the fallacy
of holistic causation, the approach calls for the specification of explicit causal
mechanisms at a lower level of aggregation (Hedström and Swedberg 1998).1 In
the present context, this entails the articulation of grievance-based mechanisms
at the level of (individual or group) agents, as opposed to focusing merely on
the empirical link between inequalities and collective violence at the macro
level.2 That is, it entails moving from “factors” to “actors.”
As argued in the previous chapter, both the assumptions of ubiquity-
of-grievances and irrelevance-of-grievances should be exposed to systematic
empirical tests rather than being treated as unquestioned postulates. However,
it is not enough to show that there is a statistical macro link between inequalities
and violence, and to demonstrate that our proposed mechanisms are plausible.
We also have to establish that alternative causal pathways connecting inequal-
ities and violence are less plausible, and that endogeneity and reverse causation
do not undermine our reasoning. We will return to these issues at the end of
the present chapter, and in connection with the empirical analyses in Parts II
and III.
For now the task at hand is to develop the theoretical argument that explains
through what mechanisms horizontal inequalities generate civil wars. In the
following two sections, we will do so by first taking the step from inequalities
to grievances, before connecting the latter to political violence.

1 Our approach is inspired by, and has much in common with, “analytical sociology” (Hedström
and Bearman 2009).
2 We adopt this disaggregation strategy that goes from “black-box regularity to a mechanism”
without making any strong, reductionist assumptions about causation being reducible to a search
for “microfoundations” (see Elster 1998).
From Horizontal Inequality to Civil War via Grievances 37

figure 3.2. The causal path from horizontal inequalities to grievances.

From Inequalities to Grievances

For inequalities to be transformed into grievances, they have to be politicized.
At this point we are concerned with politicization rather than with mass mobi-
lization. We will deal with mobilization processes in the next section that close
the gap between grievances and political violence. Throughout the chapter,
we focus primarily on horizontal inequalities. Yet, more generally speaking,
the reasoning underlying many of our causal mechanisms applies to verti-
cal inequalities that are associated with political representation and economic
interests. In Chapter 7, we will return to the link between vertical inequalities
and civil war.
It is important to realize that our argument is probabilistic rather than
deterministic. Not all inequalities produce grievances, and not all grievances
trigger violence. Therefore, in the following, we have to keep in mind that
“dead ends” of the conflict process may at any time interrupt the causal chain.
In this sense, the stages of the pathway could be seen as necessary conditions
rather than as contributing causes.
What does it take for skewed distributions of power and wealth to produce
politically relevant grievances? As illustrated in Figure 3.2, we identify several
important steps in the general causal pathway. Politicization requires a basis
of identification, an explicit comparison that reveals some injustice typically
through framing and attribution of blame, which in turn can be used to make
explicit political claims. The remainder of this section discusses each of these
steps in turn.

Group Identification
In most cases, the presence of horizontal inequalities presupposes the existence
of well-defined groups, which is not a trivial precondition (Stewart 2000; Gurr
2000b). Objective group-level inequalities will hardly develop into grievances
unless the groups themselves are perceived by, and identified with, a consider-
able part of the affected population, although articulation of grievances may
help crystallize identity. This requires at least some awareness of the relevant
38 Inequality, Grievances, and Civil War

social categories, but this awareness may not be very well developed in areas
that have not benefited from mass schooling or a minimum level of develop-
ment (Darden forthcoming). Furthermore, the salience of identities depends
critically on the availability of relatively sharp boundary markers. The difficul-
ties of constructing cohesive group identities based on socioeconomic classes
confirm this point. Thus it is not surprising that Marxists refer to the notion of
“false consciousness” in order to explain away the failures of groups to form
around class interests (Brass 1991, p. 261).
Yet, this observation does not imply that the formation of ethnonationalist,
as opposed to class-based, identities is a smooth and spontaneous process.
As we have seen in Chapter 2, some theorists have gone so far as to reject
the usefulness of collective identities and ethnic groups as analytical concepts
altogether (e.g., Brubaker 2004). While it is undoubtedly true that modern
politics is to a large degree group-based, and social life depends critically on
social categories (Simmel [1908] 1971; Gellner 1964; Hogg and Abrams 1988),
we argue that the extent to which cohesive groups can actually be said to exist is
ultimately an empirical matter that depends on the specific situation. To be sure,
premodern, clan-based societies, such as Somalia, represent a relatively poor
fit with our group-level perspective.3 Nor do we rule out a connection between
inequality and conflict in nonethnic cases, for example, based on classes and
ideology.4 Since our goal is to evaluate the conflict-inducting effect of horizontal
inequalities, however, we join Horowitz (1985), Gurr (1993a; 2000b), and
others in adopting a self-consciously group-based framework, while restricting
our substantive focus to groups defined through ethnic categorization rather
than through other cleavages.
Categorization can happen spontaneously through bottom-up processes
involving cultural innovation among intellectuals, scholars, writers, and
artists. Students of nationalism have produced volumes detailing this process
of “cultural engineering” (see, e.g., Hroch 1985). However, the formation of
ethnonationalist identities usually involves at least some measure of organi-
zational coordination, typically through the state. If imposed on previously
ethnically “unconscious” populations, classificatory schemes, such as censuses
and ethnically explicit laws regulating the use of languages and the practice
of religion, help bring about identity formation (see Anderson 1991; Brubaker
1996; Martin 2001). Indeed, the more exclusionary and discriminatory such
state policies are, the more directly they contribute to further cementing group
identities.5

3 In fact, the Ethnic Power Relations data set, on which we rely in our empirical analysis, lists
Somalia as a country in which ethnicity is irrelevant due to the dominance of clans, which we
exclude from our definition of ethnic groups (see Chapter 4).
4 See Buhaug, Cederman, and Gleditsch (forthcoming) for an analysis of geographic inequality
and the location of conflict onset; see also Chapter 7.
5 In Chapter 2, we justified the assumption of reasonably stable group identities in conflict pro-
cesses by referring to theories of reactive identity formation (e.g., Simmel 1955; Coser 1964).
From Horizontal Inequality to Civil War via Grievances 39

Group Comparison
The further path to grievances requires a relational setting within which mem-
bers of groups compare their own group’s status and wealth to that of other
groups. Social psychologists, especially those propounding social identity the-
ory (SIT) (Tajfel and Turner 1979), have long argued that human beings tend
to form groups based on trivial markers, which may serve as opportunistic
justifications for conflict. Tajfel and his colleagues conducted a famous series
of “minimal group experiments,” which indicates that group loyalty quickly
emerges based on arbitrary and imposed distinctions. In particular, Horowitz
(1985) uses these findings as the basis of his own group-level theory of ethnic
conflict. Tajfel’s experimental perspective served the purpose of proving an
important fundamental point about identification in social relations. Used as
a conflict theory, however, social identity theory appears oddly inattentive to
substantive differences relating to political or economic inequalities. In fact, it
can even be viewed as a vindication of the ubiquity-of-grievance assumption:
if trivial group differences are likely to escalate into competition and conflict,
grievances must be truly ubiquitous and therefore could hardly explain the
outbreak of conflict. Unsurprisingly, Horowitz (1985) subsequently proceeds
to consider the impact of actual, rather than merely imagined, inequality on
conflict, and does find a positive relationship.6
While constructed and misperceived comparisons may cause conflict, we
posit that actual and objectively measurable differences in group status or
economic development will make grievances more likely. Comparison, then,
proceeds primarily based on real inequalities along ethnic lines, although per-
ceived differences may also play a role in some cases. Rather than conforming
with the principles of SIT, this logic corresponds to realistic group conflict
theory, as developed by Sherif and Sherif (1953) and LeVine and Campbell
(1972). This classical, but often neglected, social-psychological approach to
group relations assumes that conflicting goals are central to conflict behavior.
In particular, under conditions of scarce resources, out-groups appear more
threatening. In the words of Williams (2003, p. 135): “Where strongly bounded
ethnies act as rival collectivities (rather than mere social categories or unorga-
nized aggregates of individuals), their relative positions become salient.” Under
such conditions, comparisons are hard to avoid: “Different aspects of group
position – economic advantages, political authority, cultural eminence, gener-
alized prestige – become central issues under differing conditions” (Williams
2003, p. 135). Ultimately, the very tangible threat of out-groups translates into

More specifically, Weingast (1998) applies a game-theoretic perspective to the conflict in the
former Yugoslavia in order to illustrate that even a hint of repressiveness suffices to trigger
interethnic polarization.
6 See also Hale’s (2008) critique of social identity theory. Yet, our and Hale’s criticisms should
not be interpreted as a wholesale rejection of social identity theory so much as a restriction of
its substantive scope.
40 Inequality, Grievances, and Civil War

heightened intragroup cohesion coupled with ethnocentrism (Campbell 1965,

p. 291).7

Evaluation of Injustice
The mere presence of real or imagined inequality clearly does not suffice to
generate grievances. After all, there are countless examples featuring social
actors who accept considerable asymmetries in the distribution of political
power and economic wealth without becoming frustrated. Failure to evaluate
the distributional situation with respect to some norms of equality typically
leads them to accept the status quo, or even to ignore the issue of distribution.
In the words of Williams (2003, p. 131):
Grievances are not merely expressions of deprivation and dissatisfaction. People can
be deprived, disappointed, frustrated, or dissatisfied without feeling that they have
been unjustly or unfairly treated – their unsatisfactory outcome may be “just the way
things are” or the result of divine judgment, or a consequence of personal ineptitude.
In contrast, a real grievance, regarded as the basis for complaint or redress, rests upon
the claim that an injustice has been inflicted upon undeserving victims. Grievances
are normative protests, claiming violations of rights or rules. Those who are intensely
aggrieved may use the language of moral outrage.

In concrete terms, this means that someone has to identify what is unfair about
the status quo. In most cases, this task falls on the group’s leadership: “Without
elite entry . . . injustices and inequalities may be accepted, cultural decline or
assimilation may occur, and grievances may be expressed in isolated, anomic,
or sporadic forms of conflict and disorder” (Brass 1991, p. 293).
From this reasoning it is clear that grievances may vary as much with changes
in the normative framework as with the actual level of inequality. Breaches of
norms may only be felt with growing relevance of a specific notion of justice.
Such evaluations are likely to change over time (Parekh 2008). In general, the
macro-historical process of democratization has prompted increasing pressures
in favor of inclusive governance and political equality, at least at the level of
individuals. Also to a large degree the product of the great American and French
Revolutions in the late nineteenth century, nationalism serves as the most
important source of normative evaluation in group relations. In particular, the
principle of national self-determination stipulates that each nation is entitled
to govern its own affairs without interference. As we have seen, wherever this
principle is, or is seen to be, violated, nationalist grievances are likely to emerge.
In the empirical chapters below, we will derive a set of specific hypotheses
that we evaluate systematically. Most obviously, if the incumbent elite of a state

7 Indeed, if real differences do exist in a sufficient number of cases, they should leave a trace in our
quantitative findings. However, we subscribe to the principles of realistic group conflict theory
without, for that matter, assuming that inequalities always have to be objectively given in order
to have an impact on conflict onset (see Herrera 2005). Hence, we agree with Brass’ (1991)
critique of attempts to reduce ethnicity to a matter of interest groups.
From Horizontal Inequality to Civil War via Grievances 41

denies an ethnonationalist group access to central power, grievances are more

likely to form (see Chapter 4). However, not even power sharing guarantees
full satisfaction with the status quo. Such agreements may fall well short of the
maximalist expectations of the group’s nationalist ideology:

It is unlikely that grievances can simply be imaginary or invented, but the minimum level
of discontent necessary for mobilization to start can be surprisingly low. Thus ethnic
mobilization may occur when there is no exclusion from political participation – even
when members have full access to political parties and state agencies, but feel deprived
or frustrated because their policy preferences are not being granted (Williams 2003,
p. 150).

Nevertheless, we postulate that, on average, grievances will be experienced

roughly in proportion to the degree of violation. For example, the more harshly
the government treats the group in question, the more frustrated its members
will become. Furthermore, sociological theories of emotions suggest that neg-
ative emotions are especially likely to be aroused following loss of power and
prestige (Kemper 1978). When the subjects blame others for their downgrad-
ing, anger and resentment increase the readiness to fight in order to change
the situation (Turner and Stets 2005). Thus, we expect that the reversals of
a group’s status, especially political exclusion from executive influence after a
long period of dominance, will provoke especially marked discontent (Brass
1991; Petersen 2002). Again, we turn to Williams (2003, p. 138) for a cogent
summary of this logic:

A predictable reaction of resentment appears in those ethnies whose members feel that
the collectivity as a whole is being unjustly subordinated in a ranking order. Especially
sharp resentment is likely when there is an actual reversal in ranking, so that when
a previously subordinate ethny acquires a dominant position we then see the familiar
resentment of displaced groupings – those pushed aside, overturned, ignored, and treated
with disdain by those earlier regarded as inferior.

In Chapter 4, we will test this mechanism explicitly by evaluating Hypothesis

H4.2.

Framing and Attribution of Blame

Even the realization that the current order is unfair may not suffice to bring
forth grievances. Indeed, our definition of grievances as a reaction to unfair
treatment raises the question of who is responsible for the mistreatment. In
the context of civil war, it is by definition the incumbent government that
serves as target. Ultimately the representatives of the state have to be seen as
sponsors and protectors of the unjust order; otherwise, grievances will not be
focused and violence is thus less likely to erupt. Reminding us that “[p]olitical
grievances are not like anomie, but directed against a target,” Goodwin (1997,
pp. 17–18) asserts that
42 Inequality, Grievances, and Civil War

grievances may become “politicized” (that is, framed as resolvable only at the level
of the state), and thereby a basis for specifically revolutionary collective action, only
when the state sponsors or protects economic and social conditions that are viewed as
grievous.

How do activists manage to pin the blame on the state? The classical socio-
logical literature suggests that the notion of injustice is central to such efforts
(Moore 1978). More recently, social movement researchers have introduced
the notion of “injustice frames” that depict the members of a social movement
as victims of societal injustice (see, especially, Gamson 1992, p. 68). According
to Mason (2009, p. 80), framing works as follows:

To activate local social networks in support of a national movement, dissident leaders

employ framing processes to persuade the member of local social networks to join a
national movement. Framing involves identifying injustices that afflict the community
and attributing them to the state or some other entity that is the intended target of
the social movement. . . . Effective framing involves using traditional symbols to attract
nonelites to a new set of values and beliefs about the state that will make them more
willing to participate in a movement that challenges the state’s sovereignty.

More generally, Benford and Snow (2000) argue that injustice frames constitute
an important class of “collective action frames.” Such constructs help social
movements arrive at “a shared understanding of some problematic condition
or situation they define as in need of change, make attributions regarding who
or what is to blame, articulate an alternative set of arrangements, and urge
others to act in concert to affect change” (p. 615). To be effective, collec-
tive action frames need to resonate with their target audiences. According to
Benford and Snow (2000, pp. 619–21), their effectiveness hinges on the inter-
nal consistency of the frame itself with the goals of the social movement, its
empirical credibility, and the political entrepreneurs’ own persuasiveness in
terms of status, moral authority, and knowledge.
Yet, while constituting a good analytical starting point, this somewhat
abstract and mostly cognitive account of framing needs to be amended since it
fails to do justice to the emotional aspects of action frames (Goodwin, Jasper,
and Polletta 2001a; 2001b). Rather than being a matter of formal ideologi-
cal consistency or general factual persuasiveness, framing in this deeper sense
requires that the call to action relates directly to individual members’ own
life circumstances and aspirations. Successful frame-making typically relies on
the construction of “political cultures of opposition” that connect popular
idioms and ideological constructs to people’s everyday lives in an emotionally
powerful way (Reed and Foran 2002). For example, Reed and Foran explain
how the Nicaraguan Revolution drew on a combination of liberation the-
ology and “Sandinismo” as well as “more diffuse idioms and sentiments of
nationalism, social justice, human dignity, and democracy” (p. 339). Besides
being interpreted through cultural lenses, the revolutionary message needs to
From Horizontal Inequality to Civil War via Grievances 43

be “packaged” in a digestible and engaging format that can be easily dissem-

inated. As Selbin (2010) explains, revolutionary and rebellions’ framing does
not merely depend on their structural appropriateness but also on the narrative
tools invoked by the mobilizers (see also Polletta 1998). Thus, it is crucial to
recognize the central role played by “myth and memory of revolution and of
the power of mimesis for the mobilization and sustenance of revolutionary
activity” (Selbin 2010, p. 4).
During the last couple of centuries, nationalism constitutes the perhaps most
prominent master template for rebellious scripts. Despite considerable diversity
as regards the specific contents of nationalist ideologies, Anderson (1991, p. 4)
explains how nationalism is “modular” in these sense that it is “capable of being
transplanted, with various degrees of self-consciousness, to a great variety of
social terrains, to merge and to be merged with a correspondingly wide variety
of political and ideological constellations.” Although the recent research on
emotions as the key to successful framing within social movements remains
somewhat inchoate and scattered, it promises to overcome the gap between
structural renderings of nationalism, as those proposed by Gellner (1983), and
those that focus on the power of ideology and symbols (e.g., Smith 2009).
As pointed out by Petersen (2002, pp. 57–8), considerable emotional power
is implicit in Gellner’s famous account of the “Ruritarians’” liberation from
the “Megalomanians.” In sum, these arguments articulate mechanisms that
connect past exclusion and discrimination to a narrative of group identity that
highlights the virtuous struggle against state oppression and unjust treatment.8
The power of framing becomes especially clear in its absence. Indeed, the
status quo can easily be maintained by the incumbent elite if no coherent protest
frame emerges. In this case the category never becomes politically relevant and
the incumbent maintains cognitive dominance through Gramscian hegemony
(Laitin 1986). Whether the current wealth distribution is perceived as being
just or not thus critically depends on how expectations are managed through
framing. In case of political exclusion and discrimination, it is usually obvious
that the incumbent state elite is the guilty party. In case of economic horizontal
inequalities, more elaborate arguments concerning the government’s blame for
underdevelopment may have to be advanced (see Chapter 5). In particular,
beliefs about unfair exploitation in terms of taxation and the incumbent’s
extraction of the group’s “own” natural resources may lead to a politicization
of inequality (e.g., Aspinall 2007; see also Stewart 2008a). Weiner’s (1978)

8 In the social movement literature, the framing of mobilization processes belongs to one of the
most intensively studied topics. Partly because of widespread grievance-skepticism as described
in Chapter 2, however, analysis of collective action frames has been relatively rare in the conflict
literature, although there are exceptions. For example, Kaufman (2011) offers a detailed frame-
based study of the Mindanao rebellion in the Philippines dating back to the 1970s. Combining
ideas from his own theory on symbolic politics, Kaufman draws heavily on the sociological
literature on frames in conflict processes. For more general explanations of ethnonationalist
violence highlighting the function of myths and symbols, see Smith (2009) and Kaufman (2001).
44 Inequality, Grievances, and Civil War

figure 3.3. The causal path from grievances to civil war onset.

notion of “sons of the soil” is a special case of this important grievance process,
as will be argued in Chapter 8.

From Grievances to Violence

Once horizontal inequalities have been politicized by intellectuals, promi-
nent dissidents, and political entrepreneurs, the question arises of how such
grievances can spread within the concerned populations to such an extent that
they trigger large-scale internal conflict. Our argument hinges on the interac-
tion between the challenger’s mobilization and the response of the state (see
Figure 3.3).9

Mobilization
Clearly, elite-level grievances can be articulated, and claims advanced, without
their finding any resonance among the masses. This observation constitutes
the core of the irrelevance-of-grievances assumption that was introduced in
Chapter 2. Much has been said about collective action, and individual fears
of governmental repression may certainly block any further increase in par-
ticipation. Olson’s (1965) theory of collective action offers the most extreme
formulation of this argument. Since overcoming inequalities and establishing
a new social and political order amount to public goods, free riding can be
expected to be widespread, especially where state-led repression increases the
risks associated with protest.
Rationalist collective action theory puts the bar very high, but it may still
be possible to find successful mobilization strategies that entrepreneurs can
resort to without violating the individualist axioms originally stipulated by
Olson. For example, a number of conflict scholars have proposed that offering

9 Obviously, the sequence of mobilization followed by state reaction is stylized. In many cases,
these steps are taken repeatedly and more or less simultaneously.
From Horizontal Inequality to Civil War via Grievances 45

material selective incentives helps overcoming the dilemma (e.g., Popkin 1979;
Lichbach 1995; Weinstein 2007).
Yet, the strictly individualist perspective associated with rationalist theoriz-
ing fails to take the social context into account (Hechter and Okamoto 2001).
Referring explicitly to the role played by social networks and institutions, social
movement theory goes a long way toward explaining how social movements
with political goals manage to overcome collective action hurdles and turn
abstract claims into actionable programs that draw on considerable popula-
tion support (whether direct or indirect).10 Preexisting social networks provide
a reservoir of trust that makes free riding harder and more risky because of
in-group punishment (Mason 2009). In the social movement literature, such
contextual reinforcement and social monitoring add up to something that has
been called “mobilizing structures.” According to Tarrow (1994, p. 22),
the mobilization of preexisting social networks lowers the social transaction costs of
mounting demonstrations, and holds participants together even after the enthusiasm
of the peak of confrontation is over. In human terms, this is what makes possible the
transformation of episodic collective action into social movements.

In addition to such structures, identity formation constitutes another impor-

tant resource that helps mobilization processes advance. Drawing on Harrison
White’s important insight, Tilly (1978) captures this logic with the notions of
“catness” and “netness.” This conceptualization implies that the effectiveness
of mobilization efforts equals the product of solidarity of categorical identifica-
tion (i.e., catness) and the density of social networks (i.e., netness). However,
Tilly does not further develop the source of categorically induced solidarity.
Those who have done so typically refer to the reduced transaction costs and
the possibility of normative enforcement among members of the identity group
in question (Goldstone 2001, p. 164; see also Opp 2009, ch. 7).11
The cognitive and structural focus of the social movement literature makes
valuable contributions to our understanding of mobilization processes that are
able to transcend the collective-action dilemma, but most of this research fails
to take into account the direct impact of grievances. As has been argued in the
context of framing processes, these explanations take on a somewhat blood-
less and clinical flavor that seems worlds apart from the emotional heat of
real-world conflict processes (Horowitz 1985). Whether based on relational
or identity-based arguments, the reasoning surveyed so far attempts to steer
clear from emotional effects by privileging strictly instrumental calculation and

10 See also Kalyvas and Kocher (2007), who argue that collective action dilemmas do not neces-
sarily apply in civil war settings, especially where noncombatants run a higher risk of getting
killed than the rebels themselves. Although insightful, the scope of this argument is limited to
particularly violent cases characterized by massive violations of human rights.
11 For example, Simpson and Macy (2004) present a model that demonstrates the effect of collec-
tive identification on costly collective action that is validated through experiments. For a more
general argument based on agent-based modeling, see Cederman (2002).
46 Inequality, Grievances, and Civil War

“cold” cognitive processes (Emirbayer and Goldberg 2005). In fact, this sani-
tized logic captures the dominant view of the literature on social movements,
whether inspired by economics or sociology:

Emotions have disappeared from models of protest. When crowds and collective behav-
ior, not social movements and collective action, were the lens for studying protest,
emotions were central. Frustration, anger, alienation, and anomie were not merely an
incidental characteristic but the motivation and motivation of protest. Such images were
displaced 30 years ago by metaphors of rational economic calculators and purposive
formal organizations, for whom social movements were just one more means of pur-
suing desired ends (Jasper 1998, p. 398; see also Goodwin, Jasper, and Polletta 2001a;
2001b).

Without belittling the conceptual coherence of mainstream theorizing, there

are good reasons to question the absence of affective mechanisms in conven-
tional explanations of collective protest action:

Mobilization theory based on the utilitarian rational choice model leaves one some-
what unsatisfied, because strong passions, group consciousness, ideological appeal, and
appeals to solidarity in group conflict are relegated to secondary place, if they are taken
into account at all (Oberschall 1993, p. 57).

To the extent emotions enter conventional explanations, they do so indirectly.

Far from being emotionally neutral, Olson’s (1965) classical rendering of the
collective action dilemma at least implicitly presupposes that potential rebels
will fear governmental retaliation (see, e.g., de Figuerido and Weingast 1999).
Yet, fear is far from the only emotion that influences mobilization. Fear-
based explanations need to be balanced against accounts that involve other
emotions, such as anger and resentment (Petersen 2002; 2011). Indeed, the
fear-resentment balance may tip over into grievance-driven collective action if
grievances are strongly felt and shared, although overwhelming deterrence on
the part of the incumbent elite may dampen the impact of resentment. Examples
from around the world show that high-risk social movements have developed
strategies of “emotion management” that serve to counterbalance the paralyz-
ing effect of fear, for instance by evoking liberating anger and castigating free
riders through shaming (Goodwin and Pfaff 2001).
Emotions provide the missing piece in the explanatory jigsaw. Failure to
accord them an active role in mobilization processes makes it difficult to
account for the emergence of solidarity, and how action is triggered by “moral
shocks” (Jasper 1998, p. 409; although see Hechter 1987). Indeed, specific
emotional patterns are associated with action tendencies that help explain
why mobilization may spill over into violence (Mercer 2005; Petersen 2002).
Even though organizational and cognitive factors are central to mobilization,
it would be a mistake to overlook the contribution of emotionally charged
grievances. Indeed, “emotional ties and investments are a potential source
From Horizontal Inequality to Civil War via Grievances 47

of power in their own right, alongside social-structural sources of power”

(Emirbayer and Goldberg 2005, p. 507).
In a seminal study of the civil war in El Salvador, Wood (2003) offers
especially convincing direct evidence showing how grievances can serve as the
main motivation for rebels despite their exposure to extremely high risks. Her
account describes how rural workers, the campesinos, defied a lethal campaign
of violence and disappearances orchestrated by a repressive state dominated by
rich, reactionary landowners. Because this study focuses on ideological, rather
than ethnically, defined solidarity, it constitutes a hard test for this collective-
action mechanism under extremely risky conditions. Relying on anthropologi-
cal methods, including a large number of interviews with active rebels, Wood
(2003, p. 18) found a very different set of motivations from Olson’s free riders:
“My insurgent informants made it clear to me that moral commitments and
emotional engagements were principal reasons for their insurgent collective
action during the civil war.”
In contrast to standard approaches to collective action in civil war, Wood
(2003) highlights two important aspects of her approach to the “pleasure of
agency.” First, she deviates from methodological individualism by assuming
that affective mobilization is profoundly “other-regarding.” In her view, the
“reasons for which participants acted referred irreducibly to the well-being of
others as well as oneself, and to processes, not just outcomes” (p. 19).12 This
is an extremely important finding as it demonstrates the limitations associated
not only with rationalist theorizing, but also with individualist renderings of
grievance-based mechanisms, such as Gurr’s relative deprivation theory (Cook
and Hegtvedt 1983; see also Chapter 2).
In this very respect, social psychology helps us appreciate the extent to which
individual actors identify themselves with the group and predicate their val-
ues and motivations on the well-being of the group rather than considering
exclusively their own self-interest. Of course, this process of “deindividual-
ization” may not be perfect and does not require that all members become
entirely exchangeable. Yet, a series of experiments has established that human
decision making depends as much on the “group in the individual” as on the
“individual in the group” (Hogg and Abrams 1988). Not unlike “sociotropic”
voting (Kinder and Kiewiet 1979), according to which voters consider macroe-
conomic indicators rather than their own pocketbook, group-level solidarity
often prevails over strictly egocentric reasoning in conflict processes (see also
Hechter 1978).
Yet, it would be a mistake to reduce such collectivizing mechanisms
to a purely cognitive phenomenon. As we have argued, the perception of
injustice generates grievances that serve as a formidable tool of recruit-
ment. In fact, Tilly’s notion of catness is entirely compatible with collective

12 This interpretation has much in common with Weber’s “value rationality” as opposed to
“instrumental rationality” (see Varshney 2003).
48 Inequality, Grievances, and Civil War

solidarity prompted by deeply felt resentment with the status quo. According to
Oberschall (1993, p. 25), the microlevel assumptions that underpin this the-
oretical idea constitute a “multiplier effect” that creates collective grievances
out of individual ones:

Members not only take offense at injuries and injustices they personally suffer, but will
react to the experiences of their peers out of fellow feeling and solidarity. . . . A single
injury is “multiplied” by the amount of group cohesion and interpersonal bonds to
become a collective grievance.13

This formulation supports Wood’s (2003) point that altruistic motives can
transcend collective action dilemmas even in the face of extreme peril. Such
emotional reactions become even more potent in the course of conflict pro-
cesses that further escalate the level of injustice. Wood’s second insight shows
that, far from being constant, the level of grievances varies over time as a
function of the conflict processes involving challengers and incumbents. Thus,
“political culture . . . was not fixed but evolved in response to the experiences
of the conflict itself, namely, previous rebellious actions, repression, and the
ongoing interpretation of events by the participants themselves” (p. 19). Most
importantly, Wood detects clear signs of reactive mobilization:

As government violence deepened, some campesinos supported the armed insurgency.

They did so as an act of defiance of long-resented authorities and a repudiation of
perceived injustices. . . . Participation per se expressed outrage and defiance, its force
was not negated by the fact that victory was unlikely and in any case was not contingent
on one’s participation. Through rebelling, insurgent campesinos asserted, and thereby
constituted in their own eyes, their dignity in the face of condescension, repression, and
indifference (p. 18).

This observation leads us directly to the interactive logic of claims and coun-
terclaims issued by challengers and incumbents. Given that grievances are
inherently relational, violence-inducing mobilization processes should also be
understood in a similarly relational context.

Rebel Claims and State Repression

A central task facing the rebel leaders is to turn the “blame game” into action-
able claims. Targeting the incumbent government, such claims serve to correct
perceived injustices. The challenger’s political demands range from reforms of
the current regime (e.g., more cabinet seats, transfers of public goods, or a larger
measure of autonomy) to the very toppling of the government or secession from
the state. In order to effect change, movements draw on collective-action reper-
toires that include peaceful protest, boycotts and strikes, and symbolic acts, but
may also spill over into violence (McAdam, Tarrow, and Tilly 2001). As we

13 Runciman (1966) introduced the notion of “fraternal deprivation” in order to overcome the
individualist character of relative deprivation theory (see also Markowsky 1985).
From Horizontal Inequality to Civil War via Grievances 49

will see, claim making may evolve over time, depending on the response of the
government, in an explicitly strategic fashion.
How do incumbent state elites react to the threat of antigovernmental mobi-
lization? Obviously, this depends directly on the nature of the claim. Demo-
cratic regimes are more likely than authoritarian ones to engage in negotia-
tions concerning the rebels’ claims as long as they entail peaceful change of
the current regime. Yet, there is large variation in the willingness to accommo-
date group-level demands, especially if they are seen to undermine the central
authority and territorial integrity of the state. Leaders of liberal democracies are
typically reluctant to grant groups privileges because they believe that the indi-
vidual right to vote guarantees all types of peaceful political change. Indeed,
few states, democratic or not, would voluntarily accept to be dismembered
through secession, although there are examples of peaceful “velvet divorces”
(e.g., the breakup of the Swedish-Norwegian Union in 1905 and Czechoslo-
vakia at the end of the Cold War). In contrast, authoritarian states, especially
of the totalitarian kind, generally refuse to accept even the most modest pro-
posals for political change while exposing peaceful protesters to the full force
of their repressive capabilities.
Under what conditions is the outbreak of internal violent conflict most
likely? A distinguished, state-centric tradition of theories on revolution insists
that the state’s reaction to challengers’ claims holds the key to the answer
of this question. While earlier instances of such theorizing highlight struc-
tural conditions and international opportunities, sometimes explicitly at the
expense of grievances (e.g., Skocpol 1979), more recent theories of revolution
factor in agency and the state’s reactions to rebel movements’ claims (e.g.,
Goldstone 1991). Here we follow Goodwin (2001), who proposes a clearly
argued example of such a theoretical perspective. On this view, a specific set of
policies is likely to provoke revolutionary, and, by extension, violent reactions.
These relate to the state’s approach to economic/social and political inequali-
ties respectively, as well as to the government’s possible resort to indiscriminate
violence.14
The approach to economic and social inequalities is especially relevant to
our attempt to make the theoretical leap from collective grievances to the
outbreak of armed conflict. State elites that are perceived as the defenders of
highly unequal economic and social arrangements, while systematically block-
ing reforms, run the risk of inviting more radical demands, thus disqualify-
ing themselves as neutral agents working for social justice. Goodwin (1997)
explains that, under such circumstances, “grievances may become ‘politicized’
(that is, framed as resolvable only at the level of the state), and thereby a basis
for specifically revolutionary collective action, only when the state sponsors or

14 Gurr (2000b) highlights two other properties of policies that relate to weak state reach and
corrupt rule. The link to violence is plausible in these cases as well, but these factors are less
directly linked to the grievance-related mechanisms discussed in this book.
50 Inequality, Grievances, and Civil War

protects economic and social conditions that are viewed as grievous” (pp. 17–
18). Thus it follows that “states that regulate or abolish perceived economic
and social injustices are less likely to become the target of political demands
(revolutionary or otherwise) than those that are seen to cause or reproduce
such injustices” (p. 17).
Second, sustained exclusion of mobilized groups from state power will also
make violent outcomes much more probable, because the state’s refusal to grant
the excluded groups a share of power blocks peaceful avenues of political
change. As it fits our own approach to exclusion, Goodwin’s (1997, p. 18)
argument deserves to be quoted at some length:

Even if aggrieved groups direct their claims at the state, they are unlikely to seek its
overthrow (or radical reorganization) if they manage to attain some significant share –
or believe they can attain such a share – of state power and influence. Indeed, even
if such groups view their political influence as unfairly limited, their access to state
resources or inclusion in policy-making deliberations – unless palpably cosmetic – will
likely prevent any radicalization of their strategic repertoire or guiding ideology. In fact,
the political “incorporation” of mobilized groups – including putatively revolutionary
proletariats – has typically served to deradicalize them.

Conversely, chronic exclusion and outright discrimination are likely to incline

the movements to opt for more radical responses that may include violent
strategies.
Third, states that resort to indiscriminate campaigns of repression targeting
their challengers leave the movements with little choice but to respond with vio-
lent means, thus triggering revolutionary civil war. Mirroring Wood’s (2003)
observation about the endogeneity of grievances, Goodwin (1997) stresses that
state reactions of this uncompromising nature leave the rebels with “no way
out”:

Like political exclusion, indiscriminate state violence against mobilized groups and
opposition figures is likely to reinforce the plausibility, justifiability, and (hence) diffu-
sion of the idea that the state needs to be violently “smashed” and radically reorganized
(p. 19).

Groups targeted in this way will have little choice but to arm themselves to pro-
tect themselves against the maltreatment of the state. Indiscriminate violence
is clearly counterproductive since it tends to deepen grievances and radicalize
the movement by undermining more moderate leaders. In his thorough analy-
sis of the effect of indiscriminate violence, Kalyvas (2006) demonstrates that,
although overwhelming force may crush the resistance, such ham-fisted poli-
cies of brute force typically deepen the level of grievances through emotional
responses, even in those cases where they were initially held by only a small
minority of a group’s members.
In such situations, radical and revolutionary ideologies thrive, includ-
ing exclusive and militarized ethnonationalism, because “a society in which
From Horizontal Inequality to Civil War via Grievances 51

figure 3.4. Grievance-based mechanism as spurious explanation.

aggrieved people are routinely denied an opportunity to redress perceived

injustices, and even murdered on the mere suspicion of political disloyalty,
is unlikely to be viewed as requiring a few minor reforms” (Goodwin 1997,
p. 19). It goes without saying that regimes of this type are exposed to grave
risks of civil war onset.

Alternative Explanations
The previous sections have attempted to render plausible the role of grievances
in conflict processes. Yet, in the absence of direct tests of aggrieved people’s
attitudes and emotions, the proposed causal pathway that we have so far
sketched can be criticized on several grounds.
The most obvious angle of attack builds directly on the ubiquity-of-
grievances argument that we introduced in Chapter 2: one could undermine
the macro link that connects group-level inequality with conflict by showing
that there are other ways to get from inequality to conflict than via grievances
(see Figure 3.4). This has indeed been the strategy of the quantitative country-
level studies that argue that grievances lose their explanatory power as soon as
self-interested and opportunistic factors are accounted for.
Methodologically this approach introduces proxies for both sets of fac-
tors and then proceeds by showing that those indicators that are (ostensibly)
associated with grievances, notably indices of ethnic diversity and income dis-
persion, become insignificant once the other factors are introduced into the
model (again, see Fearon and Laitin 2003; Collier and Hoeffler 2004). Indeed,
if it can be shown that the effect of either political or economic inequality
disappears once opportunity-based explanations are controlled for, then our
grievance account could well be spurious. In the empirical chapters, especially
52 Inequality, Grievances, and Civil War

figure 3.5. Grievances bypassed through reverse causation.

in Chapter 7, we will introduce controls of this type in order to show that the
effect of inequality remains robust.
It is not difficult to think of possible alternative stories that bypass grievances
in this way. Consider the case where poverty in specific regions drives outcomes
along the lines of Collier and Hoeffler’s (2004) argument concerning oppor-
tunity costs. Regional underdevelopment could constitute fertile ground for
rebel recruiters who are motivated by greed and power, rather than by group-
based grievances. Likewise, Fearon and Laitin’s (2003) approach suggests that
peripheral location could drive conflict onset due to the state’s weakness far
away from its center of power. But at the same time, remoteness could also
disadvantage economic activity in the relevant area, with horizontal inequality
as a consequence. If geographic position causes both onset and inequality, it
could well be that the postulated grievance link connecting the two factors is
spurious. Fortunately, we are able to control for factors of this type in the
empirical analysis in the remainder of the book (see also the Online Appendix).
Assuming that the empirical link holds, however, our explanation could
still be invalid. It could be that endogeneity undermines our grievance-based
account through reverse causation (see Figure 3.5). It stands to reason that
previous violent challenges to the state’s power could incline governments to
exclude, or even discriminate, the groups in question, or that economic inequal-
ity could grow as a direct consequence of previous combat. Furthermore, it
cannot be ruled out that acts of violence, especially those targeting civilians
indiscriminately, will generate grievances irrespective of horizontal inequali-
ties. This creates a link from civil war to grievances, thus rendering the latter
causally ineffective as an explanation of violent conflict.
Finally, even if the macrocausal link points in the right direction, our
grievance-based account may fail because of a competing mechanism that
connects inequalities with violence, however, without taking the road via
grievances. It is notoriously difficult to rule out such an observationally equiv-
alent mechanism, but fortunately, it is also very difficult to think of a plausible
From Horizontal Inequality to Civil War via Grievances 53

candidate.15 However, while we will address the issue of spurious causation

and endogeneity explicitly throughout the book, we feel that it is not our duty
to come up with observationally equivalent mechanisms that are unrelated to
grievances. We leave this specific task to our critics and are looking forward to
challenging these accounts once they materialize. For now, it seems reasonable
to focus on the claim that grievances can be dismissed just because standard
indicators associated with such explanations are statistically insignificant.

15 For example, one could imagine that large inequality implies that a group is clearly weaker than
the state. Yet, the group’s weakness would also tilt the power balance in favor of the incumbent
government, thus making conflict less likely. In fact, this story leads to exactly the opposite
expectation.
part ii

EMPIRICAL ANALYSIS OF CIVIL WAR ONSET

T he second part of the book investigates the empirical link between hori-
zontal inequality and civil war onset. Chapters 4 and 5 analyze the effect
of political and economic inequalities on ethnonationalist conflict based on
samples of ethnic groups drawn from the EPR data set. Chapter 6 extends this
analysis to transnational relations, focusing on transborder ethnic kin. Building
on the previous findings, Chapter 7 revisits the country level, which allows us
to compare factors responsible for both ethnic and nonethnic civil wars.

55
4

Political Exclusion and Civil War

Do civil wars result from political exclusion? This chapter evaluates empir-
ically the mechanisms that trigger internal conflict as a function of political
horizontal inequality, as manifested through the exclusion of ethnic groups
from central executive state power.1 Rather than viewing ethnicity as a politi-
cally disconnected, demographic characteristic reflecting linguistic and cultural
diversity, as is often done in the contemporary civil war literature, we contex-
tualize ethnicity macro-historically as a political organization form embedded
in nationalism.
In Chapter 3, we discussed how nationalist politics leads to the generation
of political horizontal inequality. In the present chapter, we shift the atten-
tion from the origins of inequalities to their consequences for internal conflict.
Nationalism defines the macro setting within which much, possibly even most,
large-scale political violence unfolds in the modern world. Drawing on Gilpin’s
(1981) terminology, this type of transformation can be classified as “nation-
alist systems change.” Viewed as a special case of the broader category of
systems change that features fundamental transformations of the main actor
types defining politics, this type of macro-historical process prompted nations
to play the main role on the world stage along with states, roughly from the
time of the French Revolution (Cederman, Warren, and Sornette 2011; see also
Hall 1999). The importance of the nation can be found in the way that this
entity legitimizes the state. As opposed to the “descending” principle of terri-
torial sovereignty that justifies governance in personal, dynastic, and possibly
even divine terms, the French Revolution introduced an “ascending” logic that
defines the people, conceived of as the nation, as the locus of political legitimacy
(Calhoun 1997).

1 This first part of this chapter builds on Cederman, Wimmer, and Min (2010), which introduced
the first group-level results on political exclusion using the original version of the EPR data set.

57
58 Inequality, Grievances, and Civil War

This transformation of political legitimacy has profound repercussions for

interactions among social actors, including those that are associated with polit-
ical violence. As first realized by Clausewitz (1984), the most obvious conse-
quences for warfare pertain to the sheer scale of combat that became possible
thanks to the levée en masse (i.e., popular conscription). By tapping into the
demographic resources of the entire state rather than relying on mercenaries
or professional troops organized in standing armies, Napoleon and his succes-
sors eroded the distinction between warriors and the citizenry (Hintze 1975;
Howard 1976; Cederman, Warren, and Sornette 2011).
Yet, nationalist systems change does not only affect geopolitics through
changes of the internal workings of states. In addition, and at least as impor-
tantly, this type of transformation influences profoundly the very shape and
boundaries of the units in question. Wherever the nation happens to coincide
with the state, the boundary issue will cause little difficulty. However, where
national and state boundaries do not coincide, nationalist systems change will
trigger tensions that increase the likelihood of both internal and external con-
flicts. According to Gellner’s classical formula, “nationalism is primarily a
political principle, which holds that the political and national unit should be
congruent” (Gellner 1983, p. 1).
In some areas, such as preunification Germany and Italy, there were too
many states per nation. Competition over who was going to lead the unified
nation in many cases led to warfare (Miller 2007). More often, and more
importantly to our study, polities can be characterized by a state-to-nation
deficit (Gellner 1983, p. 2). This was the case in premodern and colonial mul-
tiethnic empires but still applies to most decolonized states (Connor 1972). To
the extent that an ethnically homogenous ruling elite refuses to share power
and wealth with other ethnic groups, grievances are likely to emerge as dom-
inated populations become conscious of their predicament through improved
education and communications.
Since this book is devoted primarily to internal conflict, we focus on the
first type of situation, which features a state inhabited by two or more ethnic
groups. An adapted version of Tilly’s (1978) classical polity model serves as an
appropriate starting point (see Figure 4.1). The original model features an actor
configuration that centers on the government, to which a number of political
groups compete for access. In this and the following two chapters, we assume
that the groups in question are defined in ethnic terms. Some of the groups,
which we will refer to as included or incumbent groups, enjoy privileged access
to power by virtue of being represented through its elite members within the
government’s executive organs, typically through ministerial representation
within the cabinet. Beyond this set of ethnic groups, there are also ethnic
groups that are systematically excluded from power, which we will here refer
to as excluded or marginalized ethnic groups.2 In the figure, we depict three

2 Cederman and Girardin (2007) use the terms Ethnic Group In Power (EGIP) and Marginalized
Ethnic Groups (MEG) for this distinction.
Political Exclusion and Civil War 59

figure 4.1. The polity model with included and excluded groups.

groups as having privileged access to central power, namely Groups 1–3. By

contrast, Groups 4 and 5 are excluded from power and Group 6 is irrelevant,
in the sense that it plays no explicit role in national politics.
By definition, excluded ethnic groups are exposed to a situation of alien
rule, since their state is ruled by “foreign” individuals who are members of
other ethnic groups. As we have seen, this type of political inequality consti-
tutes a direct violation of the principle of nationalism and self-determination.
Under such circumstances, representatives of the aggrieved groups typically feel
inclined to advance claims securing full, or at least partial, sovereignty. Full
sovereignty can be established by dislodging the incumbent government and
replacing it with one that is under the control of the challenger group. However,
power sharing would also ameliorate the situation by allocating a fair share
of cabinet seats to the group. Alternatively, the leaders of the marginalized
group may seek independence through secession. Again, compromises are also
possible along this territorial dimension, for example through various types of
regional autonomy arrangements that delegate decision-making authorities in
important areas, such as financial or cultural matters.
To the extent that the incumbent elite belonging to the ethnic group in
power opposes the challengers’ demands, possibly even through repression,
the challengers are especially likely to respond with violent means. Such armed
resistance may finally escalate to a full-fledged ethnonationalist civil war pitting
the included incumbent group against its excluded challenger group. Civil wars
confront incumbent governments with political and military organizations that
challenge the governments’ claim to sovereign rule. This situation corresponds
to standard definitions of civil war as a conflict that pits some nonstate actor
against the state (Sambanis 2004b; Kalyvas 2007) but excludes communal con-
flicts between nonstate actors or pogroms in which the state plays less of an
60 Inequality, Grievances, and Civil War

active role (though see Horowitz 2002; Wilkinson 2009) as well as one-sided
conflicts where the state violently represses a group lacking cohesive organi-
zation (Eck and Hultman 2007). In conflicts that are fought in the name of
excluded and discriminated groups, rebel movements are composed of mobi-
lized and militarized organizations that challenge the government. In the case of
challenges launched in the name of groups that are already represented within
government, other actors such as a faction within the army or newly created
political organizations and militias might instigate a violent confrontation.

Generating Hypotheses
We now proceed to the derivation of operational hypotheses that link nation-
alist grievances to the outbreak of civil war.3 Again, there is no better place
to start than with Gellner’s (1983) seminal book on Nations and Nation-
alism. According to Gellner, tensions result from the violation of the main
nationalist principle pertaining to the congruence of political and cultural
boundaries: “Nationalist sentiment is the feeling of anger aroused by the vio-
lation of the principle, or the feeling of satisfaction aroused by its fulfillment”
(p. 1). Given the widespread state-to-nation deficit that results from the “musi-
cal chairs” logic referred to in the previous chapter, many ethnic groups will find
themselves without adequate representation within their respective countries’
executives. Such nationality problems are further exacerbated by intermingling
of ethnic groups’ settlement areas:
It follows that a territorial political unit can only become ethnically homogenous, in such
cases, if it either kills, or expels, or assimilates all non-nationals. Their unwillingness
to suffer such fates may make the peaceful implementation of the nationalist principle
difficult (Gellner 1983, p. 2, see also p. 40).

Although Gellner’s sweeping macrohistorical theory constitutes an excellent

point of departure for the structural framing of the link between grievances and
violence, it is not, and was never meant to be, directly applicable to individual
cases. In order to do precisely this, we need to identify the pivotal institutional
variable that most directly influences the probability of conflict. Clearly, access
to state power is the crucial resource that ethnonationalist groups compete
over. Rather than being a neutral arena for competing group interests, the state
itself has to be seen as a prize. Thus,
the state has become the decisive political vehicle and political arena for ethnic groups –
dominant and subordinate – in modern conditions of rapid change in which it redis-
tributes even higher proportions of its citizens’ and subjects’ general income and assets.
To protect and articulate their social, cultural, and economic interests, grievances,
claims, anxieties, and aspirations, ethnic groups must enter the political arena – that is,
they much add the quality of becoming conflict groups to their previous qualities of

3 See also Cederman et al. (2010), who derive similar hypotheses.

Political Exclusion and Civil War 61

being status groups, interest groups, cultural groups, and the like (Rothchild 1981,
p. 232).

Already introduced in Chapters 2 and 3, this institutional logic stipulates that

where the incumbent regime insists on blocking access for challenging groups
despite repeated explicit demands for an opening of the polity, the risk of
armed conflict becomes acute. Systematic and persistent refusal to open the
polity to the whole population leaves the excluded groups with few feasible
alternatives other than resorting to a direct, and possibly violent, challenge to
the incumbents’ rule. Applying the same argument specifically to ethnic groups,
Brass (1991, p. 238) identifies the conditions under which violence becomes
particularly likely:
Where . . . the government or the leading political party draws its support exclusively
or overwhelmingly from a dominant ethnic group and propagates a chauvinist or racist
ideology, then the society in question is headed for civil war and secessionism.

Rather than being primarily a matter of the group’s own actions, the central
state’s institutional and informal policies determine the power status of ethnic
groups living within its borders. Of course, the latter are sometimes able to
influence the central government’s decision by applying pressure and extracting
concessions, but wherever the state’s vested power elite refuses to back down
and even escalates the tension by resorting to violent repression, civil war
becomes a distinct possibility.
In view of this straightforward reasoning, we derive the following general
hypothesis:

H4.1. Ethnic groups that suffer from limited access to state power are more likely
to experience conflict than those that enjoy full access.

Most obviously, we should expect excluded groups to be more likely to

experience conflict than incumbents who already enjoy governmental repre-
sentation. As observed in Chapter 4, such expectations are in line with the
findings of Gurr’s research. Based on analysis of the MAR data set, he finds
that group disadvantages, including political ones, tend to generate conflict:
“Resentments about restricted access to political positions and a collective his-
tory of lost autonomy drive separatist demands and rebellion more generally”
(Gurr 1993b, p. 188).
Furthermore, we postulate that the risk of conflict decreases gradually
according to the precise degree of power access. This means that there are also
considerable differences in terms of conflict propensity within the included and
excluded categories respectively. In particular, groups that are not only merely
marginalized, but also exposed to state-led discrimination, have been found
to be even more inclined to resort to armed resistance (Gurr 1993a; 1993b).
Recent evidence confirms these results. For example, the predictive model of
62 Inequality, Grievances, and Civil War

political instability assembled by Goldstone et al. (2010) indicates that state-led

discrimination increases the probability of conflict.
So far, we have considered the effect of power status on conflict propensity
in static terms. Viewed from a dynamic standpoint, however, the impact may
be felt even more acutely, especially where a group’s status is downgraded. As
argued in Chapter 3, the shock of demotion is likely to trigger strong emotional
reactions that spill over into violence in order to “reverse a reversal” (Petersen
2002). The sense of loss will be very obvious to those groups that have enjoyed a
privileged position. As pointed out by Rothchild (1981, p. 217), “favored ethnic
groups come to take a proprietary view of their traditional overrepresentation.”
For these reasons, we articulate a second hypothesis pertaining to power access:

H4.2. Groups that have experienced recent loss of state power are especially likely
to engage in internal conflict.

It should be noted that under specific circumstances power upgrades may

also cause conflict, at least to the extent that the incumbent offers insufficient
and reluctant concessions. According to Tocqueville’s ([1856] 1956) classical
account of the French Revolution, this phenomenon, which can be referred
to as “too little, too late,” typically features vacillation between inclusive and
exclusive stances. It is a common response by authoritarian and unrepresenta-
tive regimes to challengers’ pressure for increased regime openness (see Taylor
1948; Goodwin 1997).
Beyond such status shifts, we would expect the size of the group in question
to matter directly for the perception of grievances. Other things being equal,
the larger the excluded group, the bigger the perceived injustice. Grievances
are prone to result “if the rulers of the political unit belong to a nation other
than that of the majority of the ruled, this, for nationalists, constitutes a quite
outstandingly intolerable breech of political propriety” (Gellner 1983, p. 1).
More generally,

A sense of injustice is likely to form when a majority perceives its position as “below” a
minority (when the language of the minority is the language of state and education, when
minorities hold disproportionate numbers of political positions, and so on) (Petersen
2002, pp. 51–52).

This effect operates within both democracies and nondemocracies. Democratic

systems are known to suppress representation of minorities that are viewed as
threatening and uncooperative. Referring to the individual rights of democratic
representation, such states may also refuse to grant ethnic groups any conces-
sions whether in terms of power sharing or group autonomy. A prominent
example is Turkey, where the supreme court has banned Kurdish political par-
ties alleged to challenge the “indivisible unity of the state, the country and the
Political Exclusion and Civil War 63

nation.”4 However, even autocratic political systems depend on at least some

support from the broader population. The risk of revolution increases in cases
where the state elite is viewed to be drawn from an especially narrow basis, as
was the case in South Africa under Apartheid and remains the case of the Assad
family’s Alawite minority regime in Syria at the time of writing. Suffering from
an acute deficit of political legitimacy, such regimes are particularly dependent
on threats and actual execution of violence against their own citizens to uphold
their precarious claim to power. It goes without saying that the risk of civil
war is higher in these political systems than in those that rely on a broader and
more secure basis of population support.
In principle, democratic polities cease to be democracies as soon as a major-
ity of the population is excluded, but ethnonationalist politicians sometimes
attempt to claim democratic status by simply excluding selected parts of the
populations from citizenship. Thus, the “demos” can be redefined in more nar-
row terms as an “ethnos” (Mann 2005). The European colonial empires clung
to this double standard as their metropolitan cores democratized. Ultimately,
however, the pressure for representation made such minority rule within the
colonies unbearable and sometimes violent decolonization followed.
We sum up the demographic aspects of the grievance mechanism in the
following way:

H4.3. Large excluded ethnic groups are more likely to experience conflict than are
smaller excluded groups.

It should be noted that group size does not merely capture the extent of
grievances held in a population. This dimension can also be expected to corre-
late with a group’s resources and likely ability to challenge the state. Most obvi-
ously, large numbers translate into the availability of more fighters and more
generally, material support from more individuals (Cederman, Buhaug, and
Rød 2009; Cederman, Wimmer, and Min 2010). Yet, the argument advanced
here is that sheer size, irrespective of grievances, plays no important role in
this relationship.5 Indeed, H4.3 applies to excluded groups only, since they
are the ones more subject to grievances according to H4.1. Stated differently,
opportunities for conflict alone have no independent role absent clear motiva-
tion. As we will see below, it is possible to separate this grievance-related effect
of group size from the capacity dimension. If, however, group size increases

4 See, e.g., Robert Tait, “Turkey bans main Kurdish party over alleged terror links,” The
Guardian, Saturday 12 December 2009, online at http://www.guardian.co.uk/world/2009/dec/
12/turkey-bans-main-kurdish-party.
5 Thus, as opposed to previous related studies, our test interacts the effect of relative group size
with that of group-level exclusion. Moreover, in order to capture the comparative logic, we
employ a relational indicator that compares the group’s size to that of the incumbent groups. In
contrast, Cederman et al. (2010) rely on a nonrelational measure of group size.
64 Inequality, Grievances, and Civil War

conflict propensity for incumbent groups as well, the pure resource perspective
seems more plausible.
Finally, grievances are deeply contextual, especially with respect to prior
histories of conflict. Long-standing conflict processes that involve repeated
outbreaks of violence reflect the fact that the incompatibility fought over has
not been resolved. Such situations are likely to foster an increasingly deep sense
of resentment among the conflict parties, especially among those who have lost
family members, close friends, or property in previous conflict episodes. Indeed,
following Barbalet (1998, ch. 6), we expect vengefulness to trigger a stronger
action tendency than mere resentment.
Thus, conflicts become self-reinforcing if previous losses fuel new violence in
addition to the effect exerted by the unresolved inequality at root of the conflict.
Clearly protracted conflict histories of this type involve suspicion and fear, as
argued by theorists of the security dilemma (Posen 1993), but the emotional
reactions typically surpass a general loss of trust:
Reiterated feelings of hostility can crystallize into intense and rigid sentiments of genuine
hatred so strong as to incite violence even in the absence of specific provocation. Desires
for revenge are extremely powerful components of long-continued violent conflicts
(Williams 2003, p. 139).

Tendentious oral history as well as ideologically biased writing and teaching

of history can further perpetuate the hatred and stimulate calls for revenge
decade after decade (Kalyvas 2006). Of course, this effect should not be con-
fused with essentialist accounts referring to “ancient hatred.” These cycles of
ethnonationalist violence are distinctly modern (Mann 2005), but this does not
prevent them from being long-lasting and extremely difficult to escape from.
In fact, the literature on ethnic conflict has been so rash in its desire to debunk
primordialist journalistic accounts that it typically downplays the importance
of historical sequences of violence (Petersen 2002).
The empirical literature supports the general logic of our reasoning. In inter-
state relations, there is evidence that “enduring rivalries” between pairs of
states can lead to a high likelihood of new outbreaks of violence (Diehl and
Goertz 2000). More recently, similar patterns have been detected in internal
conflict (DeRouen and Bercovitch 2008). However, these studies are pitched
at the country level and do not consider ethnonationalist conflict more specif-
ically, beyond flagging incompatibilities arising over specific disputed territo-
ries. Thanks to our group-level data, we are able to formulate and empirically
evaluate a more specific hypothesis:

H4.4. Ethnic groups that have experienced conflict in the past are more likely to
experience conflict onset than those that have not.

Since all of our hypotheses have been developed for relations involving
groups, we need to evaluate them at this analytical level as well. It is therefore
Political Exclusion and Civil War 65

necessary to move beyond the abstract theoretical schemes that we introduced

in Chapter 3. Rather than aggregating our analysis up to the level of entire
countries (though see Chapter 7), or disaggregating it all the way down to
individuals, we follow the lead of researchers who have relied on group-level
data sets such as the Minorities at Risk (Gurr 1993a; 2000b).
Our strategy is to find operational measures of horizontal political inequali-
ties affecting specific ethnic groups and to measure conflict outbreaks associated
with the corresponding entities. This approach allows us to shorten the infer-
ential leap from country-level inequality measures to political violence. Indeed,
group-level analysis has the advantage of being much more precise in its attri-
bution of independent and dependent variables than country-level studies. It is
clear that many civil wars only affect specific parts of states, such as Kashmir
in India and Chechnya in Russia, rather than engulfing their entire territories
(Buhaug and Lujala 2005; Cederman and Gleditsch 2009). Proceeding with
country-level measures, then, risks blurring important causal connections that
can be assessed only if we have access to information about substate units and
are in a position to tally this information for both sides of the causal connection.
At the same time, the intermediate-level disaggregation takes us closer to
the level of individuals, who are the units capable of harboring grievances. As
we have seen in Chapter 2, however, treating groups as units is fraught with
difficulties and certainly does not protect our analysis from ecological fallacies
deriving from group-level reification. Yet, for empirical tractability reasons,
our approach refrains from validating our theory at the level of individuals and
subgroup organizations, and we leave this for future research.

Data
It is now clear that our empirical strategy hinges critically on the availability
of suitable data on ethnic groups and their institutional access to power in a
wide and unbiased sample, as well as on conflict indicators associated with
the same ethnic groups. In fact, there are several possible data sets that offer
information about ethnic groups. The Atlas Narodov Mira (ANM, Atlas of the
Peoples of the World) was the first major data source of this type (Bruk and
Apenchenko 1964). Assembled by Soviet anthropologists in the early 1960s,
the atlas covers ethno-linguistic groups around the entire world, but says little
about religious cleavages and nothing about the groups’ political status. The
narrow conception of ethnicity, together with its apolitical approach and the
fact that the original material is now quite dated, makes it less suitable for our
purposes.6 Fearon’s (2003) more recent list of ethnic groups attempts to offer
a selection of politically relevant groups, but the criteria for inclusion are not

6 The ANM provides useful information about ethnic groups’ settlement areas, which has served as
a useful basis in the construction of new geographic information systems, such as the Geographic
Representation of Ethnic Groups (GREG) data set (Weidmann, Rød, and Cederman 2010) and
the GeoEPR data set (see Chapter 5 and Wucherpfennig et al. 2012).
66 Inequality, Grievances, and Civil War

entirely transparent, and like the ANM, it fails to provide explicit measures of
political and institutional access.
As argued in Chapter 2, the most important listing of ethnic groups remains
the Minorities at Risk (MAR) project. The work on this pioneering resource
started in the 1980s and was first presented by Gurr (1993a). Since then MAR
has undergone a number of updates and has quickly become the most popular
source of information on ethnic groups in the empirical social-science litera-
ture. Most quantitative studies of ethnic conflict rely on the MAR data (see,
e.g., Saideman 2001; Regan and Norton 2005; Olzak 2006; Walter 2006). By
focusing on the notion of minorities “at risk,” the MAR breaks new ground as
regards political relevance, but unfortunately at the price of threatening selec-
tion bias. While the sample includes some “advantaged” minorities, it remains
incomplete since Gurr and his colleagues do not include dominant majority
groups, some of which may be at risk of being challenged by disadvantaged
minorities, and the data also do not include politically relevant minority groups
that are not considered “at risk.”

Data on Ethnic Groups’ Access to Central State Power

In preparing the empirical foundations of this book, it became clear that existing
data sets would not be able to carry the weight of our empirical analysis.
Therefore we launched a tailor-made data project that allows us to test our
hypotheses directly. In an effort to improve on previous attempts to capture
ethnic groups’ power status (Cederman and Girardin 2007; Buhaug, Cederman,
and Rød 2008; Cederman, Buhaug, and Rød 2009), Cederman’s research group
at ETH Zürich teamed up with Andreas Wimmer and Brian Min, then at
University of California at Los Angeles, in a major data collection project
that relied on an online expert survey with more than 100 participants. The
fruit of this collaboration was an entirely new data set called Ethnic Power
Relations (EPR). The original version of the data set identifies all politically
relevant ethnic groups and their access to state power, where political relevance
is defined by including groups that are active in national politics and/or directly
discriminated by the government (Cederman et al. 2010). In this book, we use
a new version, EPR-ETH, assembled by the International Conflict Research
group at ETH Zürich.7 Whereas the original data contain information about
politically relevant ethnic groups around the world from 1946 through 2005,
the new version improves the coding in many cases and extends the coverage
up to 2009.
7 The new data are publically available through our data portal GROWup , see http://growup.ethz.
ch. Descriptions of the data can be found at http://www.icr.ethz.ch/data/growup/epr-eth. We
would like to thank all the experts who helped us with the coding, including Susanne Böhn, Nils-
Christian Bormann, Johan Brosché, Michael Bürge, Naomi Furnisch Yamada, Armando Geller,
Omar Kassab, Thomas Klepsys, Ether Leemann, Dirk Leuffen, Jasan Miklian, Gudrun Østby,
Navitri Putri Guillaume, Seraina Rüegger, James Scarritt, Michael Schmitz, Guido Schwellnus,
Corinne Trescher, Jessica Trisko, Manuel Vogt, Judith Vorrath, and Toru Yamada. See http://
www.icr.ethz.ch/data/other/epr old/experts for a list of experts who contributed to the coding
of the original EPR data set.
Political Exclusion and Civil War 67

The structure of the data in the new version remains essentially the same as
in the original version. In agreement with constructivist principles, the ethnic
identities are allowed to vary over time, such that the political relevance or the
main level of political identification may move from one level to another. For
each group and time period, the demographic weight and access to power are
provided.8 The latter dimension is operationalized on the basis of the influence
over the country’s executive, including cabinet seats or control of the army in
military regimes.9
The data set divides power status into three main categories depending on
whether the group in question (1) controls power alone, (2) shares power, or
(3) is excluded from power. Goups that fall into the two first categories are
classified as included and those that belong to the last category are excluded.
The EPR classification offers subcategories for each of the three main status
categories:

Undivided Power
In case a single ethnic group controls the executive, there are two types of rule,
namely:

r Monopoly: In this case, the members of the group enjoy total monopoly
on executive power at the exclusion of other ethnic groups. The power
monopoly of the Americo-Liberians in Liberia until 1981 illustrates this
situation.
r Dominance: Falling short of full-fledged monopoly, this situation allows
for some limited but politically irrelevant participation of “token” members
of other groups. Tariq Aziz, Iraq’s foreign minister until the fall of Saddam
Hussein, is Christian rather than Sunni Muslim, but by no means represented
his own ethnic group in any politically meaningful way.

Power-Sharing Regimes
The second class of power configurations entails power-sharing arrangements,
whether of a formal or informal nature. Included groups that share power play
either a senior or junior role measured by their absolute influence over the
cabinet.

r Senior partner: If the representatives of the ethnic group in question have

at least as much executive influence as other powerful partners in the
power-sharing arrangement, then the group is classified as a senior partner.

8 The EPR data set captures absolute influence irrespective of demographic weight.
9 Partly due to data limitations, EPR does not measure legislative political power. While also
relevant as an indicator of political influence (see, e.g., Birnir 2007; Brancati 2009), control over
the executive is arguably the most crucial aspect of influence since minorities can be systematically
outvoted in majority systems.
68 Inequality, Grievances, and Civil War

In Switzerland, for example, the Swiss Germans are usually allocated the
greatest number of cabinet posts by virtue of their demographic status as a
clear majority of the country’s population.
r Junior partner: If the group’s representation lies clearly below that of the
most powerful members of the power-sharing arrangement, then it is rele-
gated to the status of junior partner. In Switzerland, both the Swiss French
and Swiss Italians usually (but in the latter case not always) enjoy repre-
sentation within the executive, although clearly below the level of the Swiss
Germans. Thus, we code them both as junior partners.

Exclusion from Central Power

The third main class of power configurations features groups that have no
regular representation within the executive and can therefore be labeled as
being excluded. There are four types of categories:
r Powerless: The group’s representatives hold no executive power, whether
at the central or regional level, but they are not systematically and openly
discriminated against. For example, the Israeli Palestinians are represented
within the Knesset, but have had their representation within the Israeli cab-
inet blocked through informal measures (Shelef 2010).
r Discriminated: The members of the group in question are subjected to
explicit and targeted discrimination that effectively blocks their access to
executive power. Discrimination can be implemented through denial of
political rights, including citizenship, but may also be carried out through
a systematic ban of parties that represent the ethnic group. The Kurds have
been systematically discriminated by the Turkish regime at least since 1946.
Similarly, the governments of Estonia and Latvia discriminate against their
Russian populations by denying them citizenship, or at least severely restrict-
ing their opportunities for becoming citizens through language proficiency
requirements.
r Regional autonomy: In this case, the group’s elite members are excluded
from central power but have been granted access to regional executive power
through a scheme of territorial autonomy. By “regional” is meant a substate
unit below the level of the state as a whole, such as provinces and federal
administrative units, though not local counties. For example, within the
former Yugoslavia, the Vojvodina Hungarians enjoyed regional autonomy
of this type.
r Separatist autonomy: This condition applies if the group’s representatives
declared de facto independence from, and against the will of, the central
government. Here the group has chosen to exclude itself rather than having
been excluded. Examples include the Bosnian Serbs walking out of the cen-
tral government of Bosnia Herzegovina or the Abkhazians having declared
independence from Georgia when the country became independent in
1991.
Political Exclusion and Civil War 69

Data on Ethnic Groups’ Conflict Participation

Having described our data on ethnic groups and their power access, we now
turn to the information sources that reveal their role in conflict. In this and the
next two chapters, we rely on a dependent variable that is based on a group-
level coding of the Uppsala Conflict Data Program’s and the Peace Research
Institute Oslo’s UCDP/PRIO Armed Conflict Data Set (Gleditsch et al. 2002),
with supplementary information from the Non-State Actors data set (Cunning-
ham, Gleditsch, and Salehyan 2009). The ACD2EPR data set assures that each
conflict onset is mapped to the corresponding EPR group, provided that the
rebel organization expresses an aim to support the ethnic group and mem-
bers of the group in question participate in combat. We note that there can
be more than one organization claiming to represent an ethnic group, and we
will return to the mapping between EPR groups and rebel organizations in
Chapter 8.

Variables
In the current chapter, we use a dichotomous dependent variable that indi-
cates whether a group experienced an outbreak of conflict with the incumbent
government. The onset variable is coded one for a group year during which
a conflict started and zero otherwise. We drop observations with ongoing
conflict. Since ethnic groups that already enjoy monopoly or dominance by
definition cannot rebel against themselves, we also discard these groups from
our group-level analysis.
We proceed by describing the independent variables at the group level. Our
main independent variables pertain to the power access status of the group in
question.
r In many models, we use a dichotomous measure that captures EPR’s dis-
tinction between excluded and included groups as defined above.
r In other cases we extend the power access variables to all status categories of
EPR by using dummy variables for each of these categories: junior, power-
less, discrimination, regional, and separatist autonomy (using senior mem-
bership in power sharing as the reference category, while dropping all cases
pertaining to monopoly and dominancet status since groups cannot chal-
lenge themselves).
r Finally, we consider a dummy variable labeled downgraded, indicating if
the group suffered a decline in EPR status during the last two years.

Demographic group size is measured through EPR-ETH’s population size esti-

mates. Denoting the group’s population as G and the population of the incum-
bent as I,10 we operationalize the relative group size as follows:

10 Note that the incumbent group can consist of several groups. In this case, I is computed as the
sum of these groups’ population.
70 Inequality, Grievances, and Civil War

r Relative group size g ࢠ [0,1) in the primary dyad is defined as G / (G + I) if

the group is excluded and as G/I if the group is included (since the rebelling
group left the incumbent coalition and would otherwise be counted twice).
We also include a measure reporting on previous conflict history:
r The count variable number of previous conflicts indicates the number of
conflicts the group has experienced since 1946 or the independence of the
country.
We introduce a number of variables to control for country-level properties:
r Logged GDP per capita of the country as a whole, lagged (Penn World Table
7.0, see Heston, Summers, and Aten 2011).
r Logged population size of the country, lagged (Penn World Table 7.0, see
Heston et al. 2011).
r Dummy variable ongoing conflict based on the ACD2EPR conflict data
indicating if there was an ongoing conflict involving any other group in the
country during the preceding year.
r Number of years since the previous conflict, entered as a nonlinear function,
based on natural cubic splines with three knots (see Beck, Katz, and Tucker
1998).
r Finally, since we contrast ethnic exclusion as distinct from general differences
in degree of democracy and autocracy, we consider a measure of the extent
to which countries are democratic (Scalar Index of Polities, see Gates et al.
2006).11

Results
Having described the data and the main variables, we are now ready to evaluate
our hypotheses on the effect of political horizontal inequality. This section starts
by presenting descriptive statistics followed by multivariate regression analysis
that allows us to control for potentially confounding factors.
Our data set contains 29,740 group years all pertaining to politically rel-
evant ethnic groups in independent states from 1946 through 2009, exclud-
ing all years associated with ongoing conflict. Of all these observations, only
207 group years featured conflict onset. This suggests that the outbreak of
conflict is very much a low-probability event. As a first assessment of H4.1,

11 The Scalar Index of Polities (SIP) democracy measure combines information on the regulation of
executive recruitment and executive constraints from the Polity data (Jaggers and Gurr 1995)
with data on the extent of political participation from Vanhanen (2000). The conventional
Polity index has been criticized for relying on coding of “factionalized competition,” which
could reflect political violence (see Vreeland 2008). Using the SIP measure allows us to avoid
the problems of factionalized competition in the conventional Polity data as well to take into
account the participation dimension, which is largely ignored in the institutional focus of the
Polity data (see, e.g., Paxton 2000; Vanhanen 2000).
Political Exclusion and Civil War 71

table 4.1. Comparing Excluded and Included

Groups’ Conflict Propensities

Group Years Onsets

Group included 8,951 29 (0.32%)
Group excluded 20,582 178 (0.86%)
Total 29,533 207 (0.70%)

Table 4.1 compares excluded and included groups as regards their conflict
proneness.12 Whereas the former category experienced as many as 178 out-
breaks, the latter merely contains 29 such cases. These numbers translate into
starkly different relative frequencies: the marginalized groups attained a con-
flict onset frequency of 0.86% per year, while the rate for incumbent ones
groups is merely a third of this frequency, 0.32%. In this simple cross tabula-
tion, the difference between the two power status classes is clearly significant
(p > 0.001). This is a first, albeit crude, indication that power access does
reduce the frequency of conflict, as suggested by H4.1.
We proceed by breaking up the composite cases of included and excluded
groups into their respective EPR power status categories. Table 4.2 reports on
the results of this exercise. Again, the descriptive statistics correspond to our
expectations. The most privileged groups are those that experience the least
conflict, and not surprisingly we see no conflict onsets involving dominant
groups. Senior partners in power sharing arrangements have as low a conflict
rate as 0.23%, compared to junior partners whose rate is much higher (0.37%).
The excluded groups, however, exhibit more variation: as would be expected,
those groups that enjoy at least regional autonomy experience the least conflict,
namely in only 0.47% of the years, followed by powerless groups (0.64%).
When we turn to discriminated groups, the likelihood of onset increases to
1.24%. Finally, the special case of separatist autonomy, that is, of self-excluded
groups, corresponds to an extremely high rate of 5.23%.
Descriptive statistics are useful to provide a first assessment of whether
power access influences conflict behavior. Yet, it is difficult to test the other
hypotheses in this way, and we cannot rule out that other, confounding factors
might make the grievance effects spurious. Therefore, we introduce a series of
multivariate regression models that allow us to compare the influence of several
independent variables simultaneously. In all these models, we use the same
sample as in the descriptive analysis, namely all group years from 1946 through
2009 except politically dominant groups, and those that are not politically
relevant or characterized by ongoing conflict. Subject to data availability, this
leaves us with a sample of 28,302 observations. The analysis is conducted by

12 Again, it should be noted that groups enjoying monopoly or dominance have been dropped
from the included category since our conflict variable only records challenges to state power.
72 Inequality, Grievances, and Civil War

table 4.2. Ethnic Groups’ Conflict Propensity by

Power-Access Category

Group Years Onsets

Included groups:
Monopoly 1,846 0
Dominance 2,105 0
Senior Partner 3,074 7 (0.23%)
Junior Partner 5,877 22 (0.37%)
Excluded groups:
Regional Autonomy 5,320 25 (0.47%)
Powerless 9,735 63 (0.64%)
Discriminated 5,038 63 (1.24%)
Separatist Autonomy 489 27 (5.23%)
Total 33,484 207 (0.61%)

means of logit regression models with robust, country-clustered standard errors

and corrections for temporal dependence according to Beck et al. (1998).
Our main results are displayed in Table 4.3. All three models reported mea-
sure the effect of a series of group-level properties in addition to country-level
controls. We start by considering Model 4.1, which introduces a dichotomous
classification of power access (see Table 4.1 above). In agreement with H4.1,
the excluded dummy variable has a strong positive effect that is highly sig-
nificant as well. There is also convincing evidence in favor of H4.2, since the
downgraded indicator is also substantively and statistically significant. More-
over, we likewise detect a positive effect of relative group size in support of
H4.3. Finally, turning to our final group-level measure allows us to corroborate
H4.4, which links rebellion to the number of previous conflicts. Moreover, we
find that the influence of two out of our three country-level control variables
can be safely distinguished from zero. Ongoing conflict elsewhere in the country
significantly increases the probability of conflict. Unsurprisingly, ethnic groups
in wealthy countries are less inclined to be involved in violent conflict with the
government. The only control variable that seems to have no effect is the over-
all population size of the country, which is one of the most robust correlates
of civil war risk at the country level (Hegre and Sambanis 2006).
Proceeding to Model 4.2 enables us to disaggregate the dichotomous cate-
gorization of included and excluded groups. Using senior partners in power-
sharing arrangements as the base category, the series of categorical variables
successively decreases the access to executive power, as was done with descrip-
tive statistics in Table 4.2. As we move down this scale, we notice that the
coefficients increase and become increasingly significant for each step. More
specifically, we find that junior power-sharing partners are more conflict prone
than senior ones, although not significantly so. As we shift the attention to
Political Exclusion and Civil War 73

table 4.3. Political Horizontal Inequality and Group-Level Onset of Civil War,
1946–2009

Model 4.1 Model 4.2 Model 4.3 Model 4.4

Group-level variables:
Group Excluded 1.0958** 0.7086* 1.2182**
(0.2020) (0.3042) (0.2446)
Junior Status 0.8392
(0.5210)
Autonomy Status 1.1844*
(0.5758)
Powerless Status 1.4248**
(0.5294)
Discriminated Status 2.0008**
(0.5048)
Separatist Status 3.2109**
(0.6592)
Downgraded 1.4816** 1.5762** 1.4879** 1.4305**
(0.2988) (0.2935) (0.3004) (0.3404)
Rel. Group Size 1.1020** 1.3885** 1.3067**
(0.3406) (0.3145) (0.3941)
Rel. Group Size (Excl.) 1.3578**
(0.3978)
Rel. Group Size (Incl.) 0.0860
(0.8046)
Number of Previous Conflicts 0.6732** 0.5571** 0.6763** 0.7219**
(0.0742) (0.0972) (0.0716) (0.0897)
Country-level variables:
Ongoing Conflict, lag 0.6116* 0.3876 0.6126*
0.6664*
(0.2903) (0.2882) (0.2886)
(0.2997)
GDP/capita, lag, log − 0.1982* − 0.1672* − 0.1885*
− 0.2303*
(0.0809) (0.0727) (0.0822)
(0.0927)
Population, lag, log 0.0076 0.0827 0.0156
0.0050
(0.0929) (0.0895) (0.0885)
(0.0875)
Democracy, lag 0.6033
(0.4404)
Constant − 4.1221** − 5.7704** − 3.9416** − 4.5297**
(1.0654) (1.0971) (1.0573) (1.0498)
Observations 28,302 28,302 28,302 26,306
Robust standard errors in parentheses; estimates for peace-year correction not shown.
** p < 0.01, * p < 0.05.
74 Inequality, Grievances, and Civil War

the excluded categories, the coefficients increase even further, culminating with
discriminated status. Again, we treat separatist autonomy as a special case with
extremely high levels of conflict, due to their extraordinary challenge to the
central authorities, and in many cases involving prior mobilization. The other
group-level indicators replicate the behavior of Model 4.1, thus confirming all
four hypotheses (although H4.3 only approximately). Moreover, there are no
major differences as regards the country-level controls.
We are now ready to study Model 4.3, which provides a more nuanced
assessment of the size effect as stipulated by H4.3. While the two previous mod-
els captured these effects for all ethnic groups, the current model distinguishes
between excluded and included groups. This distinction enables a more precise
evaluation of the grievance dimension. Instead of measuring group size across
the board, we divide this variable into two components by interacting it with
the dummy variables corresponding to excluded and included groups, respec-
tively. The results indicate that relative group size increases conflict propensity
if the group in question is excluded. In contrast, the relative size of incumbent
appears leave conflict propensity mostly unchanged.
We have argued that exclusion is distinct from general differences in degree
of democracy and autocracy, and that ethnic exclusion is more relevant for the
motivations to resort to conflict than the presence or absence of democratic
institution in terms of individual representation. Although ethnic exclusion and
democracy are clearly conceptually distinct, and ethnic exclusion can occur
under relatively competitive democratic institutions, it is also the case that eth-
nic exclusion is likely to be more widespread in nondemocracies. To ensure
that our results regarding exclusion do not arise as an artifact from not consid-
ering political democracy we test the relationship between political democracy
and conflict directly in Model 4.4. Consistent with previous research using the
polity data, we find no statistically significant effect of political democracy on
the outbreak of ethnic civil war at the group level (see Wimmer, Cederman,
and Min 2009). However, the coefficients for excluded and downgraded groups
remain large and statistically significant, confirming our argument that ethnic
exclusion is more relevant for the outbreak of conflict than lack of democracy
per se. Additional robustness tests using a curvilinear relations specification
of the relationship between democracy and civil war similarly find no consis-
tent effect of democracy on civil war, while the effects of ethnic exclusion and
downgrading remain highly statistically significant and consistent with Models
4.1 and 4.4.
While promising, the regression results presented in Table 4.3 indicate effects
on the log-odds of conflict that are somewhat difficult to interpret, especially
when it comes to the interactive terms. Therefore, we introduce several graphs
that report on the implied outcomes in terms of conflict probabilities. Based on
the estimation of Model 4.1, Figure 4.2 compares the predicted probability of
conflict onset for included and excluded groups while holding all other indepen-
dent variables at their means. The comparison yields a clear picture: excluded
Political Exclusion and Civil War 75

0.010
0.008
Pred cted Probab ty of Conf ct

●
0.006
0.004
0.002

●
0.000

Included Excluded

Political Status
figure 4.2. Comparing the effect of power access status on civil war onset.

groups are much more likely to rebel than included ones. Their conflict prob-
ability exceeds 6.8%, which is about three times as high a conflict propensity
as for the ethnic groups in power. The respective confidence intervals (which
correspond to two-sided standard errors with p > 0.05) are well separated,
thus indicating that the difference between the two basic status categories is
statistically quite robust.
We illustrate the results of Model 4.2 with an expanded graph (see Figure
4.3). This illustration reports the predicted conflict propensity for each sepa-
rate EPR power status category. The figure confirms, and renders more intu-
itive, what we have already found in Table 4.3. Indeed, the conflict likelihood
increases gradually with decreasing power access, moving from left to right in
the graph. As would be expected, the most inclusive category, which pertains
to senior partners in power-sharing arrangements, is the one that is the most
pacific (predicted annual onset probability = 0.13%), followed by junior part-
ners, which experience a probability of onset more than twice as high (predicted
probability = 0.31%). When we turn to the excluded groups, we can see that the
likelihood of conflict increases still further. Groups enjoying regional autonomy
have a predicted probability of 0.44%, powerless ones a probability of 0.55%,
and discriminated ones a probability of 0.97%. In agreement with the coef-
ficient estimates reported for Model 4.2, we find that separatist autonomous
76 Inequality, Grievances, and Civil War
0 05
0 04
Predicted Probability of Conflict

●
0 03
0 02
0 01

●
●
●
●
0 00

Senior Partner Junior Partner Regional Aut. Powerless Discriminated Separatist Aut.

Political Status

figure 4.3. Comparing the effect on civil war onset for EPR categories.

groups are extremely conflict prone (predicted probability = 3.24%), however,

with large error bands due to few observations.
Having found very strong support for the general claim that power access is
negatively related to conflict risk, as stated by Hypothesis H4.1, we now turn
to the consequences of downgrading or reversals in power status. According to
H4.2, groups that lose power should experience a much higher risk of conflict
than other groups. Based on Model 4.1, Figure 4.4 provides a straightforward
comparison of the predicted probabilities for those groups that suffer a loss of
power status to those that do not. The graph reveals that downgrading leads to
a massive increase in conflict risk, from 0.46% to 2.0%, or an almost fourfold
increase.
Our graphical investigation of the key hypotheses proceeds with an evalua-
tion of H4.3, which postulates a size-dependent increase of conflict probability
for excluded groups. Using 0.05 error bands, Figure 4.5 plots the predicted
Political Exclusion and Civil War 77

0.04
Pred cted Probab ty of Conf ct

0.03
0.02

●
0.01

●
0.00

Not Downgraded Downgraded

Change of Political Status

figure 4.4. The effect of status reversals on the probability of civil war onset.

probability of conflict as a function of relative group size for excluded and

included groups (see the upper and lower curves with dark and light gray
error bands, respectively). Whereas the size of the incumbent groups makes
little difference for the probability of onset, the effect is strongly increasing for
the excluded ones. Furthermore, except for very small relative sizes, the error
bands are clearly distinct for the two categories, thus indicating that the two
types of groups are very different in terms of conflict risk. As postulated by
H4.3, demographic size should have a much higher conflict-inducing effect for
excluded groups that voice grievances more readily than included ones. The
separation in terms of size dependence is a strong indication that demographic
variables operate together with grievances rather than being mere proxies for
resource endowments.
Finally, we consider the role of conflict history as stipulated by H4.4. Based
on results drawn from Model 4.1, Figure 4.6 displays the predicted probability
of conflict onset for increasing numbers of previous conflict. The graph suggests
that the more groups are exposed to conflict, the more likely they become to
experience renewed conflict onsets. For the maximum number of prior onsets,
the probability of conflict rises above 0.2 per annum, although the small num-
ber of cases that falls into this category broadens the error bands. All in all,
H4.4 receives very strong support based on this graphical depiction.
78 Inequality, Grievances, and Civil War

0.015
Predicted Probability of Conflict

Excluded Groups
0.010
0.005

Included Groups
0.000

0.0 0.2 0.4 0.6 0.8 1.0

Relative Size
figure 4.5. Conflict propensity of included and excluded groups as a function of their
relative size.

Validating the Causal Arguments

Although suggestive, the statistical evidence presented in this chapter does
not prove that the mechanisms that we have highlighted actually generate the
observed relationship. In this section, we present a brief analysis of some of
the specific conflict onsets that are consistent with our results, focusing on the
four main hypotheses in this chapter. Our intention here is to provide examples
that can help establish whether our postulated mechanisms actually operate as
expected; presenting full-fledged case studies with original material for each
conflict is beyond the scope of what we can cover in this book. However, we
will also consider the main peaceful anomalies where we would have expected
conflict and assess what lessons can be learned from these. Finally, this section
discusses the robustness of our findings, especially as regards the threat of
endogeneity.
We first turn to our most basic claim as formulated in H4.1, which links
inhibited power access to a heightened risk of war. Table 4.1 shows that by far
the largest number onsets can be found among the excluded groups. Apart from
the special case of separatist autonomy, to which we will return, our theory
expects discrimination to generate especially explosive grievances. Of these
numerous cases (there are as many as 62 onsets), some clearly involve repressive
Political Exclusion and Civil War 79

0.4
Pred cted Probab ty of Conf ct

0.3
0.2
0.1
0.0

0 1 2 3 4 5 6

Number of Previous Conflicts

figure 4.6. Conflict propensity as a function of the number of the group’s previous
conflict onsets.

states adhering to explicitly ethnonationalist ideology and discriminated groups

reacting violently to such treatment.
We start our exploration of discriminated groups by studying the conflicts
associated with the Turkish Kurds. Modern Turkey under the leadership of
Mustafa Kemal Atatürk, emerging out of the ruins of the Ottoman Empire
after the end of World War I, has adhered to an especially ambitious pro-
gram of nation building. In an effort to modernize and centralize state power,
the Turkish nationalists distanced themselves from the religious identities of
the ancien régime. However, the commitment to secular patriotism did by no
means imply ethnic inclusion, and ironically actually weakened the religious
bonds that had previously held some ethnic groups together, most notably the
Turks and the Kurds (Kirisci and Winrow 1997, p. 97). Categorically refus-
ing to recognize the country’s ethnic diversity, the Turkish national ideology
consistently blocked the minorities’ political representation and resorted to
discriminatory and repressive measures to thwart any challenge to Turkish
identity (Gunes 2012).
Kurdish nationalism developed in direct reaction to the Turkish nation-
building project. The Kurds rebelled as early as in the 1920s and 1930s (Kirisci
and Winrow 1997, p. 100). Although the Turkish state appeared to grow
somewhat more tolerant in the early 1970s, especially under Prime Minister
80 Inequality, Grievances, and Civil War

Bülent Ecevit, the government reneged on its initial promises and did not imple-
ment any significant reforms. Consequently, the Kurdish nationalist opposition
radicalized and came under the sway of the Kurdistan Workers’ Party (PKK).
Alarmed by the expressions of Kurdish separatism, the military devoted massive
resources to control the minority in the 1980s. In 1982 a new constitution was
adopted, which reinforced the commitment to Kemalist principles. In 1983,
a special law was introduced to ban the use of the Kurdish language in any
context (Kirisci and Winrow 1997, p. 111). In response, the PKK launched its
insurgency in 1984, which is the year of onset in our data set. In sum, the Kurds
in Turkey constitute a classic case of political discrimination causing internal
conflict through a grievance mechanism.
In keeping with its historical role as the original incubator of ethnic national-
ism, Eastern Europe offers other prominent examples of discrimination trigger-
ing conflict. The former Yugoslavia stands out in the post–Cold War period.
Our data record two conflict onsets involving discriminated groups, namely
the Serb minority in Croatia in 1991 and the Albanians in Serbia (Kosovo)
in 1998. In both cases, the respective governments advanced aggressively eth-
nonationalist projects that explicitly targeted ethnic minorities. Both Croa-
tia under Tuđman and Serbia under Milošević engaged in aggressive nation-
building programs that reversed the multiethnic system of federalism of the
former Yugoslavia (Brubaker 1996). Because of their geopolitical importance
and intensity, the discriminating policies and resulting grievances have been
extensively covered in the literature. In his careful study of emotions and eth-
nic conflict, Petersen (2002) shows how the Tuđman government used anti-Serb
symbols, such as the red and white checkered flag associated with the infamous
Ustaša regime during World War II and explicitly downgraded the consti-
tutional status of the Croatian Serbian minority, thus expelling Serbs from
“almost all day-to-day positions of dominance” in Croatia (p. 228). These
ambitious and uncompromising projects triggered armed resistance through
an onset of internal conflict in 1991, primarily in the Krajina region of Croatia.
The discrimination of the Albanian minority in the former Yugoslavia
has deep roots. It was mitigated through federal reforms granting Kosovo
autonomy in 1966. However, in 1989 the Milošević regime abolished the
autonomous status of the Albanian-majority province, a move that escalated
the already considerable tensions in Serb-Kosovar relations and accelerated
the centrifugal tendencies within the federation. This move, combined with
other discriminatory practices by the Serb-dominated government in Belgrade,
triggered the most immediate violent reaction in Slovenia, Croatia, and Bosnia-
Herzegovina, but the tensions rose inexorably in Kosovo in the course of the
1990s as a direct consequence of the Milošević regime’s oppressive policies
toward the Albanian minority that amounted to “virtual military occupation”
(Gagnon 2004, p. 123; see also discussion in Chapter 5). Although the initial
organized resistance by the Kosovar community focused on non-violent resis-
tance and developing separate institutions under Rugova’s Democratic League
Political Exclusion and Civil War 81

of Kosovo, a violent insurgency broke out in 1998 with the emergence of the
Kosovo Liberation Army (UÇK), which subsequently triggered Belgrade-led
ethnic cleansing and NATO’s intervention in 1999.
The Israeli-Palestinian conflict constitutes another textbook example of eth-
nonationalist discrimination fueling conflict. At the root a classical nationality
problem where two peoples claim the same territory, the confrontation started
well before the independence of Israel in 1948. Reacting to increasing anti-
Semitism in Western Europe, the Zionist pioneers soon opted for Palestine as
their homeland. Given that they mostly stemmed from Eastern Europe, it is not
surprising that the ideological project was directly inspired by ethnic nation-
alism originating from that part of the world (Shindler 2002; Shlaim 2009).
The British withdrawal from the Palestinian Mandate left a power vacuum that
the Israeli Jews used to launch a systematic and effective campaign of ethnic
cleansing that drove the vast majority of Palestinians into exile (Pappe 2004;
Shlaim 2009). In contrast, Palestinian resistance proved erratic and badly orga-
nized, as was the armed intervention by neighboring Arab states that followed
the Israeli declaration of independence in May 1948. The onset associated
with the Palestinians in 1948 can reasonably be seen as a direct reaction to
overt and brutal discrimination as manifested by the Israeli campaign.13 As
a reflection of their intense grievances, the Palestinians refer to the events of
1948 as “Nakbah,” the catastrophe. According to Pappe (2004, p. 141), this
event has been “kindling the fire that would unite the Palestinians in a national
movement. Its self-image would be that of an indigenous population led by
a guerilla movement wishing without success to turn the clock back.” Pales-
tinians who chose not to flee were subjected to military rule that extended the
state’s explicitly discriminatory policies to the beginning of the occupation of
the West Bank and Gaza in 1967 (Pappe 2004).
This geopolitical upheaval perpetuated explicit discrimination of the Arabs
but shifted the target of discriminatory policies from the Israeli Arabs to the
Palestinian refugee populations in the occupied territories. While the Jewish
state’s relations with the former group has been tense but relatively peaceful,
relations with the latter group are still characterized by enduring conflict. The
most recent case of severe escalation is referred to as the al-Aqsa intifada, a
Palestinian uprising that reacted to the deliberately provocative visit to the Tem-
ple Mount by prime minister Ariel Sharon in September 2000 (Shlaim 2009).
Here we see a clear-cut confirmation of the conflict-fueling effect of discrim-
inatory policies: the conflict onset involving the Palestinians in 2000 reflects
the Palestinians’ intense resentment, cemented into hatred and a deep sense of
hopelessness and humiliation. These grievances resulted from several decades of
Israel’s oppressive and ethnically biased policies concerning political influence,
and more generally all aspects of everyday life.

13 However, there may be some concerns about endogeneity when it comes to the sequence of
discrimination and the onset of civil war in connection with Israeli independence.
82 Inequality, Grievances, and Civil War

Our postulated casual pathway passing through grievances receives further

support from other cases where discrimination triggers conflict outbreaks. The
ethnocratic policies of the state-controlling Burmese help account for a number
of long-standing separatist conflicts in Myanmar involving groups such as the
Karenni, Kachins, and Shan (Brown 1994). These groups have resisted state-led
oppression and assimilatory centralism and retreated into inaccessible terrain,
fighting the Burmese state with insurgency tactics (Scott 2009). Likewise, ethnic
domination by the Amhara in Ethiopia, first through the imperial regime of
Haile Selassie and subsequently the Marxist junta under Mengistu, triggered
a series of conflict onsets. These challenges to central power took the form of
anticolonial campaigns in response to the government’s persistent unwillingness
to make concessions in terms of power sharing or regional autonomy (see
Scarritt 1993; Young 1996). Whereas the Eritreans managed to secede from
Ethiopia in 1991, other groups have been involved in enduring conflict, such
as the Somalis in Ogaden.
Other conflict cases are less characterized by ideological claims and coun-
terclaims along ethnonationalist lines, but rather by de facto ethnic domi-
nation through overwhelming military force. In Nigeria, Africa’s first great
post-independence war – the Biafra war that caused over a million casualties
overall – was the immediate consequence of a military coup by Northern offi-
cers mainly from the Hausa-Fulani ethnic group and the subsequent pogroms
against Igbos in the northern region. Despite a federalist state structure, the
uneasy cohabitation of Nigeria’s three major ethnic groups – the Hausa-Fulani,
Yoruba, and Igbo groups – had deteriorated into an intense ethno-political com-
petition over power in the central government. Eventually, a military coup by
mostly Igbo officers in January 1966 – intended to reunite the country behind
a pan-Nigerian ideology – brought Nigeria’s First Republic to an end (Lloyd
1970, pp. 10–11; Ige 1995, pp. 290–6, 303, 309–11). Under the circumstances,
the coup rapidly became perceived as an Igbo strike against northern domi-
nance and ignited deep fears in the north of becoming politically excluded
(Lloyd 1970, p. 11; Morrison, Mitchell, and Paden 1972, p. 316; Ige, 1995,
pp. 319–20). This led to a much more clearly ethnically motivated counter-
coup by mostly northern officers only six months later, followed by massive
anti-Igbo violence in the north. The secession of the self-declared Republic of
Biafra – the Igbo-dominated eastern region of Nigeria – in May 1967 was por-
trayed by the secessionist leaders as a last resort to ensure Igbo survival in the
face of a northern-dominated central government. Hence, the ensuing civil war
constituted the bloody culmination of this ethno-political power struggle and
demonstrates the dangerous consequences of political exclusion in the highly
competitive political systems of Sub-Saharan Africa (Diamond 1988).
Many subsequent conflicts in Sub-Saharan Africa were fought mainly among
elites, with only modest popular support. Examples include the more recent
insurrection of the Ijaws in Nigeria (Zinn 2005), the insurgency against the
oppressive “Arabism” of the Sudanese government (Prunier 1995), and the
Political Exclusion and Civil War 83

northern rebellions in Côte d’Ivoire in 2002 and 2011 (Vogt 2007; Bouquet
2011; McGovern 2011), as well as other challenges to central rule of the
Tuareg, Arabs in Mali 1994 (Humphreys and ag Mohamed 2005), and the
Langi/Acholi in Uganda in 1986 (Lindemann 2011). Even Charles Taylor’s
invasion in Liberia in 1989 drew mostly on support from the Gio and Mano
ethnic groups – precisely those groups that had suffered the most under Doe’s
Krahn-dominated dictatorship (Outram, 1999; Bøås 2001; Ballah & Abrokwaa
2003). All these examples clearly underline the causal mechanisms behind
our statistical results, demonstrating how political exclusion or discrimination
along ethnic lines may lead to ethnic civil conflicts.
Yet, the most extreme cases of “internal colonialism” are the violent attempts
to overturn oppressive and explicitly racist minority rule over large majorities,
as illustrated by the African National Congress’ struggle against the Apartheid
regime in South Africa (which turned violent from 1981); the armed struggle
in 1967 against Ian Smith’s white supremacists in Rhodesia; the repeated Shiite
armed resistance to Saddam Hussein’s Ba’athist rule in Iraq in 1982, 1987, and
1991; as well as the violent toppling of the Americo-Liberians in 1980.
Although spanning a wide spectrum from ideologically sophisticated eth-
nonationalism to brute ethnic domination, these conflict cases all feature at
least some measure of reactive grievances protesting the unfair treatment of dis-
criminated groups. The EPR category of “powerlessness” encompasses mostly
smaller, marginalized groups. Their limited demographic size explains why
they tend to attract less attention from the government than discriminated
groups. All the same, grievances are prone to form among members of periph-
eral groups, especially where the state is seen as exploiting “their” resources.
A fitting example is the Acehnese revolt against Jakarta’s projection of power
and perceived exploitation of the natural gas fields that were discovered in the
1970s. As Aspinall (2007) observes, general grievances along these lines were
widespread in Aceh from the late 1970s:
The perception that Aceh’s resources were being drained to benefit others remained
central to Acehnese discontent over succeeding years. After the fall of Suharto, even a
casual visitor to Aceh would be regaled with stories about how “Java,” “Jakarta,” or
“the center” sucked the territory’s natural wealth away from it (p. 955).

According to Aspinall (2007), ethnic mobilization was broadened when the

Free Aceh Movement (Gerakan Aceh Merdeka, or GAM) successfully man-
aged to transform these general resentments into an effective collective action
frame, ultimately leading to a conflict onset in 1990. The occurrence of natu-
ral resource riches superficially appears to vindicate the greed-driven account
of Collier and Hoeffler (2004) and other political economists. However, the
fact that resources captured by the center generate collective action through
grievances and appeals to normative principles of fairness is consistent with
our theoretical framework and clearly inconsistent with Collier and Hoeffler’s
account.
84 Inequality, Grievances, and Civil War

It is not difficult to find other examples of small peripheral groups that

engage in armed resistance as a response to the state’s encroachment on their
perceived homeland, especially in cases where the settlement area contains oil
fields or other natural resources. Like the Acehnese, the Papuans have revolted
against a perceived pattern of exploitation. In the cases of West Papua, the
conflict has concerned, among other things, the center’s hunger for tropical
timber (Sukma 2005, p. 15). Beyond Indonesia, we find a similar pattern of
conflict in China’s colonization of Tibet (Ye 2007), the Sudanese government’s
marginalization of the South (Ali et al. 2005), and the Cabindan Mayombe
in Angola (Malaquias 2000). These cases generally coincide with Weiner’s
(1978) stylized conflict type involving the “sons of the soil,” who react to the
imposition of “alien” rule through immigration campaigns from the center to
the periphery, as illustrated by insurgencies in India triggered by the Kashmiri
Muslims, the Bodo, the Mizo, and indigenous Tripuri.
The powerless minorities discussed so far are relatively small compared to
the state-controlling groups. Yet, the EPR-ETH data set indicates that some
African conflicts concern larger, more centrally located groups, whose elites
compete for control over the state with representatives of other groups of
comparable size. Attempting to overcome ethnic exclusion, elite representatives
engage in struggles to secure their access to executive power, as illustrated by the
Bakongo who fought the Angolan government through the rebel organization
National Liberation Front of Angola (Frente Nacional da Libertação de Angola,
FNLA) from 1994 (Malaquias 2000), the Ewe in Togo, who challenged central
power in 1988 (Decalo 1996), or the Lari/Bakongo’s rebellion in Congo in
1998 and 2002 (Clark 2008). Grievances were clearly present in these cases,
even if not necessarily as widespread and deeply rooted in the respective groups’
populations as in the more ideological cases referred to above.
The EPR category of regional autonomy entails the mildest form of exclu-
sion since the government has granted ethnic groups considerable autonomous
powers. However, nationalist leaders representing such minorities may find the
concessions insufficient since they exclude meaningful influence over the state’s
executive power. Moreover, some representatives may radicalize and claim a
higher measure of autonomy all the way to full independence. Typically such
radicalized elites splinter away from more moderate politicians who are willing
to collaborate with the central government. Such splintering and radicalization
often leads to recurrent outbreaks of secessionist conflict, as shown by insur-
rections by small groups such as the aforementioned Aceh, as well as the Naga,
Tripuri, and Manipur in Northeast India (Bhaumik 2007) and the Moro in the
Philippines (Coronel Ferrer 2005).
Where the terrain is especially inaccessible, the groups in question may
even manage to escape the reach of the state though de facto secession (Scott
2009). As we have postulated, such cases of separatist autonomy will almost
certainly end in violent conflict because even weak states are loath to abandon
the aspiration of territorial sovereignty, especially given the possible precedent
Political Exclusion and Civil War 85

this might set with respect to other minorities (Walter 2006; 2009b; though
see Forsberg 2013). This stands in stark contrast to the previous category of
regional autonomy, which implies a relatively low risk of conflict compared
to other types of ethnic exclusion. The main difference, of course, is that in
the case of separatist autonomy, groups themselves decide to shift power to
the regional level, rather than autonomy being granted by the government.
Interestingly, conflict onsets involving separatist autonomy only seem to occur
in Eurasia. Prominent instances include the Abkhazians and the South Ossetians
in Georgia (Cornell 2002), the Chechens in Russia (Zürcher 2007), several
ethnic groups in Myanmar (Brown 1994; Callahan 2005), and the Kurds in
Iraq (Yildiz 2004).
So far, we have considered conflicts that erupt in response to different degrees
of ethnic exclusion, from stark cases of discrimination to partly cooperative
situations involving far-reaching group autonomy. Although there is consider-
able variation in the explicitness and intensity of ethnonationalist grievances,
the qualitative evidence offers ample support for the causal chain sketched in
Chapter 3. According to H4.1, we would expect groups that enjoy access to
executive power to experience less conflict, but the hypothesis is, of course, not
deterministic.
In fact, we record several onsets involving included ethnic groups. Clearly,
power-sharing arrangements can help manage peaceful relations between
groups but may also generate violent conflict. When is such a development
likely? Our data tell us that, even though the conflict frequency is considerably
lower in inclusive ethnic arrangements, several groups that participated in eth-
nic governing coalitions as junior partners took up arms against the rest of the
government (see also Wimmer, Cederman, and Min 2009). The most obvious
reason why such cooperation collapses is that radical members of groups that
play a secondary role in power sharing are unhappy with the power distri-
bution, especially if clearly underrepresented in relation to their demographic
weight. After years of Tutsi minority rule and military oppression in Burundi,
the regime allowed for some opening and brought Hutu members into the
government in the late 1980s. However, Hutu radicals had little trust in these
concessions and considered them “too little, too late” (Lemarchand 1994, p.
132). In 1991, fighters belonging to the radical resistance group Palipehutu
took up arms against the government. These events were followed by a larger
explosion of violence in 1993 when radical Tutsi officers murdered the demo-
cratically elected Hutu president Melchior Ndadaye and several other Tutsi
officials (Lemarchand 2004, p. 63).
Other relatively large groups in junior positions that have reacted violently
include the Oroma in Ethiopia, the Chadian Sara in 1992, Bosnian Croats in
1993, the Macedonian Albanians in 2001, and the Afar in Djibouti in 1998.
However, even where the group in question is small, splintering may lead to
extremist groups pushing for secession, as has been the case with the Assamese
and Punjabi-Sikh campaigns against Delhi’s rule (Mitra 1995).
86 Inequality, Grievances, and Civil War

For obvious reasons, such feelings of underrepresentation are unlikely to

apply to groups playing senior roles in governmental coalitions. Besides groups
in dominant or monopoly positions, senior partners are least likely to engage
in conflict at any rate. When violence does arise, however, the source of con-
flict is usually a general lack of trust in government. In particular, the fear
of being downgraded by other senior partners, particularly those who domi-
nate the armed forces, can prompt representatives of senior groups to opt for
early secession rather than risking a status reversal. Despite their formal equal
rights within the Yugoslav Federation, the Slovenes and Croats anxiously fol-
lowed, and eventually experienced first-hand, Milošević’s anti-federal, central-
izing campaign aimed at boosting Serb influence at the expense of other nation-
alities. Following the revocation of Kosovo’s autonomy in 1988, “Milošević
and his allies sought to impose their conservative line on the country as a
whole by attempting to undermine the autonomy of the other federal units”
(Gagnon 2004, p. 80). Needless to say, these developments generated fear that
ultimately undermined the support of moderates in the other republics and
tipped the balance in favor of secession in Slovenia and Croatia (Gagnon 2004,
p. 81).14
Having investigated the support for H4.1 at some length, we now turn to
H4.2, which says that downgraded groups should be especially susceptible to
civil violence. Our data set contains 20 cases of downgrading followed by
conflict. Although this is a relatively small number (and downgrading itself is
a rare event), our statistical analysis shows that downgrading entails a much
higher risk of conflict (see Figure 4.4 and Table 4.3).
The events leading up to the aforementioned secessionist war of the Igbo
against the North in 1967 correspond closely to the conflict-fueling logic of
downgrading. In fact, the path to war featured two status shifts. There was
initially a shift to perceived Igbo dominance following a coup staged by Igbo
army officers in 1966. After losing power to the Northern Hausa-Fulani groups
through a countercoup in 1966, however, the Igbo were subjected to intense
discrimination that also included massacres, which convinced the group’s elite
to launch an ultimately ill-fated and violent secessionist bid, which resulted in
the Biafran War. Although political economists find it tempting to account for
this war in terms of natural resource dependence, most importantly evidenced
by the occurrence of oil within Igbo territory, the main logic according to our
interpretation stresses a power struggle among ethnic groups, as explicated
above. Zinn (2005) likewise stresses the role of coups in changing the political
status of different groups as central for conflict.15

14 Moreover, there is evidence that this type of fear of future status loss and possible exclusion
also extends to junior partners in power-sharing arrangements (see the Online Appendix).
15 Zinn (2005) shows convincingly that the attempt to capture ethnic causes of war through blunt
ethno-demographic measures, such as Collier and Hoeffler’s (2004) “domination,” which is a
dummy variable indicating if the largest ethnic groups make up 45 to 90 percent of the total
Political Exclusion and Civil War 87

Another example of downgrading triggering conflict is the insurgent attack

launched by the Lari-Bakongo against the incumbent government of the Repub-
lic of Congo in 1998. This was a response to the dominance by the previously
excluded Mbochis’ having ousted the other groups in June 1997 (Clark 2008).
As in Nigeria, the sudden change of fortunes of competing ethnic groups was
the main driver of violence. Similar patterns can be observed in other African
countries such as Cameroon and Chad.
Another much publicized reaction to a dramatic status reversal occurred
in Iraq after the fall of Saddam Hussein. The Sunni loss of power through the
fall of the dictator in 2003 inspired armed resistance targeting the Shia-Kurdish
governing coalition, as well as invading forces led by the United States, from the
following year. Although the United States has tried to encourage cooperation
between the different ethnic groups, the December 2011 arrest order for Iraqi
vice president Hashemi, a prominent Sunni Arab politician, was seen by many
as a move to reassert Shia dominance under Prime Minister Maliki. Further
moves in this direction have a high potential for reigniting conflict.
In sum, even a superficial exploration of the conflict onsets following down-
grading indicates that the reaction to such traumatic events generates espe-
cially intense grievances that increase the risk of conflict outbreak. A sense of
unfairly lost entitlement and the lust for revenge fuel such conflict processes,
which sometimes even lead to sequences of power shifts through coups and
countercoups.
Our statistical analysis of group size confirmed that populous, excluded
groups are overrepresented in the conflict statistics, as postulated by H4.3.
While the hypothesis was tested with a continuous group size variable encom-
passing the entire size spectrum, we here focus on the instances of very large
excluded groups engaged in conflict as highlighted by our analysis. Onsets
where the excluded group is larger than the ethnic group(s) in power are of
particular interest since such settings should bring forth intense grievances in
response to the “scandal” of rampant underrepresentation.
Such a focus leads us to revisit some of the cases that involve brutal discrim-
ination that were discussed at the beginning of this section. So much has been
said and published about the Apartheid regime that we hardly need to produce
detailed evidence of how the demographic lopsidedness contributed to its lack
of legitimacy (for a summary of the conflict, see Rubenzer 2007). In addition to
the Apartheid elite’s blatant racism, the small population share of the Afrikan-
ers compared to the excluded and discriminated masses turned the regime into
an icon of colonial-style oppression. The Liberian and Rhodesian [Zimbab-
wean] conflicts ending minority rule also fall into this category. Beyond that,

population. Calling for a more explicitly power-sensitive measure of political dominance, in

line with the EPR measures of political status, she observes that “[m]inority groups could also
control the system, as occurred after the January 1966 coup, or there could be institutionalized
power sharing, akin to Nigeria’s coalition government of the early 1960s.”
88 Inequality, Grievances, and Civil War

the Middle East is, and has been, home to three other dramatic cases of repres-
sive ethnocratic minority regimes, namely Saddam Hussein’s Sunni-dominated
rule, the Sunni leadership in Bahrain, as well as the Assad family’s Alawite dom-
ination of Syria. While the former triggered a series of Shiite revolts in 1982,
1987, and 1991 that were all brutally suppressed by the Iraqi military, Syria
only shows up in our conflict data in connection with the Sunni-led challenge
to power in 1979, and Bahrain not at all. At the moment of writing, however, a
civil war is currently raging that pits Assad’s murderous regime against the other
ethnic groups. Furthermore, Bahrain also experienced a violent revolt against
the Sunni power monopoly in 2011, which is beyond the current coverage
of the EPR-ETH data. However, these conflicts constitute clear out-of-sample
predictions of high conflict risk based on our theory, and our estimation sample
does not include any information from the period of the “Arab Spring” in 2011.
We now turn to possible anomalies. Hypothesis H4.3 would be undermined
by the presence of large, excluded groups that have remained peaceful. To
explore this issue, we consider those large and excluded groups that never
saw conflict at any point during our sample period from 1946 through 2009.
We found about two dozen cases of excluded groups that are larger than the
included groups. Encouragingly, it was possible to account for these potential
anomalies relatively easily in most cases. In one case, namely Bahrain, a Shiite
rebellion erupted in 2011 targeting Sunni minority rule (see above). In many
other instances, we see clear evidence of political violence, such as coups and
electoral violence, below the 25 battle deaths threshold used by the Uppsala
Conflict Data Program, UCDP (e.g., the coup in Guinea-Bissau in 1980, the
Indo-Guyanese in Guyana, and the East Indians in Trinidad and Tobago). One
important case of an apparent non-conflict, the Palestinian Arabs in Jordan,
turns out to include violent conflict, as evidenced by the Black September
conflict in 1970 when the government cracked down on Palestinian refugees
and the Palestinians fought back.16 Another category comprises ethnic groups
that did not make it into our strict coding of onset because there was no
rebel organization advancing aims in their name despite there being significant
recruitment along ethnic lines. Examples include the Mbochi in the Republic
of Congo, the Teso in Uganda and the Hill Brahmins and Madhesi in Nepal,
and the Sharchops in Bhutan.
Still, some puzzling “non-barking dogs” remain. The relatively wealthy
Hausa in Niger were excluded during long periods. However, they dominated
the economic life of the country during the authoritarian regimes in place from
1960 to 1990 in which the Djerma-Songhai group very much controlled polit-
ical power (Ibrahim 1994; Lund 2001). Transborder trade with their kinship
in Nigeria helped the Hausa counterbalance the negative consequences of their

16 This event is not included in the UCDP/PRIO Armed Conflict Data Set as it does not fit their
criteria for incompatibilities, as the Palestinians neither demanded territory nor control over
the government (see Salehyan and Gleditsch 2006, pp. 344–5).
Political Exclusion and Civil War 89

political exclusion. This may explain why ethnic conflicts could be avoided in
Niger during that long period (Vogt 2007, pp. 38–9). In the Sudan, finally,
some of the northern Arab groups have remained both excluded and peaceful.
This could possibly be attributed to their unwillingness to get embroiled in the
larger north-south conflict that the country has suffered for decades.
Our fourth and final hypothesis, H4.4, stipulates that previous conflict
instances make new outbreaks more probable. A look at the data reveals
that the Iranian Kurds have as many as six onsets, followed by the Somali
in Ethiopia with five, and the Iraqi Kurds, the Naga in India, and Karenni in
Myanmar with four onsets. Extending the scope to three onsets, implying two
previous conflicts, adds more conflict groups to the list, including several in
Myanmar and India, as well as the Papuans and the East Timorese in Indone-
sia. These cases certainly exhibited especially harsh discrimination imposed
by uncompromising governments. Although there can be no doubt about the
reality of grievances in these cases, discrimination can be found in other con-
flicts as well, as we have seen at the beginning of this section. In addition to
persistent discrimination, and our expectation that revenge is a major driver
of recurrent strife, conflict recurrence appears to stem from state weakness
and group remoteness. Having to cope with several separatist challenges at the
same time, as well as political turmoil at the center and interstate disputes with
neighboring countries, some states such as Iran, India, Myanmar, and Ethiopia
have been unable to sustain a constant pressure on their minorities. As a result,
resistance flares up whenever the central government is ready for a new round
of repression or succumbs to a period of weakness that can be exploited by
peripheral rebel groups.
We end this section on the validity of our mechanisms by returning to a
major threat to causal inference, which we introduced in Chapter 3. As stressed
by Blattman and Miguel (2010), problems of endogeneity continue to haunt
quantitative studies of civil war. It is clear that our main explanatory variable,
ethnic groups’ power access, is not randomly assigned and could therefore be
at least partially dependent on our outcome variable, conflict onset (Fearon
2011; Fearon and Laitin 2011). As pointed out by Fearon (2011), such a bias
could emerge because of two reasons. First, it is possible that our expert coders
have overstated the extent to which groups are excluded or discriminated
in cases that involve armed conflict, while underestimating the same coding
for relatively peaceful groups. However, while this is certainly possible, the
coding of EPR-ETH was carefully monitored for this eventuality. Ultimately,
the burden of proof is on the critics to show that this is the case, and the data
set can be corrected should detailed criticism specifying erroneous or biased
coding be brought to our attention, which has not happened so far.17

17 For example, Fearon (2011, p. 19) criticizes the EPR data set explicitly: “Countries where there
has been no ethnic conflict and where ethnic relations have been calm are for that reason judged
to have a low value of ‘exclusion.’” Yet, he fails to provide evidence of specific cases where the
coding can be said to be biased in this manner.
90 Inequality, Grievances, and Civil War

Yet, the second objection has to be taken much more seriously. Indeed, in
what Fearon dubs “policy regressions,” we have to be worried that what is
treated as an independent variable is actually a governmental policy chosen in
anticipation of future outcomes that relate directly to the dependent variable
(Fearon 2011). This type of bias could lead to either underestimation or overes-
timation of the effect. If opportunistic governments tune the level of exclusion
to “what they can get away with,” thus aiming at primarily less threatening
groups, the causal effect of the policy variable will be underestimated. As such,
this would therefore hardly pose any threat to our findings. Such a pragmatic
approach is invoked by representatives of dominant ethnic groups who prefer
to keep as much of the state’s resources as possible, while at the same time being
willing to make concessions in the name of interethnic peace. This sensitivity to
power relations could be motivated by ethnic nepotism and more ideologically
motivated strategies of ethnic nationalism.
The opposite applies to situations where the government tries to preempt
anticipated conflict by excluding threatening groups for security reasons. Such
precautionary measures could be implemented in order to prevent ethnic com-
petitors from staging coups from within the government. Applying this argu-
ment to Sub-Saharan Africa, Roessler (2011, p. 313) suggests that “ethnic
exclusion serves as an expedient mechanism to eradicate perceived enemies.”
Incumbent leaders may even prefer waging low-intensity wars in peripheral
parts of the country rather than running the risk of being backstabbed by eth-
nic competitors within the governing coalition. As opposed to the previous
scenario, this logic would undermine our statistical conclusions because this
case of reverse causation biases the estimated effect of exclusion on conflict
upward, thus undermining causal inference.
Because neither of these two possibilities is inherently implausible, it is noto-
riously difficult to establish which, if any, of them applies. For sure, one should
not exaggerate the extent to which the decision to exclude groups reflects
a quickly changing governmental policy. Quite on the contrary, as demon-
strated by the cases of South African Apartheid and Israeli ethnonational-
ist exclusion of the Palestinians, many instances of political exclusion and
discrimination stem from long-standing decisions that are primarily ideolog-
ical and only indirectly security related. Where the level of political access
does change, however, the effect is very much compatible with our grievance-
based account. Indeed, the strong conflict-fueling effect associated with the
downgrading variable strengthens our belief that our findings are not due
to endogeneity, especially in cases where power reversals can be attributed
to external shocks, as experienced by Sunnis and Shiites in Iraq after the
fall of Saddam Hussein, in the wake of the United States–led invasion in
2003.
This said, it may actually be possible to arrive at a more conclusive answer
as regards the two possible directions of reverse causation. Despite the diffi-
culties of finding an exogenous measure of ethnic groups’ power access that
Political Exclusion and Civil War 91

is not related to conflict, there are historical situations that can be exploited
to construct an instrumental variable that fulfills this function. Focusing on
internal conflict in post-colonial states, Wucherpfennig, Hunziker, and Ceder-
man (2012) come to the conclusion that the distorting impact of endogeneity
has been overstated. By exploiting systematic differences between the strate-
gies of the French and British colonial empires toward ethnic groups, this study
instruments for exclusion of each group at the moment of independence. As
opposed to the French colonies where ethnic groups’ power access mostly fol-
lows a straightforward center-periphery logic privileging groups at the coast,
the British were much more likely to apply a strategy of “selective indirect
rule” that allowed peripheral groups to have a much higher influence than in
the French cases. Based on the results from an analysis using this instrument
for exclusion, we conclude that our current, uncorrected analysis if anything
tends to underestimate the effect of exclusion on conflict onset.

Conclusion
This chapter represents the first step in our empirical evaluation of the postu-
lated link between inequality and civil war onset. As opposed to highly aggre-
gate studies at the country level and research focusing on micro-dynamics, it
does so by investigating causal mechanisms at the group level. Rather than
investigating grievances directly, we have employed an indirect method that
establishes whether ethno-political exclusion of ethnic groups is more likely to
trigger conflict than less-exclusive regimes. The EPR-ETH data set allows us
to capture groups’ access of power around the world and during specific time
periods.
Our statistical analysis shows that the less power access a group has, the
more likely it is to fight the government (H4.1). This effect becomes even
stronger should the group recently have experienced a power reversal (H4.2).
As shown in the previous section, a large number of cases confirm the link
between impaired power access and conflict-proneness, ranging from groups
reacting to discrimination and exclusion in narrowly ethnonationalist regimes
to less ideologically driven conflicts.
We also found strong evidence of a demographic size effect, whereby popu-
lous excluded groups are overrepresented in the conflict statistics (H4.3). This
is an important result, because it suggests that it is not primarily group size
itself, viewed as a proxy for the group’s resources, but the political framing
of the demographic relationship that matters. This strengthens our evidence
that conflict is to a large extent driven by grievances, because the perceived
injustice should increase with the size of the excluded population, rather than
with group size in general.
Finally, the findings confirm that groups that have already been involved
in previous conflict are more likely to see new outbreaks of violence (H4.4).
However, in this case, it is less obvious that the grievance logic acts as the main
92 Inequality, Grievances, and Civil War

engine of conflict. Rather, a second look at the cases of recurrent conflict points
in the direction of limited territorial state reach.
In the previous section, we also considered the reasons why causal inference
may be less robust, most importantly the threat of endogeneity. Although such
concerns cannot be entirely eliminated, we believe that our analysis rests on firm
ground given that some of the possible biases actually strengthen our results
and short-term changes in our independent variable through downgrading fuel
conflict. We refer the reader to the Online Appendix for further details on this
issue and other matters relating to the robustness of the results.
These findings are of considerable theoretical importance, because they cast
doubt on those studies that deny or downplay the centrality of ethnic grievances
in conflict processes. We argue that the main reason why the grievance-skeptics
have failed to find evidence of such mechanisms relates to their tendency to use
blunt demographic proxies that have very little to do with political aspects of
ethnicity. By shifting the theoretical stress from ethnicity to ethnic nationalism,
and from merely cognitive to motivational mechanisms, we propose a theo-
retical framework that puts ethnicity in its proper political context. Yet, our
rendering of horizontal inequality does justice to only one dimension, even if
perhaps the most important one. In the next chapter, we broaden the concep-
tual scope of our empirical evaluation of grievance-related conflict mechanisms
by considering economic inequalities between ethnic groups.
5

Economic Inequality and Civil War

Horizontal inequality concerns more than political power. Following Stewart

(2008a), we have conceptualized it as a multidimensional concept in Chapter
3. Indeed, most studies of inequality evaluate the consequences of differences in
wealth and income rather than in terms of political influence. As a complement
to the previous chapter, we therefore continue our investigation of group-level
effects by considering the impact of economic horizontal inequality.1
Again, we are up against what is commonly perceived as a nonresult. As was
shown in Chapter 2, the contemporary quantitative literature on civil war finds
no evidence for a link between economic inequality and internal conflict. Based
on indicators measuring individual wealth distributions, these studies find no
statistically distinguishable relationship to internal conflict (e.g., Fearon and
Laitin 2003; Collier and Hoeffler 2004). Yet, inequality continues to occupy a
prominent place in the qualitative literature on civil wars and has repeatedly
been linked to conflict processes (Wood 2003; Sambanis 2005, p. 323; Boix
2008; Stewart 2008a). Earlier research relying on the Minorities at Risk data
link found that economically discriminated groups were more likely to mobilize
or protest (see, e.g., Gurr 1993a; Goldstone et al. 2010). However, this line
of research has tended to look at measures of political instability broader than
civil war. Their claims about groups have been largely ignored by much of
the conventional civil war literature, which has focused on countries rather
than groups. Moreover, in the last few years, some quantitative studies of
civil war have started to appear arguing that the current literature’s failure to
connect distributional asymmetries to conflict behavior may actually be due
to inappropriate conceptualization and imperfect measurements, rather than

1 This chapter draws heavily on the approach developed in Cederman, Weidmann, and Gled-
itsch (2011). We are grateful to Nils B. Weidmann for major conceptual and methodological
contributions.

93
94 Inequality, Grievances, and Civil War

reflecting a fundamental absence of any causal effect (Østby 2008b; see also
Cramer 2006; Stewart 2008a).
Following up the main theoretical focus on horizontal inequality, this chap-
ter joins these recent contributions in shifting the explanatory focus from indi-
vidualist to group-level accounts of inequality and conflict. Because formidable
problems of data availability associated with the uneven coverage and compa-
rability of surveys have stood in the way of assessing such differences, most
scholars have had to contend themselves with selective case studies or statistical
samples restricted to particular world regions.
In order to overcome these difficulties, we combine our newly geo-coded
data on politically relevant ethnic groups’ settlement areas with Nordhaus’
(2006) spatial wealth measures, both with global coverage. Based on this novel
strategy, we present a worldwide comparison of economic horizontal inequality
and ethnonationalist civil wars. Building directly on the previous chapter, we
show that both backward, and to some extent also advanced, ethnic groups
are more likely to experience such conflict than those groups whose wealth lies
closer to the national average. Moreover, in agreement with Stewart’s broad
conception of horizontal inequalities, we find that economic inequalities trigger
conflict primarily for excluded groups.

Previous Literature on Economic Inequality and Conflict

In Chapter 2, we offered a general survey of the literature on inequality and con-
flict. We showed that the classical literature dating back to the 1960s and 1970s
took a keen interest in the effects of economic inequality. This is hardly surpris-
ing, since the ideological battle between East and West during the Cold War
revolved around class-based, rather than ethnic, politics. Much of this debate
centered on the Third World, where insurgent activities, partly in connection
with decolonization, drew inspiration from Marxist writings and attracted
considerable Soviet support.
Marx’s theory obviously covered much more than political violence. Indeed,
this aspect does not play an important role in the overall theoretical edifice
other than as a part of his notion of socialist revolution. Within this context,
rebellion is viewed as a class-based reaction to sustained capitalist exploitation.
As ever larger numbers of workers are exploited, the theory predicts that capital
owners’ attempt to oppress the masses will give rise to collective grievances that
finally trigger a revolt of the working class. In this sense, “the expropriators
are expropriated” (Boswell and Dixon 1993).
Although the original formulation of these ideas in Marx and Engels’
writings concerned industrial societies, their more recent intellectual follow-
ers extended this intellectual heritage to encompass peasant revolts in the
Third World. After all, Russia, the most important country to come under
the influence of these ideas. was predominantly rural at the moment of the
triggering revolution. As liberation ideologies spread in the Third World after
Economic Inequality and Civil War 95

World War II, the Western academic literature responded to the challenge of
analyzing peasant rebellions from a neo-Marxist angle. For example, Paige
(1975) theorized such revolts as responses to zero-sum situations that pit land-
lords against rural workers. In his classical study of the Moral Economy of
the Peasant, Scott (1976) applied anthropological methods to uncover how a
breakdown in traditional patron-client relations triggers revolutionary condi-
tions. In a more recent study, Booth (1991) argues that persistent inequality and
exploitation of peasants by rich landowners in Central America triggered rev-
olutionary challenges to incumbent regimes in the 1970s and 1980s. Focusing
on conflict during this period, Booth (1991, p. 34) claims that
economic development trends worsened the region’s historically extreme maldistribu-
tion of wealth and income, intensifying grievances among negatively affected class
groups. . . . Such problems led the aggrieved to demand change and sparked growing
opposition to incumbent regimes by political parties, labor unions, religious commu-
nity organizers, and revolutionary groups. Violent repression of opposition demands
for reform . . . not only failed to suppress mobilization for change but actually helped
forge revolutionary coalitions that fought for control of the state.

In this account, we can identify a distinctive causal chain starting with persistent
inequality leading to grievances among the peasant population, and fueling
demands for political change and redistribution. Denied such reforms, and
possibly even encountering state-led repression, the aggrieved are more likely
to see little choice but to rebel.
An even more recent contribution to this literature has been offered by Boix
(2008), who refines the standard argument about inequality and conflict by
considering the impact of factor mobility. According to his logic, conflict is
likely only in those cases where inequality relates to immobile resources, since
wealthy elites are unable to move their wealth abroad should political change
threaten their assets. Relying on structural measures of landownership rather
than comparisons of income levels directly, Boix reports strong support for
a link between wealth differentials and conflict behavior. Likewise, influen-
tial formal politico-economic models that take classes or social interests as
actors, such as Acemoglu and Robinson (2005), postulate a strong relationship
between income distributions, policy preferences, and incentives for violent
revolution.
However, these studies stand out as exceptions to the anti-grievance trend
that we documented in Chapter 2. Generally, students of opportunity struc-
tures and resource mobilization have been critical of positing motivations as
explanations of conflict onset, a criticism that also pertains to inequality:
The basic assumption of these theories is that the sources of discontent that lead to
violent political conflict are inherent in all societies and that the occurrence of political
violence is a function of political opportunities and constraints of the immediate political
environment, rather than variations in levels of economic inequality or the intensity of
economic discontent (Schock 1996, p. 104).
96 Inequality, Grievances, and Civil War

As we have also found, most prominent studies of civil war have come to
similar conclusions. Whether inspired by Marxist or liberal principles, most
scholarship on inequality and violence adopts a class-based or individualist
perspective, respectively. Yet, another stream of research emerged that stresses
skewed wealth distributions that may stem from, or interact with, ethnic cleav-
ages. Broadly compatible with Tilly’s categorical inequalities and Stewart’s
horizontal inequalities, this strand of thinking highlights inequalities between
identity groups that are demarcated in other ways than through wealth. In
particular, ethnicity serves as the most important criterion of boundary forma-
tion. Whereas both Marxist and Liberal scholars expected ethnic differences
to fade away through “modernization,” students of ethnicity were more con-
vinced of the staying power of ethnic relations (Schock 1996; Cederman 1997,
ch. 7).
Early contributions to this research stream were inspired by Marxist ideas
of exploitation and accorded economic processes a central role. As we have
seen in Chapter 2, Hechter (1975) proposed a theory of internal colonialism
that explains how ethnic cores come to dominate the territorial peripheries,
such as Wales and Scotland. In this perspective, political dominance leads to
a cultural division of labor that allocates low-status jobs to members of the
subordinate ethnic groups (Hechter 1978). As later argued by Tilly (1999),
political entrepreneurs can use ethnic boundaries for distributional purposes in
order to exploit and hoard wealth for the ingroup. High levels of economic dis-
crimination tend to produce reactive identity formation and stronger solidarity
within the exploited groups, which in turn tends to spill over into political
mobilization and possibly violent revolts.
From a less-materialist perspective that privileges socio-psychological mech-
anisms such as group apprehension, Horowitz (1985) proposes an explanation
of secessionism among “backward” and “advanced groups.” As does Hechter’s
theory of internal colonialism, Horowitz expects less advanced groups to be
the most likely to rebel. Having little to lose, these groups are highly moti-
vated by economic grievances and thus less sensitive to the potentially high
costs associated with separatist campaigns. However, advanced groups may
also opt for exiting the state to the extent that they are unhappy with the
revenue-expenditure balance. We will return to the topic of how inequalities
in both directions can trigger violence. All in all, Horowitz’s findings derive
from his erudition and detailed knowledge of a large number of cases, but his
propositions are never subjected to systematic evaluation.
Relying on his Minorities at Risk (MAR) data set, Gurr (1993a; 2000b)
exposes similar reasoning to quantitative tests. Earlier versions of the MAR
data set offer expert assessments of the level of “economic discrimination”
but no systematic economic data. Based on these measurements, Gurr comes
to the conclusion that economic disadvantages, as well as political ones,
fuel the risk of conflict: “Economic disadvantages, especially those associated
with discrimination and poverty, are consistently correlated with economic
Economic Inequality and Civil War 97

and social grievances and with demands for greater political rights” (1993b,
p. 188).
Other recent research attempts use new data sources to evaluate the wealth of
ethnic groups.2 Recognizing the difficulties of measuring horizontal inequalities
based on explicit economic data, research emanating from Stewart’s research
group has so far primarily relied on individual country case studies or a compar-
ison of a few countries rather than large N comparisons (see, e.g., the Stewart
2008b edited volume). The picture that emerges from this research suggests that
both disadvantaged and advanced groups are more often involved in internal
conflict than groups closer to the country average (Stewart 2008a). Yet, some
quantitative studies have attempted to generalize to a larger set of countries.
Relying on household surveys conducted in 39 developing countries, Østby
(2008a) finds evidence that social horizontal inequality is associated with civil
war, although the economic dimension appears to be weaker (see also Østby
2008b). In a follow-up study based on geo-coded conflict and survey data from
Sub-Saharan Africa, Østby, Nordås, and Rød (2009) reach firmer conclusions,
showing that both economic and social group-level differences appear to drive
conflict behavior.

Deriving Hypotheses on Economic Horizontal Inequalities

and Civil War
We now turn to our own account of economic inequality and conflict. What
are the observable implications of our analytical framework? Building directly
on the derivation of causal mechanisms that we introduced in Chapter 3, we
develop straightforward hypotheses. The first, and most obvious, expectation
is a positive effect of economic horizontal inequalities on civil war onset even
when we control for political horizontal inequality. If the causal chain operates
as we have postulated, there should be a statistically discernible signal indi-
cating that ethnic groups with per capita income far from the country average
have a higher risk of experiencing conflict.
Before stating our hypotheses, however, we need to differentiate between
the conflict proneness of advanced and backward groups. As Horowitz (1985)
explains, arguments can be advanced for both types of economic horizontal
inequalities leading to a higher risk of conflict. Poorer groups, especially those
residing in backward and peripheral regions, often desire to break away from
the core of their countries regardless of the cost because they perceive them-
selves to be systematically disadvantaged compared to their wealthier com-
patriots in terms of economic development and distribution of public goods.
Yet, perceptions of disadvantage also characterize members of some relatively
wealthy groups, especially if they feel that state-level redistribution denies them

2 In a pioneering statistical test, Barrows (1976) detected an influence of group-level differences

on conflict in Sub-Saharan Africa.
98 Inequality, Grievances, and Civil War

the fruits of their success: “Advantaged regions usually generate more income
and contribute more revenue to the treasury of the undivided state than they
receive. They believe that they are subsidizing poorer regions” (Horowitz 1985,
pp. 249–50). Since these groups have more to lose, and are sometimes demo-
graphically represented outside their original settlement area, they are typically
more cost sensitive as regards secession. However, such cases do occur, as illus-
trated by the Basques in Spain and the Slovenes in the former Yugoslavia (cf.
Gourevitch 1979).
Nevertheless, there is no reason to assume that the effect of group inequality
is perfectly symmetric around relative equality. Remaining agnostic as to the
relative impact of horizontal inequalities in either direction, we therefore submit
these arguments to separate tests:

H5.1. Relatively poor ethnic groups are more likely to experience civil war than
those that are closer to the country average.
H5.2. Relatively wealthy ethnic groups are more likely to experience civil war than
those that are closer to the country average.

As we have seen, an uneven wealth distribution is not the only possible type
of structural asymmetry. Drawing on Stewart’s multidimensional conceptual-
ization of horizontal inequalities, we hypothesize that both economic and polit-
ical horizontal inequalities contribute jointly to the outbreak of civil war. Even
when controlling for political horizontal inequalities, such as groups’ exclusion
from political power, income inequalities among ethnic groups should increase
the risk of civil war. Indeed, Cederman, Weidmann, and Gleditsch (2011) find
exactly such an independent impact of the two dimensions. It goes without
saying that groups disadvantaged in both ways would be more exposed to the
risk of onset than those that suffer from horizontal inequality in one dimension
only.
Some scholars argue that there may even be an interaction effect rather than
a merely additive relationship between these two types of horizontal inequality.
Echoing Hechter’s notion of cultural division of labor, this theoretical reasoning
also mirrors Stewart’s (2008a, p. 18) hypothesis that “political mobilization
is especially likely when there are consistent horizontal inequalities, that is
both political and economic horizontal inequalities run in the same direction.”
Based on statistical evidence from Sub-Saharan Africa, Østby (2008a) finds
support for a strong interaction effect between interregional asset inequality
and political exclusion. Case studies of Côte d’Ivoire (Langer 2005) and Nepal
(Murshed and Gates 2005) confirm this finding (see also Stewart, Brown, and
Langer 2008, pp. 289–90; cf. Hegre, Østby, and Raleigh 2009).
Although the consistency argument is intuitively appealing, we derive
an asymmetric dependency between the two dimensions. More specifically,
we postulate that economic horizontal inequality is conditional on political
Economic Inequality and Civil War 99

inequality. Rather than contributing symmetrically to the risk of conflict, polit-

ical exclusion takes causal precedence because the framing of inequalities in
terms of blame attribution should be much easier to justify if the group in
question is exposed to alien rule than in power-sharing arrangements. Whether
economic inferiority stems from political marginalization or other processes,
the presence of power asymmetries makes politicization much more likely.
The first case corresponds to governments’ actual favoritism of incumbent
groups at the extent of those outside the polity. As we have argued in Chap-
ter 4, politically dominant groups are inclined to favor their own kin not only in
terms of political opportunities, but also more broadly, as regards the distribu-
tion of both private and public goods. In extreme cases, the central government
may neglect the infrastructural needs of entire provinces, as illustrated by Khar-
toum’s lack of investments in Sudan’s South before the recent secession (Ali
et al. 2005). Moreover, discrimination in terms of business opportunities may
follow directly from ethnic nepotism, or indirectly from inferior educational
investments in the excluded groups’ populations thus reducing their chances
to perform economically. Under such circumstances, rebel leaders are able to
frame economic inequality along group lines as a result of political decisions
made by incumbent groups rather than as a matter of differential resource
endowments or skill levels. Of course, there are some examples, such as Saudi
Arabia (Byman and Green 1999) and Russia in the 1990s (Treisman 1999),
where the center has been able to buy off peripheral groups, but this strategy
has its limitations, most obviously because of its costs. Persistent inequality
without political influence will give rise to a sense of frustration with the exist-
ing political system that could trigger violent protest.
The second case pertains to situations where blocked power access is not
the main cause of underdevelopment, but discriminatory governmental policies
nevertheless can be blamed for the economic gap. For example, it is possible
that the peripheral geographic position disadvantages an ethnic group, or that
prior colonial or imperial regimes may have set development back so strongly
that the group in question enjoyed a much lower living standard even before
the current exclusionary government’s rise to power. Whether it is true or
not, governmental elites are usually blamed with the economic fate of entire
countries since they constitute convenient lightning rods for discontent. As
argued by social-psychologists, blame shifting and scape goating may be a
convenient strategy for the protection of a group’s self-esteem (Hechter 1978;
Horowitz 1985; see also Herrera 2005).
The contingent effect of economic horizontal inequalities should apply to
disadvantaged groups, thus inviting us to modify H5.1. We would not expect
such a dependency in case of economically superior groups. These groups are
already likely to have at least some shared power access, and even if they do
not, being granted more power is unlikely to solve the problem since their
grievances stem from having to share their wealth with other groups. In such
cases, the urge to exit the polity will be strong indeed, as illustrated by the
100 Inequality, Grievances, and Civil War

Slovenian and Croatian declarations of independence from Yugoslavia. Little

would have placated these groups short of almost total budgetary autonomy,
and it is even less probable that the less wealthy but politically powerful Serbs
would have granted them such concessions.
Taking into account the limited scope of the interaction effect, our final
hypothesis on economic horizontal inequality summarizes these theoretical
expectations:

H5.3. Economic inequality makes less wealthy, excluded groups more likely to
experience civil war.

Global Data on Horizontal Inequalities and Other Dimensions

Our theoretical expectations must now be confronted with empirical evidence.
As we have seen, data availability constitutes a major stumbling block in studies
of inequality and conflict. So far, virtually all existing statistical studies of hor-
izontal inequalities have used survey data on economic welfare by households,
as collected by the Demographic and Health Surveys (DHS) project.3 Although
these data offer a relatively direct measure of wellbeing and are therefore use-
ful as grievance indicators, the information source and survey approach are
associated with a number of practical limitations. The DHS project is limited
to a selection of primarily developing countries, relatively few countries that
experience conflict are covered, and surveys in conflict-affected countries tend
to cover only those areas not directly affected by violence. Moreover, infor-
mation on ethnic affiliation or the geographic location of households is only
available in some of the surveys.4 Although the surveys are constructed so as
to be nationally representative, the number of responses for an ethnic group or
location may often be very low, and there is no guarantee that the samples will
be representative. Finally, survey data are subject to a host of potential response
biases, both conscious and unconscious. For example, aggregate responses from
surveys may well be systematically biased against finding evidence of inequality
if poorer individuals overstate their assets and richer individuals consistently
understate theirs. In sum, survey data may be helpful for many purposes, but
the DHS data do not provide a plausible alternative for evaluating the role of
horizontal inequalities on a global basis, although they can serve as a useful
complement (see below).
Given these difficulties, it makes sense to consider spatial data sets as an
alternative to survey-based methods. The only broadly available cross-national

3 For these data, see http://www.measuredhs.com/.

4 See, e.g., Baldwin and Huber (2010), who draw on survey data from 46 countries to evaluate
the impact of ethnic diversity on public goods provision. As noted above, Østby (2008a; 2008b;
Østby et al. 2009) also uses survey data in her studies of horizontal inequalities and conflict in
Sub-Saharan Africa.
Economic Inequality and Civil War 101

data source on variation in wealth within countries is the G-Econ data set,
developed by Nordhaus (2006; see also Nordhaus and Chen 2009).5 The G-
Econ data set tries to assemble the best available data on local economic activ-
ity within countries for geographical grid cells, and convert these to compa-
rable purchasing power parity figures to allow for meaningful comparisons.
The resolution of the spatially explicit data set is one-degree square grid
cells. The data are constructed from a variety of sources, including regional
gross product data for the lowest available political subdivision, estimates of
regional income by industry, and estimates of rural population and agricultural
income. The specific methodologies differ by country and data availability.
The database has global coverage, but the temporal scope is limited to a single
year, namely 1990.6 We therefore restrict our analysis to the post–Cold War
period.7
Despite its relatively broad coverage, there are a number of disadvantages to
the Nordhaus data for testing propositions on horizontal inequalities. Many of
the classical cases of substantially wealthier-than-average ethnic groups such
as the Chinese in East Asian Countries or Indians in East Africa that some
researchers such as Chua (2003) have argued are particularly likely to become
involved in ethnic conflicts are dispersed. Since these groups lack a spatially
concentrated homeland or region, spatial data cannot be used to estimate rel-
ative wealth differences among groups. Any measure of the value of economic
production is strictly speaking a “flow” measure and hence an imperfect proxy
for the “stock” of wealth, although this criticism obviously applies with equal
force to national-level productivity measures. Since the quality varies consider-
ably across countries, the data are likely to understate the extent of inequality
in countries with poor data coverage. Indeed, in some countries the official data
may be of such poor quality that the variance is suppressed and the advantages
over survey reports may be questionable. We will return to these issues at the
end of this chapter.
Based on the G-Econ data, Buhaug et al. (2011) present the first global anal-
ysis of the relationship between spatial inequalities and the location of civil
conflict outbreaks. However, their research design focuses on local measures
of inequality across geographic grid cells and the specific location where con-
flict first breaks out and does not capture group-level participation or wealth
differences, which may extend far beyond the original onset location. Another
useful approach estimates the wealth of regional subunits of states (Sambanis

5 Another promising avenue is to use light emissions as a proxy for economic activity, see, e.g.,
Elvidge et al. (2009). Chen and Nordhaus (2011) report that this data source primarily is useful
for poor countries with low light emissions and where official statistics are especially poor.
6 G-Econ 2.2 provides separate estimates for gross cell products in 1995 and 2000. However, these
estimates simply adjust the 1990 estimates by updated population figures for 1995 and 2000. As
such, the 1995 and 2000 figures contain no independent economic data over the 1990 values.
7 Since it is well known that relative inequalities, as opposed to absolute measures of wealth, are
characterized by considerable stability and inertia over time, relative inequalities in 1990 seem
plausible proxies for prior group inequality ratios (Tilly 1999; Stewart et al. 2008).
102 Inequality, Grievances, and Civil War

and Milanovic 2011; Deiwiks, Cederman, and Gleditsch 2012), but these do
not always overlap with ethnic group settlement patterns, sometimes intention-
ally breaking up ethnic groups or making groups minorities in larger regions.
A more direct assessment of horizontal inequalities requires geo-coded data on
ethnic groups, and, in view of H5.2, also information about their access to
executive power.
In order to obtain spatial estimates of economic performance for EPR groups
based on the Nordhaus grid, we need information on their settlement areas or
region. Since this overlay operation requires data on the precise extent of these
regions rather than a simple textual description, existing data sets such as
Minorities at Risk are insufficient. We therefore rely on the GeoEPR data set,
a comprehensive geo-coded version of the EPR groups (Wucherpfennig et al.
2012). The most recent version of GeoEPR builds directly on the newly updated
nonspatial EPR-ETH database that offers improved coding with coverage from
1946 through 2009. In short, GeoEPR provides two types of information about
ethnic groups. First, for each group in EPR, the data set categorizes the type of
settlement pattern, distinguishing between regional, urban, and migrant groups
(plus mixed categories). For all groups with a regional base, GeoEPR represents
the settlement area of the group as a polygon (or a set of polygons, if there
are more than one distinct settlement cluster for a group in a country). In
contrast to earlier geocoding attempts, GeoEPR also tracks major changes in
the settlement pattern of a group over time, including those resulting from
ethnic cleansing.

Deriving Operational Measures

The G-Econ data allow for deriving ethnic group-specific measures of wealth
by overlaying polygons indicating group settlement areas with the cells in the
Nordhaus data. Dividing the total sum of the economic production in the
settlement area by the group’s population size from the spatial data enables
us to derive group-specific measures of per-capita economic production, which
can be compared to the nationwide per capita product.
A visual illustration helps explain the estimation of group GDP per capita
from the Nordhaus data. Figure 5.1 shows the G-Econ estimates as grey cells,
where darker shadings indicate wealthier cells. The map shows the spatial vari-
ation in wealth across the Yugoslav federation in 1990, based on the CShapes
data set for the historical boundaries (Weidmann, Kuse, and Gleditsch 2010).
Relatively wealthy pockets appear primarily in the northwest of the country,
in the constituent republics of Slovenia and Croatia. Compared to the other
parts of the country, Serbia shows up as a generally poor region.
Together with the settlement areas of GeoEPR, the G-Econ data can now be
used to estimate group wealth spatially. Figure 5.1 also shows the settlement
regions for the Slovenes and the Albanians. Using similar techniques as those
pioneered in Buhaug et al. (2008) and Cederman, Buhaug, and Rød (2009), we
Economic Inequality and Civil War 103

Slovenes

Albanians

Gross Cell Product

under 1.94
1.94 − 3.88
3.88 − 5.82
5.82 − 7.76
over 7.76
figure 5.1. G-Econ cells for Yugoslavia overlaid with GeoEPR group polygons for
Slovenes and Albanians.

derive an indicator of group wealth by summing up the (population-weighted)

proportions of the Nordhaus cells covered by a group.8 For example, as a result
of this procedure, the Slovenes get a high score, since their settlement region
is located in the rich parts of Yugoslavia. Measuring horizontal inequality as
the ratio of the group’s GDP per capita estimate divided by the average value
for the entire country, Figure 5.2 shows the result of this computation for
Yugoslavia, depicting wealthier groups in darker shades and poorer ones in

8 This spatial aggregation process retrieves all the G-Econ cells that are covered by a group
polygon and computes the total wealth estimate as the sum of the cell values. However, in a
number of cases cells do not align perfectly with group polygons, and there is only a partial
overlap between a cell and a group polygon. For these cases, only the overlapping area’s wealth
should enter the group wealth computation. We estimate the wealth of a partial G-Econ cell by
distributing its total value as given in the data set to much finer cells of 2.5 arc-minutes (approx.
5 × 5 km, or 1/576 of the area of a G-Econ cell). This distribution is population weighted, that
is, it assumes that wealth is proportional to the number of people in each of the smaller cells.
This weighting was done using the Gridded Population of World (GPW) data set (version 3,
available at http://sedac.ciesin.columbia.edu/gpw/).
104 Inequality, Grievances, and Civil War

Slovenes

Croats

Serbs
Bosniaks

Montenegrins

Albanians

Inequality
Macedonians
under 0.8
0.8 − 0.95
0.95 − 1.05
1.05 − 1.2
over 1.2
figure 5.2. Result of spatial wealth estimation for groups in Yugoslavia.

lighter shades. Slovenes and Croats receive high scores, but the opposite is
true for the Albanians in Kosovo, which are among the poorest groups in the
country.
As a further illustration, Figure 5.3 shows the same information for the
Sudan. Unsurprisingly, the southern and western groups, the latter including the
Fur, emerge as the most impoverished in that state. Extending the comparison to
Myanmar, we also illustrate the limitations of our spatial approach (see Figure
5.4). Despite considerable wealth discrepancies between peripheral and central
areas, the Nordhaus data exhibit very limited variation due to underlying data
quality issues. To the extent that similar measurement problems afflict other
country cases that experienced conflict, we can expect the effect of inequality
to be underestimated by our study.
As explained by Mancini, Stewart, and Brown (2008), there are many dif-
ferent ways to operationalize horizontal inequalities, most of which apply to
entire countries. In this chapter, we use a straightforward group-level measure
of income inequality. Letting yg denote the GDP per capita of the ethnic group,
and yc the average GDP per capita of all groups in the country, we measure
Economic Inequality and Civil War 105

Beja

Fur

Nuba

Dinka
Nuer Shaygiyya, Ja'aliyyin/nand Danagla

Inequality
under 0.8
0.8 − 0.95
0.95 − 1.05
1.05 − 1.2
over 1.2
figure 5.3. Result of spatial wealth estimation for groups in the Sudan.

inequality asymmetrically with two variables that correspond to groups that

are poorer and wealthier than the country average, respectively:
low ratio = yc /yg if yg < yc ,
1 otherwise;

high ratio = yg /yc if yg > yc ,

1 otherwise.

This operationalization guarantees that deviations from the country mean

are always positive numbers above one.9 For example, a group that is twice as
wealthy as the average has low_ratio = 1 and high_ratio = 2, and a group that
is three times poorer has low_ratio = 3 and high_ratio = 1.
9 Note also that this specification differs from Cederman, Weidmann, and Gleditsch (2011), who
use zero rather than one as the default value for their asymmetric equality measures. The current
formula has the advantage of ruling out discontinuities between zero and one.
106 Inequality, Grievances, and Civil War

Kachins

Shan
Bamar/n(Barman)

Kayin (Karens)

Inequality
under 0.8
0.8 − 0.95
0.95 − 1.05
1.05 − 1.2
over 1.2
figure 5.4. Result of spatial wealth estimation for groups in Myanmar.

As regards all other variables, we refer the reader to the previous chapter.
Indeed, the current analysis relies on the same specification of the dependent
and independent variables, as well as the temporal controls.

Results
We are now ready to present the results. Given the limited temporal availability
of the inequality data (which refer to 1990), we restrict the sample to group
years after the Cold War, from 1991 through 2009. All groups represented in
GeoEPR are included.10 This leaves us with a total of 5,377 group years with
only 59 conflict onsets, which constitute a subset of the dependent variable used
in Chapter 4. Also in agreement with the previous analysis, we compensate for
country-level dependencies by estimating clustered standard errors. Because

10 We drop all dispersed groups, which normally cover the entire territory of a country, except in
those cases where dispersed dominant groups interact with minorities that cover only a part of
the country’s territory, as is the case with the Turks and Kurds in Turkey.
Economic Inequality and Civil War 107

of measurement difficulties, we reduce the sample by discarding all groups

that have a population of less than 500,000. Our spatial method becomes
unreliable for small population sizes primarily due to the low resolution of
the G-Econ data and the limited precision of the population estimates for tiny
groups.11
Table 5.1 presents the main results. Our starting point is Model 5.1, which
replicates Model 4.1 for the restricted sample as described above. The replica-
tion confirms that both power access (H4.1) and downgrading (H4.2) continue
to have a strong impact on the dependent variable. The other two hypothe-
ses, H4.3 and H4.4, receive weaker support for reasons related to the limited
sample. The effect of relative group size does not reach significance due to
the censoring of small groups. Likewise, the influence of previous conflicts
is blunted by the temporal sample restriction to the post–Cold War cases.
Finally, the control variables have the same signs as in Model 4.1, but the
variable measuring ongoing conflict fails to reach significance in the current
analysis.
We are now ready to test the first two of our new hypotheses (see Model
5.2). Still controlling for political exclusion, we introduce the two ratio
variables low_ratio and high_ratio. Whereas very little changes with regard
to political horizontal inequality, this extension of the analysis uncovers
an association between relative poverty and conflict, thus confirming H5.1.
Figure 5.5 provides a graphical illustration of this finding. The probability of
conflict onset in a given year increases steadily as a function of low_ratio. Com-
pared to parity, which corresponds to a relatively low probability of conflict of
0.004, groups that are three times poorer than the country average more than
double their conflict propensity to roughly 0.01. The error bands that mark
the 5th percentile are wide for high deviations from the country average due to
data scarcity, but the increase of the curve is very clear.
However, the conflict propensity of groups that are wealthier than the coun-
try average is even higher than for less wealthy groups, but the estimated effect
is only statistically distinguishable from zero at the level of p = 0.07. These
results suggest that groups that are wealthier than the country average may
have a higher likelihood of being involved in conflict, but there is only limited
support in the data for a clear trend, which may be due to both large variation
in conflict propensity and a limited number of groups notably wealthier than
the country average. Moreover, as previously mentioned, our results are unable
to consider horizontal inequalities involving ethnic groups that are not spatially
concentrated or discrete. All in all, we believe our results are consistent with a
tendency for wealthier groups to be more prone to see conflict, but we do not
have strong evidence for the robustness of H5.2.

11 This sample restriction drops a number of tiny groups, especially in China and Russia, for
which no reliable spatial estimate can be computed (see the Online Appendix).
108 Inequality, Grievances, and Civil War

table 5.1. Economic Horizontal Inequality and Group-Level Onset of Civil

War, 1991–2009

Model 5.1 Model 5.2 Model 5.3

Group-level variables:
Group Excluded 1.2411** 1.2730** 1.1371**
(0.3671) (0.3746) (0.3557)
Low Ratio 0.4736**
(0.1120)
High Ratio 0.6329
(0.3440)
Excluded × Low Ratio 0.5233**
(0.0944)
Excluded × High Ratio 0.5160
(0.2749)
Included × Low Ratio − 1.1450
(1.0853)
Included × High Ratio 0.8760
(0.8642)
Downgraded 1.9858** 1.9469** 1.9275**
(0.4497) (0.4562) (0.4559)
Rel. Group Size − 0.2000 − 0.0277 − 0.1441
(0.6348) (0.6721) (0.6467)
Number of Previous Conflicts 0.1784 0.2251 0.2306
(0.1581) (0.1795) (0.1758)
Country-level variables:
Ongoing Conflict, lag 0.3596 0.3701 0.3526
(0.4078) (0.4081) (0.4071)
GDP/capita, lag, log − 0.2334* − 0.3239* − 0.3224*
(0.1144) (0.1310) (0.1331)
Population, lag, log 0.0532 − 0.0039 − 0.0133
(0.1466) (0.1803) (0.1815)
Constant − 2.7347 − 2.9462 − 2.4067
(1.5729) (1.6779) (2.0276)
Observations 5,377 5,377 5,377
Robust standard errors in parentheses; estimates for peace-year correction not shown.
** p < 0.01, * p < 0.05.

The other variables behave similarly to the previous model. Again, we depict
graphically our findings. Figure 5.6 reveals that the predicted probability of
conflict increases considerably with increasing relative group wealth. Yet, as
we have discussed above, the error bands are extremely broad and almost
encompass zero.
Economic Inequality and Civil War 109

0.05
0.04
Predicted Probability of Conflict

0.03
0.02
0.01
0.00

1 2 3 4 5 6

Low Ratio
figure 5.5. Conflict probability for groups less wealthy than the country average.

Having considered H5.1 and H5.2, we now test H5.3, which postulates that
the causal impact of economic horizontal inequality hinges on power access.
Retaining the specification of the previous model, Model 5.3 introduces inter-
action terms that separate the consequences of economic inequality according
to whether the group in question is included or excluded. According to our
hypothesis, the less wealthy groups should be consistently overrepresented in
the conflict statistics. We find strong evidence that excluded and less wealthy
groups are more prone to conflict, and this is the only interaction term that
clearly reaches significance at conventional levels. However, we note that the
coefficient estimate for the interaction term between excluded and high ratio
is of similar magnitude as the interaction term with less wealthy groups and
reaches significance at 0.05 in a one-tailed test. This suggests that there is some
tendency for wealthier excluded groups to be more likely to be involved in con-
flict, even if the result may be less consistent than for relatively poorer groups.
We stress again that there are few cases of excluded wealthier groups. It may
even be the case that the greater resources for using force against the state that
wealthier excluded groups can muster could make such groups more likely
to receive accommodation and become included. For included groups we find
that both of the interaction terms are very far from significant, and the odds
110 Inequality, Grievances, and Civil War

0.05
0.04
Predicted Probability of Conflict

0.03
0.02
0.01
0.00

1 2 3 4 5 6

High Ratio
figure 5.6. Conflict probability for groups wealthier than the country average.

of conflict for included groups at any level of wealth are always lower than
those for excluded groups. For the control variables we again detect no major
changes compared to Model 4.2.
As stated in Chapter 4, interaction terms can be tricky to interpret based
on tabular results alone. Therefore we offer a graphical illustration of the
conditional logic implied by H5.3 (see Figure 5.7). The chart plots excluded
and included groups’ predicted probability of conflict against their relative
poverty compared to the country average. The upper curve, which belongs
to the marginalized groups, reflects a strongly positive effect that is clearly
separated from the conflict propensity of incumbent groups for the range of
inequality ratios up to about three. Beyond that level, the error bands become
so wide that the effects can no longer be clearly distinguished.
In order to bring out the difference between included and excluded groups
more clearly, we present in Figure 5.8 the first-order difference of conflict prob-
ability for excluded groups, using the included ones as the base of comparison.
Based on this analysis, we conclude that excluded groups are affected much
more strongly by inequality than included ones. In fact, the differential effect
is significant for the range moving from parity to ratios of relative underdevel-
opment amounting to a factor of almost 4.
Economic Inequality and Civil War 111

0.10
0.08
Pred cted Probab ty of Conf ct

0.06
0.04

Excluded Groups
0.02

Included Groups
0.00

1 2 3 4 5 6

Low Ratio
figure 5.7. Comparing the effect of relative poverty on the conflict probability of
included and excluded ethnic groups.

Validating the Causal Arguments

Parallel to the validation analysis in the previous chapter, this section offers a
brief discussion of selected conflict onsets that support our statistical results.
Again, we will evaluate each hypothesis in turn. There are a number of cases
consistent with hypothesis H5.1, which states that relatively poor groups are
more likely to rebel. Most spectacular are the two Chechen wars that broke
out in 1994 and 1999.
There can be no doubt about Chechnya’s abject poverty in comparison
with other parts of Russia. In fact, the Republic of Chechen-Ingushetia con-
stituted by far the poorest part of the country (Hale and Taagepera 2002).
However, in line with our theoretical expectations expressed in Chapter 3, it
was not Chechnya’s underdevelopment that directly triggered conflict. Rather,
it took a nationalist mobilization campaign orchestrated by ethnonationalist
entrepreneurs to bring about the rebellion (Herrera 2005; Giuliano 2011). To
be sure, the nationalists had plenty of historical grievances to work with. In fact,
large parts of the Chechen population had been deported to other parts of the
Soviet Union by Stalin’s regime after World War II under the pretext that they
had collaborated with the Nazi invaders. Although the Chechens were allowed
112 Inequality, Grievances, and Civil War

0.10
Change of Conf ct Probab ty (Exc uded Groups)

0.08
0.06
0.04
0.02
0.00

1 2 3 4 5 6

Low Ratio

figure 5.8. First-order difference between excluded and included groups.

to return to the Caucasus in the late 1950s, Stalinist repression had severely
disrupted the local economy and caused considerable hardship that worsened
the economic position of an already-underdeveloped region (Derluguian 2005).
Thus, there was a strong sense of nationalist grievance that lingered until the
1990s, despite Soviet transfers and development projects.
Among the ethnic groups in the Russian Federation, the Chechens are not
alone in exhibiting nationalist mobilization driven by economic horizontal
inequality. Arguing that this pattern was also present in Tatarstan, Tuva,
Bashkorostan, and Yakutia, Giuliano (2011, p. 208) uses discourse analysis
to trace the grievance mechanism that triggered separatism:

Nationalist leaders framed issues about ethnic economic inequality in order to estab-
lish boundaries between titular and nontitular persons. They claimed that titulars were
unfairly limited in their own republics because they lagged behind Russians in terms of
urbanization, education, and professional achievement. The current system, the nation-
alist claimed, unjustly privileged Russians and denied full participation to titulars in the
local economy. Using this framing, nationalist leaders defined titulars as victims, placed
blame on a discriminatory state, and claimed that obtaining state sovereignty would
restore justice to the nation.

This pattern illustrates neatly the postulated mechanism that transforms

inequality into grievances via group identification, intergroup comparison,
Economic Inequality and Civil War 113

evaluation of injustice, and framing and blaming (see Chapter 3). Ethnic
distinctiveness due to lacking state penetration and previous suffering had
already helped set the stage for the Chechen nationalism. According to Der-
luguian (2005, p. 244), the economic activities of the Soviet state, especially
in the oil industry, created a situation that clearly favored the local Russian
population:

Many Chechens found themselves driven into a semi-proletarian existence on the out-
skirts of Grozny and in the sprawling villages. In contrast to the industrial cities, in such
locations the state provisions of employment, housing, and welfare benefits remained
minimal – which only served to perpetuate among the Chechens a widespread distrust
of the state after the deportation.

It was these “sub-proletarians” who enthusiastically supported Dzokhar

Dudayev’s bid for sovereignty in the early 1990s. A successful air force general,
Dudayev was inspired by Baltic nationalism when he returned to Chechnya to
lead the resistance against Moscow (Gall and de Waal 1998). Unfortunately
for the Chechen nationalist project, quasi-independence from Russia failed to
deliver the economic windfall through oil revenue that Dudayev had hoped
for, and soon a new, even more destructive confrontation with Moscow fol-
lowed. In sum, the Chechen wars were to a large extent caused by economic
grievances, even though both sides of the conflict were guilty of widespread
criminality and thuggery (Zürcher 2007). Thus, rather than being incompati-
ble alternative motivations, greed comfortably coexisted with grievance in this
case.
Although ultimately successful in their secessionist endeavor, the Kosovo
Albanians’ campaign resembles the Chechen case in more than one sense.
Also the most backward region in a communist ethnofederal state, Kosovo
early on developed a strong sense of economic grievance vis-à-vis Belgrade
that exploded in student unrest in Pristina in 1981 shortly after Tito’s death
(Koktsidis and Ten Dam 2008; see also Woodward 1995). As with the Chechen
case, the discriminating policies of the central state directly contributed to
escalating the conflict. As we discussed in Chapter 4, Serb centralization efforts
led to the revocation of Kosovo’s autonomy in 1989, followed by increasingly
repressive measures targeting the Kosovars throughout the 1990s. Ballentine
and Nitzschke’s (2005, pp. 8–9) summary account of the causes of the conflict
is clearly consistent with our grievance interpretation:

Driven by long-standing nationalism and the ever-multiplying political and socio-

economic grievances of Kosovo Albanians, the crisis was sparked by destabilizing pro-
cesses resulting from the disintegration of the Socialist Federal Republic of Yugoslavia.
Socioeconomic underdevelopment, massive unemployment, and widening inter-ethnic
inequalities brought about by the policies of systematic discrimination and exclusion
carried out by the Milošević regime led to the establishment of the shadow “Kosovo
Republic,” a parallel government under the leadership of Ibrahim Rugova.
114 Inequality, Grievances, and Civil War

The Palestinians’ resistance to Israeli ethnonationalist repression represents

another prominent case of economic inequality causing conflict. In Chapter
4, we discussed the political aspects of horizontal inequality in this conflict
at some length. Apart from having their access to state power systematically
blocked, the Palestinians have been persistently disadvantaged in the economic
realm. Starting with the hardship imposed by the dismal conditions in refugee
camps, this discrimination extends to unlawful and overtly ethnically motivated
land expropriations in favor of settlers as well as disruptive limits on inter-
nal and external movements and economic activity under occupation (Pappe
2004). In fact, Israeli settlers and entrepreneurs have come to exploit cheap
Palestinian labor to such an extent that Khalidi (2008, p. 13) refers to the
“Arab-Israeli economy as a marginalized, impoverished, and largely subservient
region of a national (Jewish/Zionist) economy.” In addition to the intense and
long-standing political oppression of the Palestinians by the Israeli state, these
omnipresent frustrations in everyday life have done much to fuel Palestinian
grievances.
Similar tales of economic inequality triggering conflict can be told about
other ethnic groups around the world. The indigenous peoples of Mexico were
exposed to what González Casanova (1965) characterized as “internal colonial-
ism.”12 Their grievances were partly a consequence of political exclusion by the
Mexican state, but poverty combined with perceived threatening land reforms
in the early 1990s as well as various forms of land grabs by agro-industrialists
paved the way for the Chiapas rebellion of 1994 (Alschuler 1998–9).
The predicaments of peripheral peoples within Thailand and the Philippines
also constitute clear cases of similar internal colonialism. Warning against a
one-sided focus on religion in recent analyses of conflict in Southeast Asia, Liow
(2006, p. 7) stresses the centrality of ethnonationalist grievances of Muslim
minorities in these two countries and concludes that

a major reason that global events gain currency in local discourses and narratives
is the existence of very real and longstanding grievances among the Malay-Muslims in
southern Thailand and the Bangsamoro Muslims in southern Philippines stemming from
palpable socio-political and economic marginalization and in some instances outright
victimization by repressive central authorities.

In another study of group-level conflict in Southeast Asia, Brown (2008, p. 281)

finds that “a combination of regional and ethnic horizontal inequalities and
demographic marginalization on ‘ethnic peripheries’ of modern nation states
create the potential for violent separatism.”
The complicated nature of the Darfur conflict makes it difficult to disentangle
the role of economic grievances. However, like the southern provinces, Darfur

12 The notion of internal colonialism and its economic dimensions was further developed by
Hechter (1975), who applied it to peripheral nationalism in the United Kingdom.
Economic Inequality and Civil War 115

has suffered from an almost total neglect on the part of Khartoum with regards
the provision of public goods, a neglect that started already during the colonial
period (Prunier 2005). Thus, the horizontal inequality in favor of the Arabs of
the Nile valley is not just political but also explicitly economic. Thus it is not
surprising that the Fur hold both political and economic grievances (Ylönen
2005).
As indicated by the quantitative analysis in this chapter, we have been unable
to find robust evidence for H5.2, which states that relatively wealthy groups are
associated with conflict. For sure, there exist some cases of wealthy groups that
have expressed wealth-related grievances with their respective capitals, such as
the Slovenes and Croats of Yugoslavia (Woodward 1995) and the Basques of
Spain (Gourevitch 1979), indicating that this mechanism has generated con-
flict in specific cases. However, the dearth of cases fails to yield statistically
significant results with regards to violent conflict.
At one level, this may reflect that there simply exist relatively few cases
of wealthier groups that are politically peripheral. The weakness of H5.2 is
hardly surprising in the light of Horowitz’s (1985) original theory of group-
level inequality and conflict. It also lends credence to Giuliano’s (2011) criticism
of the “wealth hypothesis” according to which secessionism tends to emerge
in wealthy regions as a reflection of the material rewards that would accrue to
secessionist elites in case of independence. Several students of the former Soviet
Union have advanced variations on this argument, including Treisman (1999)
and Hale (2008). Yet, putting most of her explanatory weight on the framing of
grievances, Giuliano prematurely rejects structural inequalities along the lines
of Hechter’s cultural division of labor and our own theoretical perspective
based on horizontal inequality. Although Giuliano argues that these authors
have come to the wrong conclusion by ignoring secessionism in several poor,
peripheral republics beyond Chechnya, we find strong evidence that backward
groups are more likely to rebel, and some support for the claim that wealthy
groups are more likely to challenge the government if one limits the sample
to Eastern Europe and the former Soviet Union (see the Online Appendix).
Finally, we stress that we have only considered resort to violence, and that
many cases of developed separatist groups have relied largely on non-violent
means, perhaps in part because they have greater resources to wield political
influence.
By contrast, the qualitative evidence is very solid in support of H5.3, which
postulates that the inequality effect of H5.1 is conditional on political hori-
zontal inequality. Our case illustrations in this section suggest that economic
grievances are typically expressed together with sometimes intense complaints
concerning adverse political conditions. Indeed, it is difficult for nationalist
activists to boost mobilization in the absence of a political logic targeting
the state as the main problem. Relatively poor groups that enjoy privileged
access to central power are much less likely to resort to arms since they have
116 Inequality, Grievances, and Civil War

peaceful ways to further their economic development, as illustrated by several

black groups in South Africa as well as the Sunni Wahhabi minority in Saudi
Arabia.
We end this section with a few remarks about the robustness of our findings.
It should be recalled that data limitations have forced us to drop all Cold War
cases as well as many smaller groups. Because of the small number of conflict
outbreaks, our findings are sensitive to specific conflict cases, in particular
the two Chechen wars. All the same, we retain these models for consistency
with the other chapters, but more careful analysis using rare events estimation
(King and Zeng 2001), and focusing on territorial onsets only (Buhaug 2006),
yield robust results even if extreme cases are removed from the sample (see
the Online Appendix). Encouragingly, our findings get even less sensitive if we
complement the spatial data from G-Econ with group-level inequality estimates
from surveys around the world.13
As mentioned above, we expect the data limitations to be especially severe in
poor countries with weak statistical data collection. In general, the G-Econ data
set exhibits less variation in such countries, thus making it harder to establish a
link between horizontal inequality and conflict onset. For example, the relative
poverty of the Tuareg in Mali and Niger has been cited as a source of conflict
(see Humphreys and ag Mohamed 2005 and Krings 1995, respectively), but our
data do not exhibit any significant inequality in these cases. Nor do they register
that the northern groups in Côte d’Ivoire suffer from economic disadvantages,
which have been articulated as grievances against the government (Langer
2005). Thus, we have good reasons to believe that we are underestimating the
effect of economic horizontal inequality with our spatial approach.
As we saw in the previous chapter, endogeneity always poses a potential
threat to causal inference. Fortunately, structural indicators of economic hori-
zontal inequality are much less likely to be endogenous to governmental policies
compared to those measuring political exclusion (see our discussion in Chapter
4). The main reason for this is that economic asymmetries among ethnic groups
are much more stable over time than governmental policies of inclusion and
exclusion (Tilly 1999; Stewart et al. 2008). Although at least partly dependent
on the government’s provision of public goods, ethnic groups’ relative eco-
nomic performance typically does not exhibit much change over the decades.
For example, the Kosovo Albanians were at the bottom of the Yugoslav league
table in the 1950s and remained so in the 1970s (Lang 1975).

13 We are indebted to John Huber and his colleagues at Columbia University for providing the
survey data (see Baldwin and Huber 2010). Relying on surveys of household income comple-
ments our spatial data in useful ways, especially since the G-Econ data record all economic
activity including natural resource extraction, which may not benefit the local population as
is the case with the Ijaw in Nigeria and the Aceh in Indonesia. Cederman, Weidmann, and
Gleditsch (2011) find that excluding income from resource extraction in the G-Econ data does
not change the main findings with respect to the effect of inequality on civil conflict.
Economic Inequality and Civil War 117

Conclusion
This chapter has taken a second step in the empirical evaluation of our grievance
perspective. The findings strongly suggest that economic horizontal inequality
is linked to the outbreak of civil war. Previous studies have struggled to find
systematic data at the group level for a large number of countries. This chapter
overcomes this problem by relying on a spatial method that measures wealth
differentials through overlay of spatial income data and ethnic groups’ settle-
ment areas. This approach makes it possible to estimate differences in economic
performance across all ethnic groups that can be reasonably pinned down geo-
graphically.
In agreement with an earlier study of this kind (see Cederman, Weidmann,
and Gleditsch 2011), we have found solid support for the proposition that
groups that are poorer than the country average fight more wars than groups
closer to the country norm (H5.1). As opposed to the previous study, however,
we concede that the evidence for a higher risk of conflict is weaker when it
comes to the opposite end of the wealth spectrum when we do not impose a
symmetric effect of inequality. Although specific cases, such as Slovenia and
Croatia in the former Yugoslavia, indicate that relatively wealthy groups may
harbor grievances because of the perceived injustice associated with fiscal redis-
tribution in favor of less-wealthy groups, this effect appears to be less robust
for violent conflict and cannot be reliability established with much confidence
on a global basis (H5.2).
This chapter also shows that the impact of economic horizontal inequality on
conflict risk hinges on the group in question being politically excluded (H5.3),
which is an entirely new finding. As with the politically conditional size effect
that we investigated in the previous chapter (H4.3), this result indicates that
political horizontal inequality plays a pivotal role in the perception and framing
of grievances. While economic comparisons make a difference, they appear to
be less conflict-inducing in, and possibly compensated by, inclusive political
settings. We will return to the policy implications of this finding in Chapter
9, but for now we note that this offers some hope that short-term political
reforms may partly help to compensate for long-term economic inequalities.
While suggestive, our findings do suffer from limitations that will have to
be addressed in future research. The spatial data offer important advantages
in terms of global coverage and uniformity, but are associated with important
weaknesses as regards data quality in the least developed countries. Moreover,
inequalities among groups whose settlement areas are not distinct cannot be
measured. For these reasons, it would be useful to complement our data with
information from other data sources, including survey research (Baldwin and
Huber 2010) and satellite data (Agnew et al. 2008; Elvidge et al. 2009; though
see Chen and Nordhaus 2011).
Furthermore, it should be kept in mind that, even though our analy-
sis is more disaggregated than the customary country-level proxies used in
118 Inequality, Grievances, and Civil War

quantitative civil war research, the group-level analysis presented here also
hinges on theoretical interpolation in order to connect structural inequalities
with collective-level violence. Although our proposed causal mechanisms are
potentially capable of closing this explanatory gap, as suggested by our brief
case discussion in the previous section, we cannot provide more extensive,
direct evidence of their operation in this book. Beyond citing separate case
studies in confirmation of these mechanisms, it would be desirable to improve
and expand existing data sets such that interactions between incumbent gov-
ernments and their challengers can be traced in greater detail, while relying on
systematic information on repression and mobilization before violence breaks
out (see, e.g., Cunningham 2010). Building on the pioneering efforts of Gurr
(1993a; 2000b) and his team, who have also collected extensive data on social
and cultural horizontal inequalities, such information would help disentan-
gle the process at lower levels of aggregation and help us establish whether
the causal imputations remain robust to such scrutiny. Fine-grained temporal
measurements could also help develop an explicitly endogenous account of hor-
izontal inequalities, which have been kept exogenous in this book. As we have
pointed out, although this assumption may be less precarious with respect to
political horizontal inequalities, studying endogenous changes in public goods
is interesting and valuable in its own right.
6

Transborder Ethnic Kin and Civil War

In the previous two chapters, we have explained how the likelihood of civil
war outbreak depends on internal conditions within states. In this chapter, we
relax this limitation.1 Indeed, beyond the demonstrated importance of domes-
tic grievances and opportunities, civil wars are often influenced by the wider
international environment. A series of recent studies has shown that civil wars
can be caused not only by factors inside countries, but also by effects operating
across state borders (see, e.g., Salehyan and Gleditsch 2006; Gleditsch 2007).
Whereas a first wave of quantitative studies demonstrated that such effects
make the “closed-polity” assumption untenable, more recently researchers have
made efforts to identify and examine particular causal mechanisms driving con-
flict, including the role of transborder ethnic kin (TEK) in conflict processes.
In contrast to earlier contributions to this literature, which tended to stress
identity as the key motivating factor, contemporary scholarship typically high-
lights how the conjunction of ethnicity and power politics increases the risk
of conflict, suggesting that “identity politics is often more about politics than
about identity” (King and Melvin 1999/2000, p. 109; see also Mousseau 2001;
Thyne 2007; Cederman, Girardin, and Gleditsch 2009). We have emphasized
the importance of ethnic exclusion in states, but the political role of ethnicity
is of course equally relevant for groups spanning national boundaries.
Despite these recent advances in studying transnational ethnic linkages,
a central puzzle remains unresolved, namely how ethnic groups that at
least in theory could count on support from large TEK groups often have
remained surprisingly peaceful. The most prominent “dog that didn’t bark,” or

1 This chapter draws directly on Cederman, Gleditsch, Salehyan, and Wucherpfennig (2013).
We are grateful to Idean Salehyan and Julian Wucherpfennig for major contributions to the
current analysis. Arman Grigorian and Andreas Wimmer offered important conceptual input at
an earlier stage of the project.

119
120 Inequality, Grievances, and Civil War

perhaps better “bear that didn’t growl,” in this sense are the Russians in the
“near abroad.” There was widespread anticipation that the Russian diasporas
stranded in former Soviet republics after the collapse of the Soviet Union in
the early 1990s would resort to violence, but these communities have remained
relatively calm. However, the picture looks entirely different if we consider the
Albanians in the former Yugoslavia, or the Armenians in Azerbaijan. These
groups have been involved in major fighting with active support from TEK
groups. Given a similar pattern of domestic marginalization and cross-border
ethnic kin, why then do some groups rebel while others do not?
In this chapter we seek to ascertain whether there is any systematic link
between the transnational ethno-demographic balance and the probability of
conflict. We conjecture that the effect is not linear: as the size and power
of the TEK group rises, conflict should be more likely, but very strong kin
groups – such as the Russians – can in fact deter conflict. Our argument has
been anticipated by others. In an article on “hypotheses on nationalism and
war” that appeared shortly after the end of the Cold War, Van Evera (1994)
proposed the existence of such a curvilinear effect. Offering a more thorough
theoretical derivation, Van Houten (1998) addressed the same puzzle, arguing
that large external states controlled by TEK groups could have a pacifying
effect on ethnic politics. Yet, the author himself admits that his study should
be seen as a plausibility probe rather than a definitive test, since the evidence is
limited to selected case studies. Thus, almost two decades after the publication
of Van Evera’s initial conjecture, evidence of the conflict-reducing TEK effect
remains uncertain.
Here we test whether the curvilinear proposition between the size of TEK
groups and the risk of conflict applies more generally. Progress in data collec-
tion now allows us to evaluate this proposition systematically across a large
number of cases. Drawing on a new transnational extension to the Ethnic Power
Relations (EPR) data set that we have relied on in the previous two chapters
(EPR-ETH), we find strong quantitative support for the hypothesis. Indeed, the
propensity of conflict is highest for TEK groups in the intermediate range of
relative sizes. As “power parity” theories would expect, external groups that
have approximately the same relative size as the incumbent groups tend to see
the most conflict. However, compared to groups without transnational kin, the
net effect is contingent on the political status of the TEK group: TEK groups
that are excluded from power in their home countries increase the probability
of political violence. Included TEK groups, in contrast, appear to decrease the
risk of violent conflict, especially when groups are large, relative to the ethnic
group(s) in power.

The Literature on Transborder Ethnic Kin and Conflict

Much research on civil war has treated countries experiencing civil wars as
“closed polities.” It is often assumed that the causes of “internal” conflicts
Transborder Ethnic Kin and Civil War 121

must be found inside the boundaries of those states. However, even a cursory
glance reveals that many civil wars display strong linkages to actors and events
extending beyond the boundaries of individual countries. Military or non-
military support for either the government or rebel side can be decisive for the
dynamics of conflict. Soviet support helped sustain the Marxist government
in Afghanistan as well as the Mujahedin opposition. Support from Rwanda
helped Kabila’s Alliance of Democratic Forces for the Liberation of Congo
transform from a marginal movement in Eastern Zaire to a military force
able to overthrow the Mobutu government. These ties extend to ethnic kin
groups as well. The civil wars in the former Yugoslavia, for example, included
widespread participation by diaspora communities in other states. Moreover,
many of the Albanian National Liberation Army fighting units in Macedonian
had previously fought with the Kosovo Liberation Army in Kosovo.
The clear transnational linkages in many civil wars have fostered interest in
how civil wars may be shaped or affected by external characteristics. A first
wave of research looked at intervention in civil wars, examining either whether
certain types of civil war were more likely to see outside intervention or how
external intervention shape conflict outcomes (e.g., Moore and Davis 1998;
Balch-Lindsay and Enterline 2000; Regan 2000), emphasizing factors such as
interstate rivalry, concerns over the externalities of conflict, and shared ethnic
ties (Woodwell 2004). However, if transnational linkages can be shown to
influence the outcome of ongoing civil wars, then transnational linkages and
the prospects for external support should also affect the decision to resort to
conflict in the first place.
Another wave of studies notes that civil wars cluster geographically and
shows that proximity to conflict can increase the risk that a country will expe-
rience a civil war (Hegre and Sambanis 2006; Buhaug and Gleditsch 2008;
Bosker and de Ree 2010). Although the tendency for conflict to spread from
one state to another could stem from externalities entirely unrelated to the
actors, such as cheaper weapons, social links, and affinities between communi-
ties across national boundaries seem a very likely source of conflict transmission
(Buhaug and Gleditsch 2008; Bosker and de Ree 2010). However, as noted by
Cederman, Girardin, and Gleditsch (2009) and Salehyan, Gleditsch, and Cun-
ningham (2011), transnational support from ethnic kin can also help facilitate
resort to violence, even when the transnational kin are not themselves involved
in a conflict in their home state.
There is little consensus in the literature with regard to the effect of TEK on
civil war outbreak. In fact, virtually all conceivable causal relationships have
been proposed at some point or another, including both positive and negative
influences, as well as no connection at all. Some studies argue that TEK groups
have offensive motives linked to ethnic nationalism and postulate a positive
effect on conflict. In particular, aggressive interventions driven by irredentism
could trigger a spiral of escalating tensions on both sides of the border (Weiner
1971). There are many examples of such border-transgressing processes in
122 Inequality, Grievances, and Civil War

the breakup of the former Yugoslavia. Although most of this literature relies
on qualitative evidence (e.g., Horowitz 1985; Heraclides 1990; Chazan 1991;
Jenne 2007), some scholars have conducted quantitative tests, usually based
on Minorities at Risk data, that generally suggest a positive impact on conflict
exerted by TEK groups (see Moore and Davis 1997; Saideman 2001; Woodwell
2004).
From the vantage point of bargaining theory, Cetinyan (2002) contends that
group sizes and transnational kin groups should not influence the probability
of conflict onset. This interpretation recognizes that groups’ demands vary with
their demographic weight, but rejects that this would have any implications for
the outbreak of violence. Potential threats from external kin, he argues, should
be reflected in domestic bargaining processes as groups with strong TEK ties
receive greater concessions from the state. Cetinyan’s analysis is creative, but
there are at least two reasons to doubt the theoretical and empirical null results
he presents. First, the restricted sample of groups in the MAR data, which
by construction are limited to threatened minorities, makes them less suitable
for general tests of demographic size. Second, Cetinyan’s model rules out the
possibility that group size could be correlated with uncertainty. By contrast,
a number of scholars postulate a curvilinear relationship between the two,
since strategic misrepresentation is most important at parity (e.g., Reiter 2003;
Walter 2009a). We present a similar argument below.
In an attempt to overcome some of the sample restrictions associated with
the MAR data by relying on alternative, group-level data, Cederman, Girardin,
and Gleditsch (2009) find that TEK groups increase the probability of internal
conflict by shifting the shape of the observed effects of the power balance
between the groups associated with the rebels and the incumbent government.
While innovative in its use of geographic information systems, their article has a
limited scope in several respects. First, the GREG data on ethnic groups depend
directly on the Soviet Atlas Narodov Mira, which are defined primarily based
on language, and may not correspond to the relevant identities. Moreover,
the data are from the 1960s and could thus be said to be outdated (even
though ethnic settlement patterns are surprisingly stable over time). Second,
their analysis is limited to Eurasia, thus leaving cases in Africa and the Western
Hemisphere outside the analysis. Third, and most importantly, their model
does not factor in the size of the TEK group itself and therefore does not speak
to the size puzzle that is the main focus of this chapter.
In an early discussion of “hypotheses on nationalism and war” that we have
previously referred to, Van Evera (1994) develops a subtle argument stating
that TEK groups will be less likely to intervene if rescue is either impossi-
ble or relatively easy. It is, instead, “in-between situations . . . that are most
dangerous” (p. 20). However, Van Evera’s article never moves beyond listing
hypotheses accompanied by selected case illustrations.
Relatedly, Van Houten (1998) offers a more sophisticated attempt to explain
the relative peacefulness of the Russian near abroad and similar cases. Focusing
Transborder Ethnic Kin and Civil War 123

on kin states, referred to as “reference states,” Van Houten extends Fearon’s

(1998) theory of credible commitments from minority-majority relations to
transborder actors. Noting that Fearon’s argument overpredicts violence in
cases such as the Baltic states, he argues that external relations may pacify
domestic settings under specific conditions. Irredentist intensions may exist
among strong states, in which case conflict is very likely. However, ceteris
paribus, increasing strength of the reference state should stabilize minority-
majority relations: “the presence of a militarily strong but only moderately
irredentist reference state is most likely to overcome the commitment problem
and prevent ethnic violence” (p. 112). Thus, strong external kin states are in a
much better position to make credible threats to intervene than their weaker
counterparts:

On the one hand, if it is known that the reference state is likely to intervene in the
new state if the minority is oppressed, then the ruling majority in the new state has an
incentive to protect the minority. This threat allows the majority to credibly commit to
do this. On the other hand, if the reference state is too aggressive, it may induce a war
with the new state (Van Houten 1998, p. 112).

Recognizing the difficulties of testing an argument that hinges on intensions,

Van Houten limits the application of his game-theoretic model to a small
number of cases in the Eastern Europe: “I make no attempt to actually prove the
correctness of the model; rather, I try to show its plausibility.” Thus, building
on, and further developing, Van Houten’s ideas, we attempt to go beyond the
task of establishing plausibility by testing our hypotheses systematically across
a large number of cases.

Theorizing the Effect of Transborder Ethnic Kin

Following pioneering conceptual work by Weiner (1971) and Brubaker (1996),
we start our theoretical derivation of the TEK effect by identifying the ethno-
nationalist triad as the critical constellation of actors within which border-
transgressing influences on conflict operate. Building directly on our dyadic
polity model in Figure 4.1, an ethno-nationalist triad can be conceptualized as
an extended center-periphery configuration by adding the TEK group to the
primary dyad, which involves the group challenging the power of an incumbent
ethnic group (Cederman, Girardin, and Gleditsch 2009). Through the addition
of a TEK group on the challenger’s side, the incumbent is now faced with an
additional, potentially opposed actor in a secondary dyad. The presumably
supportive relationship between the TEK group and the challenger constitutes
the third leg of the triad.
There are two versions of the triadic constellation depending on the power
status of the TEK group in its own state. We illustrate these two possibilities in
Figures 6.1 and 6.2, respectively. These figures extend Figure 4.1 with a second
state, referred to as State B, which is the host of the TEK group. The primary
124 Inequality, Grievances, and Civil War

figure 6.1. The ethnonationalist triad with an excluded TEK group.

dyad still connects the government with the challenger group in State A. The
crucial addition is the secondary dyad that links the incumbent group in State
A with the TEK group in State B.
Let us first consider Figure 6.1. As illustrated by the situation of the Kurds
in Turkey (here State A), if the TEK group is excluded in State B (e.g., in
Iraq before the fall of Saddam Hussein), we can expect there to be a tension
between the TEK group and its government, but also between the governments
of States A and B, that is, between Iraq and Turkey. Figure 6.2 depicts a
constellation where the TEK group happens to be included in the government.
Here we can use Estonia and Russia as examples of States A and B. Within
Estonia, the challenger group refers to the Russian minority. Potentially, this

figure 6.2. The ethnonationalist triad with an included TEK group.

Transborder Ethnic Kin and Civil War 125

group can expect support from its ethnic kin in Russia, where the Russians
control the state. Under these circumstances, friction in the secondary dyad
could trigger irredentist conflict between States A and B.
Since demographic group size is relatively easy to measure and tends to be
exogenous to conflict except under extreme cases of ethnic cleansing, we will
continue to rely on this dimension to reflect power in our analysis (see also the
justification for group-level analysis in Chapters 2 and 3). For sure, there are
instances where the participants of conflict processes constitute tiny fractions
of the group’s population, such as ruthless warlords and armed bands, but we
expect the size of the group as a share of a country’s population to be relevant
to conflict onset in most cases, in terms of both potential resources and overall
support or legitimacy.
Our main task, then, is to extend the study of conflict within the primary
dyad to the transnational dimension. To what extent does this conflict process
involve the TEK group through the secondary dyad? Assuming that a nation-
ality conflict breaks out between the incumbent group(s) and the challenger,
we derive the conditions of conflict as a function of the conditions within the
primary and secondary dyads. Our analysis of relative group size in the primary
dyad is captured in H4.3, for which we found strong evidence in Chapter 4.
Indeed, the larger the challenger group is compared to the population of the
incumbent, the more likely violent outcomes become.
We are now ready to consider the secondary dyad’s influence on internal
conflict. Before discussing the different ways that the TEK group’s size can
affect the propensity of political violence, we start by considering a categorical
effect. Such a formulation would be compatible with a strictly essentialist inter-
pretation of ethnic kin that is unrelated to power relationships. As a reflection
of strong ethnic solidarity, all TEK groups should be highly motivated to inter-
vene. According to this approach, such groups are motivated by “reasons of
affinity and sentiment rather than . . . power or more hard-headed cost-benefit
analyses” (Holsti 1996, p. 127; see also Connor 1994). Based on this essentialist
logic, we derive the following hypothesis:

H6.1. The probability of conflict increases in the presence of a TEK group.

Nevertheless, we do not expect this hypothesis to be empirically accurate.

Although clearly emotionally loaded, ethnic politics is as much about power
and politics as it is about ethnic solidarity:

Disputes over the allocation of scarce resources, competing visions of foreign policy
directions, domestic political contests, and other prosaic features of political life fre-
quently trump any putative duty that political elites might feel toward individuals who
share their language or culture beyond their own frontiers (King and Melvin 1999/2000,
p. 109).
126 Inequality, Grievances, and Civil War

Whether orchestrated by state-owning or state-seeking ethnic kin, intervention

in transborder conflicts is a risky business that can have massively negative
consequences for the TEK group in question (Horowitz 1985). Indeed, it is
not surprising that quantitative studies, including Cederman, Girardin, and
Gleditsch (2009), have generally failed to find support for the simple hypothesis
that the presence of a TEK group by itself influences the risk of conflict.
An alternative account takes power politics more seriously by letting the
probability of conflict increase with the relative demographic weight of the TEK
group compared to the incumbent. The larger the group, the more confident
it will be that its intervention will contribute to the successful challenge of
the government in the primary dyad without exposing it to inordinate risks.
For example, Jenne (2007) argues that powerful secondary dyads encourage
groups in the primary dyad to radicalize and articulate larger demands, which
can lead to violence. This logic runs parallel to that of H4.3 in the primary
dyad by using relative demographic weight as a rough measure of the dyadic
power balance:

H6.2. The probability of conflict increases with the relative size of the TEK group
in the secondary dyad.

There are few examples of empirical studies measuring power relations in

the secondary dyad and those that do typically find no proportional effect. For
example, in their quantitative investigation of irredentism, Saideman and Ayres
(2000) do not detect any influence of relative group size, which is also in line
with Cetinyan’s (2002) findings. Nonetheless, we include this hypothesis and
test it, using our new data resource.
There are also important theoretical reasons to discount hypothesis H6.2.
The postulated monotonic effect of TEK group size fails to account for the
puzzle of the Russian near abroad, because it overlooks the strategic nature of
the secondary dyad. As the size of the TEK group increases, more resources
will become available to the challenger in the primary dyad, but the incumbent
regime should also take this into account when bargaining with the group.
We have postulated a linear effect of group size on conflict in the primary
dyad, but there are good reasons to believe that such a relationship needs to be
amended when it comes to the influence of the secondary dyad. Why should
large, powerful TEK groups be able to “protect” their kin, but large groups
in the primary dyad be more associated with conflict? Deviating somewhat
from Van Houten’s (1998) emphasis on commitment problems, we propose
two causal mechanisms that generate the same relationship between the size
and conflict variables.
First, large challengers in the primary dyad are unlikely to extract bargain-
ing concessions by threatening ruling minorities. Majority groups challenging
minority incumbents will remain excluded from state power unless there are
Transborder Ethnic Kin and Civil War 127

exceptional circumstances that make small ruling coalitions unstable. Some

small ruling coalitions – such as the White Apartheid government in South
Africa – established their dominance through successful coercion and managed
to cling to power for decades. Oppressive incumbents are unlikely to compro-
mise with challengers as they fear that concessions will threaten their long-term
survival and the benefits of rule. Going down the reform path may lead to the
dissolution of the regime at best, and exile, imprisonment, or death at worst.
Thus, elites defending minority rule will be inclined to rely heavily on coercion
to stifle challenges to their supremacy (Heger and Salehyan 2007).
However, such an uncompromising attitude may be much less viable if
overwhelming external groups get involved. While incumbents have greater
means to maintain control domestically, transnational actors are inherently
more difficult to police and coerce since they can mobilize resources beyond the
shadow of repression (Salehyan 2009). Because of the potential threat posed by
such interventions, we would expect incumbent regimes to be mindful of ethnic
opposition groups’ respective “homelands.” In other words, large domestic
challengers – while potentially threatening – can be successfully controlled by
the state’s coercive apparatus, but large external challengers are much more
difficult to deal with.
This type of caution should increase with the size of TEK groups relative
to the incumbent group. Put differently, the latter’s willingness to risk violent
conflict should be a decreasing function of the TEK group’s relative size. Yet
since the probability of conflict is shaped by both opportunity and willingness,
the combination of external resources and external treats produces a curvilinear
effect. For small TEK groups, the conflict propensity would be close to zero
since these groups would make little difference even if they intervene. Where
the TEK groups are relatively large, however, states will be more constrained
in their behavior because of the deterrent effect of a possible intervention.2
Bargaining theory suggests a second reason to expect more conflict in the
mid-range of TEK sizes. Originally applied by Blainey (1973) to interstate rela-
tions, this argument hinges critically on uncertainty and strategic incentives
to misrepresent one’s true strength. According to this reasoning, conflicts are
especially likely to break out where the power balance is roughly even (see
Reiter 2003). Rough parity makes it difficult for the parties to assess who
would prevail in case of combat, and so attempts to misrepresent gain signif-
icance (Geller 1993; Lemke and Werner 1996; Wagner 2000). Conflict may
then occur as the result of information problems leading to bargaining failure

2 A simple formalization shows why the resulting effect should be curvilinear in terms of the
relative size r of the TEK group. Assuming that the effect of the relative weight of the TEK group
compared to the incumbent is r[0,1), the capacity argument according to H6.2 would be linear
in r, and the incumbent’s willingness to apply force a decreasing linear function 1–r. Modeling
the combined effect as the product of both these functions, we get r(1–r) = 1/4 – (r–1/2)2 , i.e.,
an inverted U-shaped curve with its maximum value at r = 0.5.
128 Inequality, Grievances, and Civil War

(Fearon 1995). This reasoning has generally not been extended to civil wars,
which are typically thought to be caused by commitment problems (e.g., Fearon
1995; 1998), but there are clear reasons to believe that uncertainty could be
a cause of conflict in such settings. For example, Walter (2009a) points to
obvious sources of uncertainty, such as the sources of rebel financing and gov-
ernmental resolve, to which one might add the extent of popular mobilization
on either side.
Yet, the lack of information should be even more pronounced as regards
the possible involvement of TEK groups. Transnational interactions tend to be
much less transparent than domestic politics. This fundamental information
asymmetry is especially severe for the incumbent group, because its intelli-
gence services are likely to encounter much more resistance in their attempts
to gather information about the resources and intentions of foreign groups
(Salehyan 2009, p. 48). To some degree, the situation resembles the difficulties
surrounding extended deterrence in international relations, which is deemed to
be especially challenging. The challenge derives above all from the uncertainty
and lacking credibility surrounding third parties’ propensity to intervene in
defense of proxy states (e.g., Huth 1988). Thus, it can be surmised that there
will be considerable uncertainties, in terms of both estimating the capabili-
ties of a TEK group and its willingness to intervene, especially if the power
balance is close to parity. Furthermore, these uncertainties should dwarf the
corresponding ambiguity in domestic politics.
We are thus ready to derive our main hypothesis regarding the functional
form of the TEK effect:

H6.3. The probability of conflict follows an inverted U-shape for the relative size of
the TEK group in the secondary dyad.

So far, our hypotheses have said nothing about the influence of TEK links
on conflict compared to situations without kin. We argue that there is a medi-
ating effect depending on whether the TEK group is included in, or excluded
from, executive power in their own state. The most straightforward argument
focuses on the superior power of state-controlling TEK (see Figure 6.2). Other
things being equal, we would expect TEK groups that control a state to be
more powerful. Yet as we have argued above, the relative power balance in the
secondary dyad should constrain conflict as TEK groups become more threat-
ening. While we focused on power-as-population above, control of a state and
its military apparatus should augment power even further. Thus, while eth-
nic Russians and ethnic Chinese are numerically large in global terms, they
are even stronger than their size alone would suggest as they control powerful
states. Marginalized TEK groups, on the other hand, may be willing and able to
mobilize resources for their kin in the primary dyad (see Figure 6.1). Yet, there
will be considerable uncertainty over the resources and support that they can
Transborder Ethnic Kin and Civil War 129

garner as their capabilities are unknown and resource pools are insecure. TEKs
that control states usually operate through official diplomatic channels and
have known resource streams and capabilities, which bolster credible threats
and reduce bargaining uncertainty.
In addition, we must consider the TEK group’s willingness to intervene.
Incumbent groups have much more to lose than stateless groups, especially in
the case of multiethnic polities that could be seriously destabilized by actions
that undermine the sanctity of borders (Horowitz 1985; Cederman, Girardin,
and Gleditsch 2009). International norms of territorial integrity applying to
states, rather than to nonstate actors, should also restrain incumbent groups
from interfering in the internal affairs of neighboring states (Zacher 2001).
In contrast, TEK groups without access to state power, such as the Kurds in
Turkey, are less bound by such considerations, and generally more likely to
have combat experience as a consequence of their marginal position.
In short, TEK groups that control a state can “speak softly and carry a big
stick.” They are more able to intervene given their control over military forces
and these credible threats should be reflected in bargaining in the primary dyad.
They may be more reserved in their use of force, given international norms
against war, but can still bring the power of the state to bear in diplomatic
relations. Marginalized TEK groups, on the other hand, do not face similar
constraints. Support for their kin may be uncertain and difficult to mobilize
since they do not have the same military capabilities. For these reasons, large
included TEK groups tend to deter conflict, while excluded TEK groups can
foster bargaining uncertainty and promote conflict. Thus, we state our final
hypothesis:

H6.4. The conflict-dampening effect occurs for relatively large TEK groups that are
included, and not for those that are excluded.

Data
The empirical assessment of our hypotheses depends on the availability of
suitable data on ethnic groups and their transnational connections in a wide and
unbiased sample. Thus, we extend the EPR-ETH data introduced in Chapter 4
to encompass TEK links. Since the coding of EPR groups is country-specific,
there is no automatic way of identifying such connections. In contrast, the Atlas
Narodov Mira features one global list of groups, which simplifies its use in the
identification of transborder kin (see Cederman, Girardin, and Gleditsch 2009).
We therefore decided to code TEK links based on nominal identification by
matching group names, including the possibility of synonyms for groups with
shared languages. Thus, for example, Swiss Italians are linked to the Italians
through the TEK group for all Italians. There is, of course, no guarantee that
matching names will yield politically relevant bonds, and indeed in the case of
130 Inequality, Grievances, and Civil War

the Swiss Italians automatic identification with “Italian Italians” would be far
from obvious. However, this method is less afflicted by bias than attempts to
code constructivist interpretations based on partial information (see our online
appendix for a full listing of all TEK groups).3
The coders were asked to identify all ethnic groups that appear in more
than one country and to give them a special TEK code that differs from the
country-specific group identifiers. Groups that appear only in one country were
not given any TEK code. EPR also features “umbrella groups” composed of
several ethnic subgroups. For example, in Mali the umbrella group “blacks”
includes ethnic Mande, Peul, Voltaic, and others. In these cases, the composite
group can be associated with more than one TEK code, with three as the
maximum number of links allowed in the data set.
Before entering the analysis, TEK connections featuring groups from non-
contiguous countries were removed from the data set. Long-distance nation-
alism involving diaspora has also been listed as a possible source of conflict
(e.g., Shain and Barth 2003), as has funding of rebel campaigns in the Third
World through migrant diasporas in the United States (Collier and Hoeffler
2004). These conflict mechanisms are different from our concern in this chap-
ter, where we focus entirely on transborder ethnic kin. Moreover, few countries
have the ability to project force across long distances, and such kin group con-
nections are less relevant for threat perceptions.

Variables
The next step is to introduce our main variables. Throughout this chapter,
we continue to rely on a dependent variable that is based on a group-level
coding of the UCDP/PRIO Armed Conflict Data Set (Gleditsch et al. 2002). As
described in Chapter 4, the resulting coding assures that each conflict onset is
mapped to the corresponding EPR group provided that the rebel organization
expresses an aim to support the ethnic group and members of the group in
question participate in combat. Again, the onset variable is coded one for a
group-year during which a conflict started and zero otherwise, with ongoing
conflict excluded.
We proceed by introducing the independent variables by extending our vari-
able definitions that were introduced in Chapter 4. Denoting the TEK group’s
population K,4 and the population of the incumbent I, we operationalize the
main independent variables related to TEK as follows:

3 Our team has started coding constructivist exceptions from the name-matching rule, but the
coverage of this more refined coding is still only partial and we refrain from using it here.
However, using the existing partial corrections makes no substantive difference to the results
reported here.
4 Note that there may be more than one TEK group. In such cases, K is the sum of the population
of segments in all neighboring countries.
Transborder Ethnic Kin and Civil War 131

r Relative TEK size k [0,1) in the secondary dyad is defined as K / (K+I).

r Relative size for TEK groups that are state controlling and stateless, kincl
and kexcl respectively, are computed on the basis of TEK group populations
from either group category.
r We also employ a TEK dummy variable, which is coded 1 if the group has
at least one TEK group, and 0 otherwise.

The remaining independent and control variables are identical to those used
in Chapter 4.

Results
Having described the data and defined our variables, we are now ready to
present our main results (see Table 6.1). Using Model 4.2 as the reference
model, our sample encompasses all EPR-ETH groups from 1946 through 2009.
Again relying on robust country-clustered standard errors, we conduct all anal-
ysis with logit models, using the onset of ethnonationalist conflict at the group
level as the dependent variable.
As a first step, Model 6.1 introduces a dummy variable indicating whether
the group has TEK. Since the effect is indistinguishable from zero, we reject
H6.1, consistent with our theoretical expectations. Indeed, the introduction of
the TEK dummy variable does little to affect the other coefficients. Confirming
the findings of the models in Chapter 4, a group’s power access is negatively
related to conflict whereas downgraded groups and those with a history of
conflict are more likely to experience violence. At the country level, GDP per
capita has a negative and significant effect, but the population variable again
fails to reach significance.
As a direct assessment of H6.2, Model 6.2 evaluates whether the probability
of conflict increases monotonically with the relative size of the TEK group. We
retain the TEK dummy, which now serves as an intercept for the linear effect
for groups without TEK, where the size is set to zero. Again, the estimate of the
size effect in the secondary dyad is far from significant, although it is positive.
In agreement with our theoretical anticipation, we reject H6.3 as well.
Our next task is to test the postulated inverted U-shaped curvilinear rela-
tionship between TEK size and conflict. Since the indicator for the demographic
balance ranges from 0 to 1, both terms should be of roughly equal magnitude
and the quadratic term should be negative if the hypothesis is correct. In sup-
port of H6.3, the estimates of Model 6.3 reveal that while the monotonic effect
of relative group size in the primary dyad remains strong, the relative size of
the TEK group influences conflict propensity in an inverted U-shaped fashion.
Encouragingly, both the linear and square terms are strongly significant, with
the former being positive and the latter negative. Using 90 percent confidence
intervals, Figure 6.2 plots the predicted probability of conflict as a function of
relative TEK size, indicating that if the other variables are held at their means,
132 Inequality, Grievances, and Civil War

table 6.1. TEK and Group-Level Onset of Civil War, 1946–2009

Model 6.1 Model 6.2 Model 6.3 Model 6.4

Rel. Group Size 1.3807** 1.2713** 1.2622** 1.1783**
(0.3131) (0.3347) (0.3404) (0.3659)
TEK Group 0.1009 − 0.0179 − 0.4115
(0.1927) (0.2092) (0.2440)
Rel. TEK Size 0.3441 4.2382**
(0.3306) (1.3420)
Rel. TEK Size, sq. − 4.2058**
(1.3620)
Excluded TEK Group − 0.1967
(0.2237)
Rel. TEK Size (excl.) 3.3767*
(1.6882)
Rel. TEK Size, sq. (excl.) − 2.6340
(1.9853)
Included TEK Group − 0.3549
(0.2852)
Rel. TEK Size (incl.) 4.5133**
(1.6719)
Rel. TEK Size, sq. (incl.) − 5.1434**
(1.6867)
Junior Status 0.8415 0.7939 0.7823 0.7781
(0.5207) (0.5295) (0.5248) (0.5323)
Autonomy Status 1.2021* 1.1249 1.1647* 1.1097
(0.5741) (0.5840) (0.5811) (0.5852)
Powerless Status 1.4237** 1.3527* 1.3675* 1.3039*
(0.5309) (0.5430) (0.5396) (0.5458)
Discriminated Status 1.9905** 1.9315** 1.9273** 1.8952**
(0.5075) (0.5194) (0.5162) (0.5177)
Separatist Status 3.2137** 3.1275** 3.2512** 3.2108**
(0.6609) (0.6818) (0.6625) (0.6660)
Downgraded 1.5768** 1.5756** 1.5585** 1.5537**
(0.2942) (0.2950) (0.2922) (0.2898)
Number Previous Conflicts 0.5481** 0.5469** 0.5173** 0.5070**
(0.0968) (0.0959) (0.0963) (0.0933)
Ongoing Conflict, lag 0.3903 0.3960 0.4421 0.4634
(0.2856) (0.2866) (0.2864) (0.2856)
GDP/capita, lag, log − 0.1707* − 0.1674* − 0.1777* − 0.1918*
(0.0722) (0.0726) (0.0728) (0.0756)
Population, lag, log 0.0866 0.0989 0.1198 0.1386
(0.0889) (0.0937) (0.0957) (0.0969)
Constant − 5.8275** − 5.9053** − 6.1146** − 6.2274**
(1.0901) (1.1246) (1.1326) (1.1281)
Observations 28,302 28,298 28,298 28,298
Robust standard errors in parentheses; estimates for peace-year correction not shown.
** p < 0.01, * p < 0.05.
Transborder Ethnic Kin and Civil War 133

0.012
0.010
Pred cted Probab ty of Conf ct

0.008
0.006
0.004
0.002
0.000

0.0 0.2 0.4 0.6 0.8 1.0

Relative TEK Size

figure 6.3. The probability of conflict as a function of relative TEK size.

conflict propensity increases from around 0.3% to as much as 0.8% per group
and year if the relative TEK size is shifted from zero to rough parity. Illustrating
the distribution of the observations, the rug plot along the x-axis of Figure 6.3
demonstrates considerable variation in the relative size of TEK groups.
Our analysis continues with an evaluation of H6.4, which considers the
separate effect of different types of TEK groups compared to the non-TEK
baseline. The results of Model 6.4 suggest that both included and excluded TEK
groups produce a curvilinear effect on conflict onset. However, it is difficult
to compare the net influence of both of these based on the coefficients alone.
We therefore refer to Figure 6.4, which displays the respective marginal TEK
effects in two separate panels, with the left indicating included TEK groups
and the right corresponding to excluded groups. Figure 6.4 thus depicts the
first difference in predicted probabilities for given levels of relative TEK size
and status, compared to groups without TEK.
It immediately becomes clear that the effect differs significantly between
the two types of TEK groups. In support of H6.4, the graphs tell us that for
comparable TEK sizes, the conflict propensity is considerably higher for large
excluded TEK groups than for included ones. In fact, the conflict-restraining
effect can only be observed for state-controlling TEK groups. For very large rel-
ative group sizes, the effect appears to be significant since the error bands do not
intersect zero in that range. By contrast, the marginal effect of excluded TEKs
134

Included TEK Groups Excluded TEK Groups

0.010

0.010
0.005

0.005
F rst D fference

F rst D fference
0.000

0.000
−0.005

−0.005
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0

Relative TEK Size Relative TEK Size

figure 6.4. Conflict probability varying with size of included and excluded TEK groups.
Transborder Ethnic Kin and Civil War 135

is generally positive and statistically significant for the intermediate size range,
but becomes indistinguishable from the corresponding non-TEK situation for
extreme values of the size spectrum. However, despite a falling tendency for
large groups, the net effect never becomes conflict dampening. Echoing the
findings of Saideman and Ayres (2008), we conclude from this that irredentism
may be much less important as a general trigger of civil war than often assumed,
although we cannot exclude the possibility that other types of violence, such
as interstate disputes and wars, could be triggered by included TEK groups.
Indeed, our results indicate that the conflict-dampening influence appears to
be limited to included TEK groups, which is in line with Van Houten’s (1998)
account of the relatively peaceful Russian diaspora.
Finally, we explore possible interactions between the relative demographic
weights in the primary and secondary dyads. In fact, it is reasonable to expect
the effects to be relatively independent. Following the stylized narrative of
Weiner’s (1971) “Macedonian syndrome,” we assume that tensions in the
primary dyad usually precede possible activation of the secondary dyad. Thus,
decision makers participating in the primary dyad would have to consider a pos-
sible shift in the power balance that intervention would entail. This sequential
logic should be roughly compatible with additive effects, as we have modeled
them above in the regression models. Yet, since we have no strong deductive
theory to back up these expectations, it is helpful to evaluate the interaction
of the two dyads by estimating both dimensions together by letting the data
speak without the constraints of parametric assumptions.
Based on a two-dimensional nonparametric generalized additive model
(GAM), Figure 6.5 displays a contour plot of a surface, indicating effects on
conflict probability over combinations of relative group size and relative TEK
size. As in the previous figures, the x-axis shows the size of the TEK group.
Along the y-axis, we plot the relative group size in the primary dyad. The result
reflects a high degree of independence between the two dimensions. Moving
from the bottom of the plot upward, the predicted probability increases, as
indicated by the contours. Moving from left to right over the x-axis, however,
implies traveling over a “ridge” with higher conflict propensity, but the values
start descending again with very high relative TEK size. Since the ridge is more
or less perpendicular to the x-axis, or size of the TEK group, we conclude that
the size effects in the primary and secondary dyads by and large are largely
orthogonal to one another.

Validating the Causal Arguments

To follow up the quantitative analysis of this chapter, we turn to a few illustra-
tive examples to highlight the causal relationships we discuss above. In Chap-
ter 4, we argued that large excluded ethnic groups are more likely to rebel than
relatively small ones. Less intuitive, perhaps, is our claim in this chapter that
there is a curvilinear relationship between the size of a group’s transnational
136 Inequality, Grievances, and Civil War

1.0

0.02
0.8

0.018

0 016
Relative Group Size

0 014
0.6
0.4

0 012
0.2

0 01

0 008

0 006 4
00
0.00
0.0

04

0.0 0.2 0.4 0.6 0.8 1.0

Relative TEK Size

figure 6.5. Contour plot showing the effect on conflict probability of relative group
and TEK sizes.

kin and civil war (see H6.3). This qualitative discussion is less suitable to assess
whether excluded or included TEK groups are more conflict prone as postu-
lated by H6.4 since it is difficult to assess proportions or relative rates without
population data. Therefore, in this section we limit ourselves to examining how
kin relations are linked to conflict rather than assessing relative frequencies for
different types of relations.
There are quite a number of politically active TEK ties, where stateless ethnic
brethren on both sides of the border assist each other. For instance, Kurdish
populations in Turkey, Iraq, and Iran have all engaged in separatist activity in
their respective countries (McDowall 2005; Marcus 2007; Gunter 2008; Gunes
2012). Kurds represent roughly 20–25% of the population in Turkey, 15–
20% in Iraq, and about 10% of the population in Iran. These three countries
vary dramatically in terms of their political institutions (Islamic theocracy,
authoritarian rule, democracy/semi-democracy) and the dominant ethnic group
controlling the state. Yet, all three have historically discriminated against the
Kurds and have experienced violence by Kurdish separatist movements.
As we saw in Chapter 4, the Kurdistan Worker’s Party (PKK) in Turkey
emerged as a socialist-nationalist movement in the 1970s and shifted to violent
tactics in the 1980s (Marcus 2007; Gunes 2012). During the period of authori-
tarian rule in Turkey, the military attempted to crush the organization by force,
Transborder Ethnic Kin and Civil War 137

while denying Kurdish social, political, and cultural rights. More recently, with
the Justice and Development Party (AK) in control of a democratic Turkish
state, greater Kurdish autonomy has been granted, even as military operations
against the PKK continue. In Iraq, two main parties – the Kurdish Democratic
Party (KDP) and the Patriotic Union of Kurdistan – opposed harsh authoritar-
ian rule under Saddam Hussein’s Sunni-Arab dominated regime. Given rough
parity in size between Sunni Arabs in Iraq and the Kurdish population, the
Kurds potentially posed an existential threat to Hussein’s regime, particularly
if allied with the Shia majority. Thus, the Hussein regime used especially dra-
conian tactics, including the 1988 poison gas attack in Halabja, which killed
an estimated 5,000 people (McDowall 2005). Finally, the Kurdish Democratic
Party of Iran (KDPI) initially challenged the Shah’s regime, but later came to
oppose the Islamic Republic. While conflicts between the KDPI and the Iranian
state have not been as bloody as those in neighboring countries, the group
continues to fight for democratic reform and cultural rights.
Even as the particular organizations differ, it is clear that Kurds in all three
countries share the nationalist goal of a greater Kurdistan. Such bonds of
solidarity are echoed on the KDP of Iraq’s webpage:

The KDP believes that the Kurds are one nation. We put forth effort for the strengthening
of brotherly relations between our party and the democratic parties and organizations
in all parts of Kurdistan. Kurdistan was divided after World War I with no regard to
the demands of the Kurdish people for self-determination. The feeling of solidarity and
sympathy among people in different parts of Kurdistan is natural and spontaneous,
and every uprising that has ever taken place in any part of Kurdistan has always been
supported by Kurds in the other parts.5

Relations between Kurdish organizations have not always been collegial

(Gunter 2008); yet, sympathies and even family ties do extend across national
boundaries. Beyond the resources they can garner domestically, it is clear that
militant Kurdish groups benefit from resources and physical sanctuaries beyond
the borders of their targeted state(s). For instance, although Iraqi Kurdish par-
ties do not publicly condone PKK activities on its side of the border, the
organization is firmly entrenched in bases in Northern Iraq. And while there
have been calls for cooperation between Turkey and Iraq on security matters,
including the PKK question, kinship ties make it very difficult for Iraqi Kurds to
turn on their brethren. Moreover, as our theory suggests, while Turkey is able
to maintain a relative degree of control over Kurdish activities domestically,
the transnational aspects of the conflict are especially hard to monitor and
police.
Albanians in Kosovo and Macedonia shared not only ethnic ties, but
also material resources and fighters during their independence struggle

5 Kurdish Democratic Party of Iraq website, “General Information about the KDP,” http://www.
kdp.se/?do=general .
138 Inequality, Grievances, and Civil War

(Phillips 2004; Gleditsch 2007; Pettifer and Vickers 2007). Although Albanian
parties had been politically represented in Macedonia since the country’s seces-
sion from the former Yugoslavia in 1991, many in the large Albanian minority
resented the effort to set up a Slav-dominated nation state and demanded
greater recognition for their cultural and political rights. The Kosovo war,
which ended in the de facto termination of Serb control of the Albanian-
dominated province, had a radicalizing effect, and many Slavs feared that the
large influx of Albanian refugees would destabilize Macedonia. The (Albanian)
National Liberation Army (NLA) in Macedonia was set up in 1999 in the after-
math of the Kosovo war. It was headed by a former commander in the Kosovo
Liberation Army,6 and it is widely alleged that many of the NLA combatants
had previously participated in the war in Kosovo.7 The NLA in Macedonia
operated out of bases in Kosovo, and virtually all of the fighting in 2001 took
place in border areas. The timing and spatial location of the conflict highlights
the importance of the transnational linkages. Moreover, transnational factors
were also important in the outcome of the conflict, as the refusal of the gov-
ernment of Albania to fully endorse the National Liberation Army may have
contributed to prevent further escalation of the 2001 conflict.
The Tuaregs are another important case of a stateless people straddling inter-
national borders. Political exclusion and economic marginalization in Mali and
Niger provoked grievances and triggered resistance to state power in both coun-
tries (Krings 1995; Humphreys and ag Mohamed 2005). There were major
refugee flows to Algeria, Mauretania, and Libya as a response to repressive
policies that targeted the group in Mali and Niger in the early 1990s. Ethnona-
tionalist mobilization among these refugees has recently come to focus on a
secessionist campaign in Mali that increasingly undermined the government’s
control of the northern part of the country in 2012.8
So far, we have surveyed conflicts where excluded kin groups increased
the risk of internal conflict. While H6.4 states that TEK connections involv-
ing included groups should be less conflict prone than those featuring state-
less groups, we still find conflict cases that involve the former. The Rwan-
dan government, led by the Tutsi-dominated Rwandan Patriotic Front, has
offered resources and support to its coethnics across the border in the Demo-
cratic Republic of the Congo (Prunier 2009). Somalia provided resources
and bases to separatist groups in Ethiopia, and even invaded directly in
1977 as it claimed the Ogaden region as its own (Kornprobst 2002). Even
though the Serbian regime did not prevent the breakup of Bosnia-Herzegovina
against the hopes of the Bosnian Serbs, the link to Belgrade was strong

6 http://www.balkaninsight.com/en/article/ali%20ahmeti.
7 See http://news.bbc.co.uk/1/hi/world/europe/1231596.stm.
8 Markus M. Haefliger, “Alt-neuer Wunsch nach Selbstbestimmung,” Neue Zürcher Zeitung,
2 April 2012, p. 7.
Transborder Ethnic Kin and Civil War 139

before and during the Bosnian war that started in 1992. Fueling irredentist
rhetoric as well as arming and organizing their ethnic kin in both Croa-
tia and Bosnia, the Milošević regime certainly raised the expectations that
violence and ethnic cleansing would result in the creation of a greater Ser-
bia (Saideman and Ayres 2008, p. 55). In fact, the identification between
the segments of the Serbian group was so strong that Milošević repre-
sented the Bosnian Serbs during the Dayton peace negotiations (Holbrooke
1998).
As we have shown, there are many examples where TEKs are likely to have
contributed to increasing the risk of conflict. However, we have also argued
that TEK ties can decrease the risk of violent conflict if the cross-border segment
is so large as to deter incumbent governments from becoming too belligerent.
As our primary case, we discuss the Russian diaspora. In the early post-Soviet
era, many observers feared that sizable Russian minorities in the “near abroad”
would cause significant conflict and irredentist pressures in newly independent
regimes (Shlapentokh, Sendich, and Payin 1994). This was especially true in
the Baltics, where the combination of a large Russian minority and stridently
anti-Russian nationalist elites led to significant fear and anxiety (Laitin 1998).
There have certainly been tensions in post-Soviet states, such as over the sta-
tus of Crimean Russians and the establishment of an independence movement
in Kazakhstan. Yet, none of these latent conflicts have escalated to signifi-
cant violence. The case of Estonia is particularly instructive. At independence,
approximately 30% of the resident population in Estonia consisted of ethnic
Russians who arrived during various waves of migration stemming back to the
pre-Soviet era. Yet, despite their large social and political influence, nationalist
elites moved to deprive Russians of Estonian citizenship and economic benefits
(Kolstoe 1995, pp. 120–2). Although the 1995 Nationality Law attempted to
resolve the question of citizenship, by 2010 some 100,000 Russians still lack
citizenship in Estonia or Russia. In other terms, 7.5% of Estonia’s population
is officially “stateless.”9
Latent tensions have at times sparked protests and violent riots. In Tallinn,
ethnic Russians protested against the demolition of a Soviet war memorial and
clashed with the police in 2007. But while Estonian Russians are a coherent and
cohesive group, constitute a sizeable minority, and have a legitimate claim of
official discrimination, they have not resorted to large scale organized violence
or formed active rebel organizations. Estonia has softened its stance somewhat
in recent years, granting greater rights and privileges to Russians. This may
in part be due to Estonia’s application and ascension to the European Union
(Kelley 2004). Yet, this cannot explain similar restraint regarding Russian

9 Clifford Levy, “Soviet Legacy Lingers as Estonia Defines its People,” New York Times, 15
August 2010.
140 Inequality, Grievances, and Civil War

minorities in Belarus, Ukraine, Moldova, Kazakhstan, and elsewhere who are

not part of the EU, nor likely candidates for membership.10
Another potential explanation that also applies to the other cases mentioned
is Estonia’s possible fear of an ascendant Russia, which is increasingly ready
to flex its muscle in what it perceives as “its” neighborhood, as evidenced
when Russia invaded Georgia in 2008. Estonia is militarily and economically
weak compared to its eastern neighbor. Indeed, some conciliatory actions were
directly related to Russian threats. For instance, in June 1993, Estonia adopted
an Alien Registration Law, which was largely seen as discriminating against
ethnic Russians (Kelley 2004). In response, Russia cut off the natural gas supply
to Estonia, leading the President to suspend the law just a few days later.11 In
addition, Russian military doctrine, approved by presidential decree in 2000,
explicitly states that foreign threats include discrimination against Russians
abroad and the suppression of their rights and interests (Zevelev 2001). Zevelev
argues (2001, p. 168) that the fact that “the leaders of all newly independent
states are well aware of the possible consequences of a crisis involving Russian
diasporas” helps preserve stability in the Baltics and elsewhere. As Russian
power and influence has grown in the Putin years, countries with notable
Russian minorities are likely to be especially cautious of jeopardizing relations.
Similar patterns can be found with respect to the overseas Chinese. Ethnic
Chinese are numerically and economically important across Southeast Asia,
including in Vietnam, Malaysia, Indonesia, and Myanmar, where they have
existed for several centuries. There have been latent anti-Chinese sentiments
across Southeast Asia, and occasional riots targeting the group, particularly in
Malaysia and Indonesia, where the Chinese are seen as economically power-
ful. Yet, the Chinese community, while relatively organized, has not sought to
establish an armed movement.12 Beginning in the late 1970s, the Chinese gov-
ernment began to articulate a policy of forging links with the overseas ethnic
Chinese communities. Accordingly, foreign governments were asked to “pro-
tect the legitimate rights and interests of overseas Chinese and respect their
national traditions, customs and habits” (Bolt 2000, p. 54). The backing of
such a powerful state may cause host countries to behave more cautiously.
In sum, cases such as Tutsis in the Congo and Kurds across the Middle
East demonstrate that ethnic kin groups can and do support their transborder
brethren. However, as the Russian case illustrates, kin groups that are especially

10 We find some evidence that actual or prospective membership in the European Union influences
the likelihood of irredentist conflict when we included a dummy variable for such countries
in our models. However, this modification does not change our main results (see the Online
Appendix).
11 Minorities at Risk, “Chronology for Russians in Estonia.” Accessed online at: http://www.
cidcm.umd.edu/mar/chronology.asp?groupId=36601.
12 The Malayan Communist Party, which has engaged in violence, was supported by many mem-
bers of the Chinese-Malaysian community. However, it cannot be said that the movement’s
goals were strictly ethnic in nature or that the membership was exclusively Chinese.
Transborder Ethnic Kin and Civil War 141

powerful – in terms of both their numeric size and their control of a militarily
superior state – can obviate the need for rebellion. In such cases, incumbent
regimes will be wary of trampling over the rights of the group in question and
incurring the wrath of their much more powerful neighbors.

Conclusion
Thanks to improved data and analysis, this chapter has been able to provide
considerable empirical support for conjectures that have so far only been postu-
lated theoretically or shown to hold for selected cases. In agreement with earlier
studies, we show that as the relative strength of transnational ethnic kin groups
increases, the risk of internal conflict also grows, but only up to a certain point.
After this point, further increases in the ethnic kin’s demographic weight have
a dampening impact on conflict propensity. Whereas the former part of the
relationship is quite straightforward, the latter part is much less obvious, but
nevertheless of great importance. Indeed, it is this conflict-reducing influence
of large “homelands” that enables us to resolve our main puzzle relating to
the relative peacefulness of the Russian diaspora. It is not surprising that the
calming influence is limited to included, rather than excluded, TEK groups,
such as the Russians in Russia.
These findings do not imply that transnational politics are always more
peaceful than domestic dynamics. In contrast to the conflict-inhibiting influence
of large state-controlling TEK groups, we find exactly the opposite effect for
intermediate size kin groups. Indeed, the probability of conflict hinges critically
on the power status of the TEK group. Stateless communities, such as the
Kurds (outside Iraq), exhibit a much higher conflict potential compared to the
within-country baseline. Thus, the pernicious effect of political exclusion has a
tendency to spill over state borders. Since marginalized communities have little
to lose at home, they also are more willing to upset the status quo abroad in
the name of ethnic solidarity and commonly felt grievances.
Although our main task has been to test empirically the curvilinear the-
oretical conjecture and to evaluate TEK groups’ net impact on conflict, our
empirical analysis has important additional theoretical implications. By intro-
ducing the conceptual distinction between primary and secondary dyads, we
are able to unpack the “ethnic triad” originally introduced by Weiner (1971),
while at the same time going beyond analyses that study the influence of ethnic
kin as a dichotomous feature (e.g., Cederman, Girardin, and Gleditsch 2009).
Indeed, our results show that the power balance in the primary and secondary
dyads affects the risk of civil war in potentially quite different ways. The central
curvilinear finding for the secondary, transnational dyad dovetails nicely with
international relations theory, especially the strands thereof highlighting the
importance of uncertainty and deterrence. As mentioned, several studies using
conventional balance-of-power theory have shown that the risk of conflict is
the highest at power parity. However, we have not been able to find strong
142 Inequality, Grievances, and Civil War

evidence of such an inverted U-shape relationship in the primary, domestic

dyad. It seems intuitive that the secondary dyad involves strategic considera-
tions that have more to do with interstate relations than with domestic politics,
especially since transnational relations can be expected to feature more uncer-
tainty than domestic dyads. Yet, more research is needed to clarify the causal
mechanisms operating in the primary and secondary dyads, as well as possible
interdependencies between them.
7

Country-Level Inequalities and Civil War

The previous three chapters have focused on the group level and explored
how political exclusion and economic inequalities may generate widespread
grievances conducive to ethnonationalist mobilization and armed conflict with
the state.1 In this chapter, we take a step back to the country level and exam-
ine how particular configurations of political and economic privileges among
groups in a country give rise to differences in the risk of civil war.
Given our focus on disaggregation and the importance of a dyadic approach
to civil war, the move back up to the country level might at first seem puzzling.
However, comparing the causes of civil war across different levels of analysis is
helpful for a number of reasons. Many existing analyses of civil war have been
conducted at the country level, which makes it difficult to compare their findings
directly with our group-level results. By aggregating the properties of groups
and other actors up to the country level, we are able to explicitly explore the
relationship between group-level and country-level findings and whether these
may differ due to potential scaling effects. Moreover, the number and size of
ethnic groups varies greatly between countries. Many existing country profile
measures, such as ethno-linguistic fractionalization indices, depend on atheoret-
ical population-weighting procedures that disregard the political status of eth-
nic groups, and where small groups by construction will carry little weight in the
measure. However, as we will show, it is possible to construct more informative
country-level measures, reflecting group-level characteristics in a more theoret-
ically informed manner. As such, a country-level analysis allows us to compare
the explanatory power of our new proposed measures of horizontal inequalities
with existing proxies for societal grievances related to restrictions on political
participation and uneven distributions of wealth among individuals.

1 This chapter builds on an article by Buhaug, Cederman, and Gleditsch (forthcoming).

143
144 Inequality, Grievances, and Civil War

Country-level analysis also allows us to extend the comparisons of the risk

of civil war from the group-level analysis. While some countries have multiple
ethnic groups, others have no politically relevant ethnic cleavages. The group-
level models presented in the previous chapters by construction cannot consider
the risk of conflict in countries where ethnic identity has no political relevance.
Hence, aggregate analysis of this type allows us to compare the risk-inducing
effects of political exclusion and economic marginalization among countries
with politically relevant groups to the risk of civil war in countries without
politically relevant ethnic groups. The move to the country level also allows us
to expand the scope of our analysis by moving beyond strictly ethnic conflict to
consider all forms of severe insurgencies and civil wars that pit non-state chal-
lengers against the government, regardless of whether nonstate organizations
have some kind of ethnic affiliation or not.
Although our unit of analysis is quite different from previous chapters, the
theoretical arguments follow a common logic. Building on results from the
group-level analyses, we highlight the empirical validation of grievance argu-
ments at the country level. In the following two sections, we derive hypotheses
on the effects of horizontal and vertical inequality, respectively, on the risk of
civil war.

Aggregating Horizontal Inequalities to the Country Level

Originally framed as a dynamic theory of frustration over time (Davies 1962),
the logic of relative deprivation can also be extended to a strictly comparative
setting (Gurr 1970). Particular individuals or members of specific social groups
may be dissatisfied with their status or resource allocation vis-à-vis other indi-
viduals or groups (see Chapter 3). Shifting the theoretical focus from individuals
to groups, the three previous chapters investigated how economic and political
horizontal inequalities are particularly prone to mobilization and conflict if
these overlap with other social cleavages, notably ethnic ones. More precisely,
we have showed that civil wars are more common when ethnic groups are
excluded from access to executive power, and the risk of conflict increases even
more when politically excluded groups are also subject to economic marginal-
ization.
However, many of the existing debates about civil war have been couched
in terms of the risk that specific countries will experience civil war, and it is not
directly clear from our previous results how horizontal inequality and rebellion
at the group level may transfer to the country level or help reveal information
about what countries are more likely to see armed challenges to the state. At an
immediate level, the units of analysis differ, and one country can give rise to a
different number of distinct group observations that are all collapsed to one
observation in country-level analyses. More fundamentally, the so-called levels-
of-analysis, or scaling, problem tells us that theories and empirical patterns at
one level often cannot be applied across levels of aggregation, for example from
Country-Level Inequalities and Civil War 145

individuals via groups and organizations to countries. Doing so risks various

forms of inference to the wrong level or ecological fallacies (Robinson 1950).
For example, many studies have examined theoretical accounts of civil war
that focus on inequality and distribution in resource dependent economies or
rough terrain at the country level, by identifying different forms of country
profiles intended to reflect how common such features are (e.g., Fearon and
Laitin 2003; Ross 2012). However, the fact that a country has high inequality
or a high share of mountainous terrain does not by itself tell us whether the
specific actions of the rebels are consistent with the theory or whether the
conflict occurs in the place or manner suggested by the theory. Hence, many
such propositions are best assessed at a local level (Buhaug and Lujala 2005).
However, too much disaggregation is not necessarily always better and may
miss the proverbial forest for the trees (Cederman and Gleditsch 2009). Many
studies of civil war analyze the consequences for the state and argue that
features such as natural resource dependence or rough terrain primarily affect
the risk of war through its effect on state fragility, and that features of ethnic
groups or local characteristics are generally irrelevant once we consider state
attributes (e.g., Fearon and Laitin 2003). Contrary to these claims, we argue
that the distribution of political and material inequalities and likely grievances
among social groups within a state tell us a great deal about its general conflict
propensity.
In this chapter, we introduce new empirical indicators that operate according
to the principle of the weakest link as a way to overcome the information loss
associated with aggregation from substate- to state-level analysis (Hirshleifer
1983; Goertz 2007). Instead of relying on measures that use country-level
averages or aggregates, we try to identify measures reflecting the specific actors
or subunits most at risk of civil war within a country. Aggregate measures
based on averages or summed features will by construction discount small
and atypical groups, especially in large countries.2 For example, there are
many ethnic groups in Russia, some of which are considerably poorer than the
dominant Russian population. Some of the groups in the Caucasus, such as
the Chechens, fall dramatically below the country mean in terms of income per
capita (see Chapter 5).3 However, given their small relative size, these groups
will essentially have no influence on national average statistics.
In Chapter 4 we proposed, and demonstrated empirically, that the risk
of conflict among ethnic groups decreases gradually by their access to state
power. The most conflict-prone groups are excluded groups that explicitly seek

2 This does not mean that the average level of exclusion is not statistically associated with civil-war
onset; see Wimmer, Cederman, and Min (2009).
3 Cederman and Girardin (2007) introduce a country-level measure of the risk of civil war, N*,
based on the product of the dyadic conflict probabilities for each individual ethnic group.
The results obtained with that specification are very similar to those emanating from the simpler
“weakest link” indicator used in this chapter. See the Online Appendix of this book for sensitivity
analysis featuring N*.
146 Inequality, Grievances, and Civil War

independence from the state, as well as groups subject to active discrimination

by the central government. A parallel logic underpins the theoretical argument
developed in Chapter 5, according to which economically marginalized groups
challenge the state over the distribution of wealth and public goods through
violent means. In other words, relative political and economic marginalization
generates grievances among groups and increases the likelihood that they will be
able to overcome collective action problems and mobilize for armed opposition.
An individual country may contain a large number of dyadic relations
between included and excluded groups. In line with the weakest link logic,
we postulate that the overall risk of conflict in a country is a function of the
degree of political and economic marginalization of the least privileged group.
Drawing on arguments about the importance of relative group size for oppor-
tunities to wage conflict, we posit that the likelihood of violent conflict in a
country should increase with the demographic size of the largest politically
excluded or discriminated group relative to the incumbent ethnic group(s) in
power. Similarly, expecting a higher risk of conflict the larger the gap, we
consider the role of horizontal economic inequality in terms of the difference
between the poorest group and the national average.
In a manner analogous to our group-level hypotheses H4.1 and H5.1, the
first pair of country-level hypotheses can thus be expressed as follows:

H7.1. Countries where a large group suffers from limited access to power are more
likely to experience conflict than countries with less exclusion.
H7.2. Countries where the least wealthy group falls far below the national average
are more likely to experience conflict than more equal countries.

Aggregating Horizontal Inequalities to the Country Level

Extending the focus beyond ethnic politics, we now turn to more generalized
measures of inequality based on individual differences rather than between
groups. Many scholars argue that interpersonal inequalities are central to gen-
erating political grievances (e.g., Boix 2008; see also references in Chapter 5).
If so, we need to consider what could be the appropriate measures to identify
what country profiles have higher risk. Most attempts to evaluate individual-
level accounts of how grievances may be related to civil war use ethnic diver-
sity or other ethno-demographic characteristics as proxies for ethno-political
grievances. The rationale behind this approach is that more ethnically diverse
populations will be characterized by larger heterogeneity in values and pref-
erences as well as corresponding dissatisfaction with any single policy. This
is especially likely to be the case in the periphery of countries, assuming that
preferences are distributed so that similarity declines at greater distance from
the center or capital (e.g., Alesina and Spolaore 2005).
Country-Level Inequalities and Civil War 147

Political economists have long argued that ethnic diversity can lead to insta-
bility and unrest. In a classical study, Rabushka and Shepsle (1972) contended
that ethnic pluralism is incompatible with democratic stability. A series of
more recent studies indicates that ethnically diverse societies face greater polit-
ical contention arising from diverging preferences and differential skills and
habits (for overviews, see Alesina and La Ferrara 2005; Kanbur, Rajaram, and
Varsheney 2010). Drawing on socio-biological reasoning about ethnic groups
and nepotism, Vanhanen (1999) reaches a similar conclusion about hetero-
geneity and conflict and finds that significant ethnic divisions tend to produce
violent conflict in an extensive cross-national sample. More broadly, Sambanis
(2001) and Fearon and Laitin (2003) associate ethnic diversity with a class
of arguments outlining the role of ethnic and nationalist grievances in conflict
processes, whether primordialist as Vanhanen (1999), or explicitly modernist
along the lines of Gellner (1983), Anderson (1991), and other prominent the-
orists of nationalism.4
Yet, ethnic fractionalization, polarization, and related ethno-demographic
indices fail to account for political inequality. We argue that the most natu-
ral measure of vertical political inequality is simply to consider democracy as
an indicator of population-wide access to political power. In contrast to the
demographic and societal indices based on individual-level data, a democracy
indicator (or some subcomponent, such as participation) explicitly reflects the
level of political exclusion in a country. In strictly authoritarian systems, politi-
cal influence is a zero-sum game in the sense that the privileges of the elite come
at the expense of the excluded population. In this sense, our notion of vertical
political inequality resembles one of Dahl’s (1971, p. 6) two main dimensions
of democracy, namely the right to participate. This logic, in turn, is part of
a more general argument about the inverse relation between democracy and
social grievances. In fact, Fearon and Laitin (2003, p. 79) rely explicitly on
regime type as a proxy for political frustrations:
Other things being equal, political democracy should be associated with less discrim-
ination and repression along cultural or other lines, since democracy endows citizens
with political power (the vote) they do not have in dictatorships. Even more directly,
measures of state observance of civil rights such as freedom of association, expression,
and due process should be associated with less repression and thus lower grievances.

4 Others suggest a nonlinear relationship where the risk of conflict is assumed to be the highest for
intermediate levels of diversity and lower in both highly diverse and fully homogenous societies,
based on the notion that greater diversity beyond some level can encourage cooperation (e.g.,
Sambanis 2001; Collier and Hoeffler 2004). Some also argue that is not so much diversity
that increases the risk of conflict but polarization, or a situation where two roughly equal
ethnic groups may compete for influence (Horowitz 1985; Montalvo and Reynal-Querol 2005).
Although these arguments are clearly distinct and point to different operational measures, they
all emphasize demographic configurations as the origin of insecurity and conflict rather than the
role of political power and access.
148 Inequality, Grievances, and Civil War

State policies that discriminate in favor of a particular group’s language or religion

should be associated with greater minority grievances.

In the light of these arguments, we postulate the following effect of vertical

political inequality on the risk of armed conflict:

H7.3. Democratic countries are less likely to experience conflict than nondemocra-
cies.

Even the most egalitarian society is characterized by some level of asymme-

try in resource allocation and public goods provision. Such inequalities need
not follow other social identities shaped by features such as ethnic affiliation or
religious denomination. In culturally homogenous countries (and conceivably
heterogeneous societies too), resource distributions may themselves delimit
social classes, and grievances resulting from persistent inequalities are likely
drivers of discontent and revolutionary activities (e.g., Davies 1962). Argu-
ments inspired by Marxist theory constitute a special case of this approach, for
example as peasants may rebel in order to force a transfer of land ownership
and other privileges from the landed elite to the masses (see Chapter 5).
Attempting to assess the implications of economic inequality for conflict,
prior empirical research has typically relied on summary indicators of individ-
ual or household income distributions. The most prominent example of such
measures is the Gini coefficient, which reflects the extent to which an observed
individual income distribution differs from a completely equal distribution. A
Gini value approaching 0 would indicate perfect equality, while values closer
to 100 (or 1, if measured as a fraction) reflect greater inequality. The afore-
mentioned arguments about social inequality and grievances suggest a causal
chain where persistent inequality leads to grievances in disadvantaged sections
of a country’s population, in turn fueling demands for political change and
redistribution. Denied such reforms, and possibly even encountering state-led
repression, the aggrieved may see little choice but to rebel by violent means:

H7.4. Countries with high levels of individual-level inequality are more likely to
experience conflict than more equal ones.

We note that our summary of the literature in Chapter 5 shows that most
studies find no relationship between violence and general social inequality as
reflected by the Gini coefficient (e.g., Fearon and Laitin 2003; Collier and
Hoeffler 2004). However, existing studies have failed to jointly consider both
horizontal and vertical inequalities, that is, across groups and individuals
respectively.
Country-Level Inequalities and Civil War 149

Data and Research Design

Our analyses in this chapter aggregate information from the group-level data
used in Chapters 4 and 5 up to the country level. The complete data set contains
yearly observations of all independent countries between 1991 and 2009 in the
Gleditsch and Ward (1999) list. This amounts to around 2,900 observations,
with the specific number of observations depending on model specification and
data availability.
The dependent variable, civil war onset, is derived from the ACD2EPR
data, in turn based on the Non-State Actors data (Cunningham et al. 2009)
and UCDP/PRIO Armed Conflict Data (Gleditsch et al. 2002; Themnér and
Wallensteen 2011). In line with the other chapters, we adopt the most inclusive
definition of civil war, counting all conflicts with a minimum of 25 battle-
related deaths per calendar year. In its simplest form, we use a dichotomous
civil war onset indicator that codes initial year of fighting between the main
actors as 1 and all years of subsequent conflict as 0. In case of lulls in the
fighting or failed peace agreements, we consider renewed fighting to constitute
a new civil war onset only if periods of active conflict are separated by at least
two calendar years. Aggregated to the country level, the onset variable takes a
value of 1 in country years with one or more onsets of intrastate armed conflict.
In total, our data contain 99 civil war onsets for the post-1990 period.5
As a complement to the simple dichotomous civil war onset indicator, we
also specify a categorical onset indicator that separates between different types
of civil wars based on the specific incompatibility of a conflict. Our proxies for
group-level grievances rely on the categorization of politically relevant ethnic
groups, and we expect these to relate primarily to the risk of ethnic conflict,
pitting excluded groups against ethnic groups in power. Moreover, previous
research has shown that different factors and conditions appear to shape the
risk of territorial insurgencies and conflicts over the control of the central
government (e.g., Buhaug 2006; Wucherpfennig et al. 2012).
We disaggregate civil war onset along two different dimensions, asking
whether the conflict is ethnic and whether the conflict concerns issues of self-
determination or claims on the central government. The subset of ethnic con-
flicts is identified by the ACD2EPR data, linking the UCDP Non-State Actors
to the ethnic groups in the EPR-ETH data. Conflicts are coded as ethnic where

5 We do not exclude observations with ongoing conflict from the main models since a country with
ongoing conflict in our setup is still at risk of a new conflict onset, as a conflict could arise over a
distinct incompatibility. In the UCDP/PRIO data, countries may host several intrastate conflicts
simultaneously over different incompatibilities, so it is not the case that countries undergoing
civil war cannot be at risk of another conflict. Still, one might argue that observations with
an ongoing conflict may be systematically different from countries without any conflict. In
sensitivity tests reported in the Online Appendix, we show that the reported results are robust
to changes in the treatment of these observations, including dropping ongoing conflicts from the
sample altogether.
150 Inequality, Grievances, and Civil War

claims were made on behalf of specific ethnic groups and mobilization and
recruitment occur along ethnic lines. The nature of the claims that organiza-
tions make – whether related to the central government or a specific territory –
determines whether a conflict is classified as governmental or territorial, based
on the incompatibility indicator in the UCDP/PRIO Armed Conflict Data Set.
This conflict typology in principle encompasses four categories. However, we
separate between territorial and governmental conflicts only for the subset of
ethnic conflicts, because virtually all nonethnic conflicts concern aspects of the
central government.
Finally, we aggregate the conflict data to a country-year format. More specif-
ically, our dependent variable takes on four possible values:

DVcat = 0; no onset
DVcat = 1; ethnic territorial conflict onset
DVcat = 2; ethnic governmental conflict onset
DVcat = 3; non-ethnic conflict onset

In the post–Cold War sample, 1991–2009, the dominant form of civil war
is ethnic territorial conflict with 42 onsets, followed by 36 nonethnic conflict
onsets, and 21 ethnic governmental conflict onsets.
We use a number of indicators to test the proposed links between grievances
stemming from intergroup and individual differences in political and economic
privileges.
r Horizontal ethno-political inequality (H7.1) is represented by two alter-
native indicators. The most general variable, max exclusion, considers the
demographic size of the largest politically excluded group relative to the
combined size of the excluded group and the ethnic group(s) in power. This
indicator is a country-level version of the relative group size indicator, g,
introduced in Chapter 4. In addition, we include an indicator that gives
the relative demographic power of the largest ethnic group subject to active
discrimination, max discrimination. Both indicators are bounded within
the interval [0, 1), where higher values denote higher levels of political
inequality.
r Horizontal economic inequality (H7.2) is captured by the difference between
the national average per capita income level and the per capita income of
the poorest ethnic group in the country. More specifically, we define max
low ratio as the relative income gap between the poorest group and the
national average. In other words, this indicator takes the largest value of the
low_ratio indicator introduced in Chapter 5 for each country year.
r Vertical political inequality (H7.3) is captured by the extent of democracy,
based on an updated version of the Scalar Index of Polities (SIP) data set
(Gates et al. 2006). The SIP scale combines data on executive recruitment
and constraints from the Polity IV data set (Marshall, Jaggers, and Gurr
2011) with data on political participation from the Polyarchy data set
Country-Level Inequalities and Civil War 151

(Vanhanen 2000). In models shown, we use the linear democracy term with
a one-year time lag. This indicator has a theoretical range from 0 (autocratic
ideal type) to 1 (democratic ideal type). In tests not shown, we also include
a squared term to allow for a nonlinear democracy-civil war relationship.
This specification did not change the other substantive results reported.
r Lastly, vertical income inequality (H7.4) is represented by a Gini coefficient
of income dispersion from the World Income Inequality Database (WIID, v.
2.0, see UNU-WIDER 2008). These data cover 140 countries in various years
between 1946 and 2005. To minimize missing data problems, we applied
linear interpolation between data points and extended the time series by
copying the earliest/latest known value to earlier/later years for each country.
In our sample, the variable takes on values between 20.4 (Slovakia 1993)
and 73.9 (Namibia, multiple years).

In the interest of parsimony and following the model specification of earlier

chapters, we consider three control variables plausible related to vertical and
horizontal inequality and conflict – ethnic diversity, represented by the ELF
index (Fearon and Laitin 2003), logged GDP per capita, and logged population
size (Heston et al. 2011). The two latter indicators are lagged by one year.
In addition, we include a set of controls for time dependence in the risk
of conflict. To capture the legacy of past conflict, we again use Beck, Katz,
and Tucker’s (1998) nonparametric approach with a count of time since the
previous conflict or first year of observation, measured in years (peace years),
as well as three natural cubic splines to capture the anticipated non-linear
decaying effect of time. Lastly, we control for the (log-transformed) number of
previous conflicts in each country in the sample period.

Analysis
Before presenting the results from the multivariate regression models, it is
useful to briefly compare our grievance indicators and assess to what extent
these reflect different features. Figure 7.1 offers a comparison of the most
general group-based measure of ethno-political inequality (max exclusion) and
the SIP index of democracy. It is clear that the vertical and horizontal inequality
indicators tap partly different societal phenomena, even though there is a cluster
of observations in the lower right quadrant (consolidated democracies with low
levels of horizontal exclusion). Note that although we observe no countries with
high democracy score and high exclusion (a logically impossible combination),
there are several authoritarian regimes where restrictions on political influence
are not accompanied with systematic ethnic exclusion (examples include Cuba,
North Korea, and many Middle Eastern countries). Overall, the two indicators
for political grievances are moderately negatively correlated (r = −0.30).
Along similar lines, Figure 7.2 visualizes the correspondence between the
Gini coefficient and our proxy for horizontal economic inequality (max low
152 Inequality, Grievances, and Civil War

RWA BUI
SYR
.8

IRQ

BAH
BHU GUI
SUD TOG
.6

CON ANG JOR

Max Exclusion

GUA
.4 .2
0

0 .2 .4 .6 .8 1
Democracy
figure 7.1. Vertical versus horizontal political inequality in 2000.

ratio), capturing the relative income gap between the poorest ethnic group
and the national average. Countries with low levels of economic disparities
between groups display a wide variety in interpersonal wealth differences. Some
countries, such as Zimbabwe and Namibia, have high interpersonal inequality
but small differences between ethnic groups. Similarly, there is substantial
variation in intergroup inequality among countries with similar Gini scores.
Russia, for example, is characterized by relatively low-income inequality, which
conceals large systematic differences for specific ethnic groups, such as the
Chechens and other groups in the Caucasus. It is clear from the figure that
interpersonal and intergroup inequalities are hardly related at all (r = 0.04).
Figures 7.1 and 7.2 show that our group-based inequality indicators are
qualitatively different from standard measures of democracy and income dis-
persion, and that a number of cases with high interpersonal inequality have
low intergroup inequalities and vice versa. We proceed with a brief descrip-
tive assessment of the relationship between societal inequality and civil war
onset. Table 7.1 lists the mean values for the various proxies for economic
and political grievances by conflict outcome, 1991–2009. Unsurprisingly,
we find that the 1,711 country years without civil war onset in this period
are less exclusionary and contain fewer interethnic economic differences than
the average country at civil war. On average, the no-onset sample is also more
Country-Level Inequalities and Civil War 153

7
RUS
5
Max Low Ratio

SAF

INS
3

UKG

SUD
CHN PER
NAM
ZIM
1

20 40 60 80
Gini
figure 7.2. Vertical versus horizontal economic inequality in 2000.

democratic.6 Only for the Gini coefficient of vertical income dispersion do we

see nearly indistinguishable values for the two subsets, perhaps reflecting the
lack of an association suggested in previous research.
However, a more nuanced picture emerges if we disaggregate the dependent
variable by conflict type. Countries that have seen one or more new ethnic
territorial conflict in the period tend to have substantial intergroup economic
differences, as represented by the max low ratio indicator. Yet, intriguingly,
they appear to exhibit less interpersonal inequality and are on average almost
as democratic as the peaceful sample. This fact reflects how several developed
and consolidated democracies have been unable to avoid armed separatist con-
flict (e.g., India, Israel, and the United Kingdom). Conversely, we find that
countries that faced new ethnic challenges to governmental power are char-
acterized by significant ethnopolitical inequality. In fact, the mean exclusion
score for these countries is three times higher than for countries without civil
war, and the relative difference in terms of active discrimination is larger still.

6 The “no conflict” subsample is hardly homogenous, however, as it includes both most Western
democracies and peaceful countries in high-risk environments such as Bangladesh and Kenya,
as well as countries with ongoing insurgencies but no new civil war onset since 1991 (e.g., Sri
Lanka and Uganda). Note that countries that experienced multiple types of conflicts in the period
are accounted for under each relevant outcome type. Hence, the total N for the three civil war
categories is higher than the N for the aggregate onset category.
154 Inequality, Grievances, and Civil War

table 7.1. Mean Sample Values by Conflict Outcome, 1991–2009

Max Max Max Low

Outcome Exclusion Discr. Ratio Democracy Gini Obs.
No onset 0.10 0.03 1.13 0.69 40.92 1,711
Civil war onset 0.20 0.07 1.38 0.50 41.05 756
Ethnic territorial 0.11 0.04 1.72 0.59 37.42 293
Ethnic govermental 0.28 0.13 1.16 0.41 42.82 223
Nonethnic 0.18 0.04 1.42 0.49 42.49 466

The subset of countries experiencing nonethnic (and almost exclusively govern-

mental) rebellion falls somewhere in between the ethnic conflict profiles and is
characterized by moderate values on the inequality indicators. Overall, Table
7.1 offers tentative support for Hypotheses 7.1, 7.2, and 7.3, although they
appear to hold for different subcategories of civil war.
Next, we evaluate the horizontal and vertical inequality indicators in a
multivariate setting. We first consider a series of models that test the proposed
hypotheses with respect to general civil war risk, using the dichotomous conflict
onset variable. Table 7.2 presents four models: two models that include the
vertical inequality (VI) proxies and differ only in choice of horizontal political
inequality measure (exclusion vs discrimination), and two models that include
only the horizontal inequality (HI) indicators.
As shown in Models 7.1 and 7.2, none of the estimates for the interpersonal
inequality measures are significantly different from zero at conventional sta-
tistical levels. This runs counter to the expectations expressed in Hypotheses
7.3 and 7.4, although it is consistent with earlier failures to find systematic or
consistent links between vertical inequalities and civil war risk (e.g., Fearon
and Laitin 2003).7 Perhaps more surprisingly in light of the results reported in
Chapter 4, Model 7.1 also fails to produce evidence that a high level of ethnop-
olitical exclusion makes conflict more likely. Model 7.2 reveals that intergroup
political inequality in terms of the smaller set of discriminated groups also
seems largely irrelevant. The only proxy for social grievances that behaves in
accordance with the hypothesis is horizontal economic inequality; civil wars
are overrepresented among countries with large income differentials, other
factors held constant. Still, the effect is quite weak at a substantive level. Of
the other control variables, the most influential covariates in Models 7.1 and
7.2 are GDP per capita and conflict history, whereas population size and ethnic

7 Note that this nonresult is not due to a failure to specify a nonlinear relationship between
democracy and the log odds of civil war onset (e.g., Hegre et al. 2001). In the Online Appendix,
we provide results demonstrating that specifications with other functional forms for democracy
do not alter the results in any substantive manner.
Country-Level Inequalities and Civil War 155

table 7.2. Logit Model of Civil War Onset, 1991–2009

Model 7.1 Model 7.2 Model 7.3 Model 7.4

Full Model Full Model HI Model HI Model
Democracy − 0.2302 − 0.0974
(0.5487) (0.5569)
Gini − 0.0211 − 0.0228
(0.0150) (0.0155)
Max exclusion − 0.1115 0.0610
(0.7258) (0.5261)
Max discrimination 1.0537 1.1891
(0.8463) (0.6313)
Max low ratio 0.2856** 0.3042** 0.2941** 0.3147**
(0.0952) (0.0987) (0.0961) (0.1003)
Ethnic diversity 0.7875 0.8271 0.6051 0.6262
(0.5594) (0.5566) (0.4914) (0.4903)
Population size 0.1365 0.1451 0.2017 0.2154
(0.1218) (0.1234) (0.1087) (0.1129)
GDP/capita − 0.2581 − 0.2742 − 0.3209** − 0.3290**
(0.1727) (0.1723) (0.1145) (0.1141)
Previos conflicts 0.4714 0.4355 0.3867 0.3502
(0.3379) (0.3412) (0.3272) (0.3292)
Peace years 0.3531** 0.3684** 0.3753** 0.4024**
(0.1269) (0.1292) (0.1142) (0.1157)
Constant − 2.8187 − 2.8974 − 3.8044** − 3.9780**
(1.9517) (1.9956) (1.3108) (1.3815)
Observations 2,467 2,467 2,860 2,860
Robust standard errors in parentheses; estimates for three natural cubic splines not shown.
** p < 0.01, * p < 0.05.

fractionalization are only weakly correlated with civil war outbreak in the post–
Cold War period.
Missing data pose a potentially serious problem in the first two models in
Table 7.2 (as well as in earlier attempts to assess the effect of interpersonal
inequalities on conflict). Despite interpolation between data points and gentle
extrapolation to expand the time series, we still have missing Gini coefficients
for many cases. The SIP democracy indicator, too, has missing values for a
number of observations in our sample. We have little reason to believe that
the excluded observations are missing completely at random. Consequently,
in the last two models of Table 7.2, we drop the proxies that do not affect
vertical inequality and reinvestigate the impact of horizontal inequality on
general conflict risk. Aside from a moderate increase in sample size, this has
little impact on the performance of the model. The effect for relative size of
the largest excluded group is now positive, as expected, but still statistically
156 Inequality, Grievances, and Civil War

insignificant (see Model 7.3). The estimated effect for discrimination becomes
somewhat larger (see Model 7.4). Economic marginalization of ethnic groups
remains positive and significant throughout, which adds further support to the
rationale underlying Hypothesis 7.2. The most visible changes from the full
models are found among the control variables. The coefficient for population
size increases by about 30% although it still just misses the conventional 95%
confidence level. This improvement is primarily due to inclusion of a number of
smaller countries that lack data on income dispersion and/or democracy. Also,
we now report lowered standard errors and higher confidence in the parameter
estimates for GDP per capita. This is not a result of the increased sample size but
rather due to exclusion of the collinear democracy index, which has a bivariate
correlation with logged per capita income of r = 0.47. The ethnolingustic
fractionalization (ELF) index remains insignificant throughout.
As we have seen, there is little evidence of an association between income
or political power differentials and a higher risk of civil war in general when
they do not overlap with politically relevant ethnic cleavages. In contrast, the
general models presented in Table 7.2 indicate that regimes that discriminate
sizable ethnic groups are more often challenged by rebel groups, although
this effect is considerably weaker for the more sweeping measure of political
exclusion. Moreover, we find large economic inequalities across ethnic groups
to be consistently and positively linked to civil war.
Yet, it would be premature to draw conclusions about the relationship
between grievances and civil war based on these models alone, since we believe
particular configurations of material and political status should motivate con-
flicts of particular kinds (see Table 7.1). In Table 7.3, we evaluate our inequality
indicators across the three categories of civil war outlined above. We offer two
complementary models, the first one using the extent of political exclusion as
proxy for political HI, and the second one focusing specifically on ethnopoliti-
cal discrimination.
Again, we find that structural restrictions on political participation consti-
tute a poor predictor of civil war risk; the effect of democracy is insignificant
across all outcome types, with the exception of the negative association with
ethnic governmental conflict found in Model 7.6, and the marginal impact is
generally quite small. The estimate for the Gini coefficient also fails to behave
as expected. Indeed, the estimated effect of vertical income inequality is nega-
tive and significant for ethnic territorial conflict in both Models 7.5 and 7.6. In
line with theories of class-based revolutions, we find a positive coefficient for
the individual-level inequality indicator for nonethnic conflict, but this effect is
far from statistically reliable. Hypotheses 7.3 and 7.4, postulating a systematic
relationship between higher inequalities among individuals and higher conflict
risk, both fail to receive support.
By contrast, we find considerably stronger support for political inequal-
ity among ethnic groups increasing the risk of ethnic civil war in a country.
However, this relationship is more subtle than the simple logit models above
 table 7.3. Multinomial Logit Models of Civil War Onset, 1991–2009

Model 7.5 Model 7.6

Ethnic Territorial Ethnic Gov’t Nonethnic Ethnic Territorial Ethnic Gov’t Nonethnic
Democracy − 0.2066 − 1.0381 − 0.1378 0.1116 − 1.0110 − 0.1137
(0.6585) (1.3649) (0.6591) (0.7819) (1.2511) (0.6709)
Gini − 0.0604* − 0.0205 0.0049 − 0.0611* − 0.0172 0.0047
(0.0243) (0.0339) (0.0196) (0.0283) (0.0355) (0.0195)
Max excluion − 6.0190** 2.6146* − 0.1297
(1.8503) (1.1081) (0.8752)
Max discrimination − 1.7719 3.4690** 0.1009
(2.3153) (1.0932) (1.7558)
Max low ratio 0.2776** 0.0105 0.2071 0.3067** 0.1573 0.2107
(0.0896) (0.6150) (0.1342) (0.1086) (0.4701) (0.1333)
Ethnic diversity 1.2416 0.1514 0.4263 1.1920 0.5375 0.4128
(1.2131) (1.0605) (0.7204) (1.4342) (1.1589) (0.7093)
Population size − 0.0452 − 0.0652 0.2634 0.0388 − 0.1299 0.2640
(0.1932) (0.3179) (0.1711) (0.2082) (0.3193) (0.1730)
GDP/capita − 0.1280 − 0.3552 − 0.2484 − 0.1001 − 0.4908 − 0.2509
(0.2659) (0.4150) (0.2229) (0.2702) (0.4263) (0.2228)
Previous conflicts 1.4108* 0.2162 − 0.2729 1.3567 0.1167 − 0.2814
(0.6682) (0.5530) (0.3943) (0.7471) (0.4548) (0.3915)
Peace years 0.3533* 0.6991** 0.2078 0.3644* 0.7903** 0.2096
(0.1748) (0.2175) (0.2166) (0.1694) (0.2857) (0.2052)
Constant − 2.2612 − 2.0174 − 4.8279* − 4.1131 − 0.5813 − 4.8396*
(3.3828) (4.8360) (2.3400) (3.3995) (5.5293) (2.3349)
Observations 2,467 2,467
Robust standard errors are in parentheses; three natural cubic splines estimated but not shown.
157

** p < 0.01, * p < 0.05.

158 Inequality, Grievances, and Civil War

suggest. Political exclusion, or outright discrimination, of large ethnic groups

increases the estimated risk of ethnic governmental conflict, and the effect is
quite substantial. Holding all other covariates in Model 7.6 at median values, a
shift in max discrimination from the 5th percentile value to the 95th percentile
value yields a threefold increase in conflict risk.
The estimate for ethnopolitical exclusion suggests that it contributes to
reducing the risk of ethnic territorial conflict at the country level, and a simi-
lar, if weaker, pattern is found for discrimination in Model 7.6. Although this
appears puzzling at first, we believe that this primarily reflects the relationship
between relative group size and capacity. Excluded groups that are large and
relatively strong compared to incumbent ethnic groups are unlikely to settle for
self-determination in a limited territory but instead are likely to perceive that
they have a legitimate claim to central governmental power given their relative
demographic size. Hypothesis 7.1 is thus supported, but only for the subset of
civil wars that include interethnic competition for central governmental power.
Complementing the effect of horizontal political exclusion, we find that vari-
ation in economic marginalization of ethnic groups contributes to explaining
ethnic territorial (but not ethnic governmental) conflict onset. Again, we believe
this pattern is at least in part a function of the size of marginalized groups. Our
proxy for horizontal income inequality compares the mean income for the
poorest group in each country with the national average. Smaller groups are
more likely to exhibit larger discrepancies from the mean as these groups will
have little impact on national income statistics. Since very poor groups tend to
be small relative to powerful ones, and also more likely to reside in peripheral
areas of a country, increased autonomy or secession may be their only viable
political objectives. These observations seem to fit well with contemporary
insurgencies in Northeast India and the Philippines.
Most other covariates in the two models in Table 7.3 fail to yield signif-
icant coefficients within standard confidence intervals. This primarily reflects
the short time span of the sample and the relatively few conflict observations
for each outcome category. Nevertheless, in view of this limitation, it is even
more remarkable that our inequality indicators seem to perform better than the
standard covariates used in country-level studies of civil war, such as popula-
tion size, per capita income, and ethnic fractionalization (see, e.g., Hegre and
Sambanis 2006).
We use visualizations to further demonstrate the effects of intergroup
inequalities on civil war onset. Figure 7.3 shows the estimated risk of ethnic
governmental conflict as a function of political discrimination, with a confi-
dence interval based on simulated data from estimates in Model 7.6. This makes
it clear that states are much more likely to be challenged by nonstate actors as
the relative size of the largest discriminated group increases. Likewise, Figure
7.4 plots the risk of ethnic territorial conflict over relative wealth gap between
average national income and income per capita for the poorest ethnic group.
Again, we see a clear positive association, although the substantive effect is
smaller than for discrimination.
Country-Level Inequalities and Civil War 159

3
Pred cted Probab ty of Conf ct
2
1
0

0 .2 .4 .6 .8 1
Max Discrimination

figure 7.3. Estimated risk of ethnic governmental conflict as a function of extent of

political discrimination (black line) with 90% confidence interval (gray area).

Since the effects in logit models are linear in the log-odds of outcomes, the
exact shape of such effect plots – in this case ethnopolitical discrimination and
horizontal economic inequality – will depend on the specific values of the other
covariates in the underlying regression model that influence the baseline risk of
conflict. In these plots we fixed all other variables in Model 7.6 at their means.
However, the absolute risk scores would differ notably if we had considered
a high-risk country profile, and it is clear that conflict-inducing factors often
tend to go together (e.g., Buhaug and Gleditsch 2008).
We now examine if any particular cases are especially influential in driving
these results, and whether the conflicts that “fit” the model in the sense of
falling on the regression line plausibly can be linked to economic and polit-
ical marginalization of ethnic groups, Put differently, can we replicate at the
country level the patterns discussed for groups in earlier chapters, where polit-
ically excluded and discriminated groups (Chapter 4) as well as economically
marginalized ethnic groups (Chapter 5) are significantly more likely to rebel
against the state? In this regard, examining some of the conflict cases with high
values on the inequality indicators may provide further insights. Table 7.4 lists
the five countries with the highest scores of political and economic horizontal
inequality, respectively.
For example, Russia appears to be one important case contributing to the
main finding in Models 7.5 and 7.6 that economic inequality among ethnic
160 Inequality, Grievances, and Civil War

05
04
Pred cted Probab ty of Conf ct
03
02
01
0

1 2 3 4 5 6
Max Low Ratio

figure 7.4. Estimated risk of ethnic territorial conflict as a function of extent of eco-
nomic inequality (black line) with 90% confidence interval (gray area).

groups increases ethnic territorial conflict risk. Russia is the country in our
data that has the highest level of income inequality (max low ratio = 6.8), and
it hosted three onsets of ethnic separatist conflict, in Chechnya in 1994 and
1999 and in a larger part of the Caucasus in 2007. From the discussion in
Chapter 5, it is clear that these conflicts are strongly rooted in local grievances
related to the dismal economic situation relative to the national center. Like-
wise, the uprising against Saddam Hussein’s regime in Iraq in 1991 was moti-
vated in large part by the narrow concentration of political power, excluding
the large Shia majority from participation and influence, and this political
configuration is well captured by the country-aggregated exclusion indicator.

Out-of-Sample Prediction
Another way to assess model performance is to evaluate the ability to correctly
predict new conflicts outside of the estimation sample. Sometimes a poor indi-
cator of predictive power or theoretical leverage, statistical significance often
favors more complex models that capture idiosyncrasies in the estimation sam-
ple but predict worse out of sample (Ward, Greenhill, and Bakke 2010). The
regression results shown in Tables 7.1 and 7.2 provide strong evidence that
inequalities that follow ethnic cleavages are more strongly associated with civil
Country-Level Inequalities and Civil War 161

table 7.4. High Horizontal Inequality in 1999 and Civil War Onset, 2000–9

Max Exclusion Max Low Ratio Pr(onset)

HI Country 1999 1999 2000–9
Max exclusion

Rwanda 0.85 1.00 0.47

Iraq 0.77 1.06 0.78
Niger 0.72 1.13 0.62
Gunea 0.67 1.01 0.28
Angola 0.59 1.15 0.62

Russia
Max low ratio

0.05 6.77 0.89

Peru 0.40 2.06 0.38
Central Afr. Rep. 0 1.78 0.31
Turkey 0.19 1.74 0.33
Israel 0.36 1.62 0.23

war onset in the post–Cold War period than inequalities unrelated to ethnicity.
However, although this is the case for the current sample, it does not necessar-
ily follow that horizontal inequalities can help improve our ability to predict
where and when new conflicts will break out in the future.
Because of limitations in data for the group-based income inequality indica-
tor, we have so far estimated empirical models for the post–Cold War period
only. In order to obtain a reasonable training sample and a reasonable period
for out-of-sample prediction, we relax this inclusion criterion. Based on the
full global sample for all years since 1960, our prediction analysis is conducted
in a number of steps. We start by estimating a horizontal inequality (HI) logit
model of civil war onset (with a model specification identical to Model 7.4) for
the period 1960–99. From this model we then calculate the accumulated pre-
dicted conflict risk for the subsequent decade, 2000–9, based on the observed
country characteristics in the final year of observation (1999). We then esti-
mate an alternative model on the same training sample 1960–99, substituting
the intergroup inequality measures with our proxies for vertical inequality
(VI) – democracy and the Gini coefficient of income dispersion. Following the
procedures for the HI assessment, we calculate the compound conflict risk
for the 10 year interval 2000–9, based on the model’s prediction scores for
1999.8 The final step compares our predictions with the record of observed
civil wars for the period 2000–9 from the UCDP/PRIO Armed Conflict
Data Set.

8 If the annual conflict probability in 1999 is p, we can compute the compound probability over
the ten-year period 2000–9 as 1−(1−p)10 , based on the complement of peace in all the successive
ten years.
162 Inequality, Grievances, and Civil War

1.00
0.75
True Positive
0.50
0.25
0.00

0.00 0.25 0.50 0.75 1.00

False Positive

VI model HI model

figure 7.5. ROC plot of out-of-sample prediction, 2000–9.

As an illustration, we use a receiver operating characteristic (ROC) plot to

compare the performance of the out-of-sample predictions in terms of the share
of missed conflicts and false positives across the range of possible prediction
thresholds. A better performing model would yield a higher ROC curve relative
to an alternative model, reflecting a higher ratio of correct to false conflict pre-
dictions. As shown in Figure 7.5, the two models seem to perform quite well;
when half of all countries (13 of 26) that experienced civil war onset during the
first decade of the twenty-first century are correctly identified, less than 20% of
the nonconflict observations are falsely predicted to host a new civil war. Over-
all, the model containing our group-based (HI) indices of economic inequality
and political discrimination performs better than a similarly specified model
based on individual-level inequality (VI) measures – especially when it comes to
predicting some of the less obvious cases of civil war outbreak. The area under
the curve (AUC) statistic provides a measure of overall predictive power that
is useful for comparing models drawn from the same sample. The AUC scores
for the HI and VI models are 0.822 and 0.783, respectively. The difference
between these estimates, while less than massive, is statistically significant with
a 5% margin of error. In other words, we have strong evidence to conclude
Country-Level Inequalities and Civil War 163

1 no onset
onset RUS
.8

IRQ

SAF
.6
HI Model
.4
.2
0

0 .2 .4 .6 .8 1
VI Model
figure 7.6. Scatter plot of individual out-of-sample predictions, 2000–9.

the HI model outperforms “standard” civil war models in terms of providing

accurate predictions out-of-sample.
In a further evaluation of the prediction analysis, we consider the predictions
for countries that had one or more civil war outbreak(s) between 2000 and
2009. Table 7.4 lists the five relevant countries with the highest exclusion
and inequality scores, respectively, in the final year of the training period
(1999). Quite tellingly, there is no overlap between these two lists; countries
with very high ethnopolitical exclusion tend to score relatively low on the
economic inequality indicator, and the reverse is equally true. The former group
comprises states with highly exclusionary policies, where minority groups were
politically dominant. Likewise, the civil war sample includes some countries
with substantial interethnic economic differences, several of which have been
discussed in earlier chapters.
Figure 7.6 provides a more complete overview of how the out-of-sample
predictions match observed conflict outcomes for the VI and HI model. The
top “true” predictions from Table 7.4 are easily detectable, but the plot also
identifies one false positive, South Africa, which is considered at greater risk
in the HI model than in the VI model. The relatively high prediction score is
driven by South Africa’s severe horizontal economic inequalities – only Russia
scores higher on the max low ratio indicator.
164 Inequality, Grievances, and Civil War

table 7.5. High Horizontal Inequality in 2009 without Conflict, 1991–2009

HI Country Max Exclusion 2009 Max Low Ratio 2009

Max exclusion

Syria 0.84 1.29

Bahrain 0.70 1.00
Bhutan 0.67 1.02
Jordan 0.59 1.00
Ecuador 0.42 1.05

South Africa
Max low ratio

<0.01 4.29
Saud Arabia 0.19 2.06
China 0.11 1.97
Vietnam 0.02 1.75
Namibia 0.03 1.71

A brief inspection of country characteristics for the final year with data
may provide some insights for where new civil wars may be likely (see Table
7.5). When we focus on countries without a recent history of armed conflict
(no conflict since 1990 until the end of 2009), we find the highest level of
ethnopolitical exclusion in Syria, followed by Bahrain. The recent developments
associated with the so-called Arab Spring in the Middle East 2011 would seem
to add considerable credibility to this list. Yet, we should be cautious about
making too strong claims here as the turmoil in Bahrain and the civil war in
Syria clearly were influenced by preceding events, violent as well as nonviolent,
in Tunisia, Egypt, and Libya, all of which have low values on the exclusion
indicator as our present data are unable to capture nonethnic and nonspatial
social inequalities.
Focusing on economic disparities, our model suggests that unless severe
intergroup income inequalities are reduced, South Africa could see renewed
violence and civil war, with the caveats noted previously.

Sensitivity Analysis
As in the other empirical chapters, we have conducted a number of additional
sensitivity tests, with respect to sample inclusion criteria, variable operational-
ization, estimation technique, and potentially influential observations. These
are documented in greater detail in the online supplementary material, and in
this section we briefly synthesize the main tests.
In general, the results for ethnopolitical marginalization and horizontal eco-
nomic inequality become stronger if we relax the stringent sample definition
and extend our data to include the entire postcolonial period (1960–2009) or
all years since 1946, the first year of the UCDP/PRIO conflict data.
A harder test is to replace the dependent variable by a separate data source.
We reestimated all models with Fearon’s (2010) updated civil war data,
Country-Level Inequalities and Civil War 165

table 7.6. Summary of Findings

Hyp. Prediction Result

H7.1 political HI –> civil war Yes, but only for ethnic governmental conflict;
exclusion of large groups reduces risk of
ethnic territorial conflict
H7.2 economic HI –> civil war Yes, but only for ethnic territorial conflict;
countries with poor, peripheral groups more
likely to host self-determination conflicts
H7.3 political I –> civil war Not confirmed; no robust association found
H7.4 economic VI –> civil war Not confirmed; income dispersion weakly
negatively associated with ethnic territorial
conflict

classifying conflicts based on the original data set’s coding of ethnic versus
nonethnic and center versus exit wars. These data differ from the UCDP/PRIO
data set by identifying only civil wars that caused at least 1,000 deaths in
total, including at least 100 on each side, and with a yearly average of at
least 100 deaths. Despite the considerable reduction in the number of civil war
observations with these data, we are able to reproduce the distinct pattern of
inequality and conflict, whereby exclusion or discrimination of large groups is
associated with civil wars over governmental control while economic marginal-
ization increases the risk of separatist claims (see the Online Appendix).
We have also evaluated the sensitivity of the main results to changes in the
model specification, notably by modifying the sample inclusion criteria (time
and space), by selection of control variables and the functional form of the
grievance proxies, and by using rare events logit to correct for possible biases
stemming from the rareness of civil war onset in our data (King and Zeng
2001). None of these modifications change the results substantively from those
presented in this chapter.
Overall, these tests contribute to bolster our confidence in our overarching
conclusion, namely that the risk of seeing civil war in a country should be con-
sidered a function of collective grievances arising from political and economic
inequalities across groups. By contrast, interpersonal differences seem largely
unrelated to civil war (see Table 7.6).

Conclusions
In this chapter we have revisited the relationship between inequality, grievances,
and civil war at the country level. We have previously criticized many exist-
ing analyses that have tried to evaluate the relationship between inequality
grievances and civil war for using measures with poor theoretical ground-
ing, and argued that it is important to focus on group-specific inequality
166 Inequality, Grievances, and Civil War

characteristics to understand the implications for grievances and conflict.

However, although most previous analyses at the country level have dismissed
the role of inequality, this is in our view primarily a result of the specific
atheoretical measures used rather than due to the irrelevance of inequality or
grievances per se. Nor does it follow that horizontal inequalities have no impli-
cation for the risk of conflict at the country level. In this chapter we have shown
that it is possible to derive more informative country-level measures of hori-
zontal inequality that specifically draw on underlying group-level data. When
we compare these measures informed by our group-level theories to standard
proxies for vertical economic and political inequalities, as represented by the
Gini coefficient and democracy indices, we find strong empirical evidence in
support of our approach highlighting the weakest link mechanism in relation
to horizontal inequalities. Countries where the poorest ethnic groups are far
below the national average income level face a significantly higher risk of civil
war than more equal societies. Our evidence also suggests that ethnopolitical
exclusion and discrimination are systematically related to a higher risk of civil
war in the sense that countries with larger excluded groups with both plausible
motivation and opportunity for conflict are more likely to see civil war. These
findings show that there is no inherent level of analysis problem separating civil
war studies at the country and the group level; using appropriate measures, it
will often be possible to scale information up or down from one level to another.
These findings lend strong support to our idea that risk of conflict depends on
the worst-case scenario, or the weakest link, in the sense that risk is positively
associated with extent of inequality for the most economically marginalized
and/or largest excluded group. This analysis further demonstrates that it is
entirely possible to create country-level profiles that reflect such variation at
the group level.
The move to the country level also allows us to consider differences between
types of civil war and how the risk of conflict for countries involving groups
with plausible grievances compares to the risk of civil war in other countries
that do not have ethnic cleavages. Distinguishing between types of conflict by
actors and incompatibility in our view further strengthens the relevance of how
grievances can arise over political and economic marginalization along ethnic
lines. Our results clearly indicate that economic horizontal inequality above
all tends to promote ethnic separatist conflict, where poorer groups seek to
claim autonomy or separation from the state. This finding corresponds well
with the patterns of contemporary conflicts, where separatist insurgencies typ-
ically involve small, economically disadvantaged groups and tend to be located
in the peripheral hinterland of larger states. The physical distance from the
group’s homeland to the center of national politics – which often goes together
with a substantial cultural divide – and the small size of a peripheral group
will typically lead to political marginalization, even under more competitive
political institutions. This, in turn, can imply that demanding full or partial
Country-Level Inequalities and Civil War 167

self-determination may be perceived as the only viable option for groups dis-
satisfied with the status quo.
Distinguishing between specific incompatibilities also allows us to go fur-
ther in terms of understanding the relationship between group characteristics
and the specific demands or claims that groups are likely to pursue in violent
conflicts. Whereas small marginalized groups are likely to seek secession or
autonomy, large groups are more likely to demand control of the state itself.
Our results strongly suggest that the risk of ethnic civil wars over control of the
state increases with the size of the largest politically excluded group relative to
the ethnic group(s) in power. Large marginalized groups can more easily make
a plausible cause for demanding inclusion in national politics. Although such
claims by themselves may not always be successful or overturn a restricted or
minority rule, the sheer size of the excluded group means that it would consti-
tute a potent military force if mobilized and hence is a clear threat. Cases like
those in Iraq, Rwanda, and South Africa demonstrate the probabilistic rather
than deterministic nature of this relationship, and strong ethnocratic regimes
may be able to hold on to power for generations in the absence of a coordinated
opposition.
Finally, the third form of civil war that we have considered in this chapter,
namely nonethnic conflicts, comprises a more heterogeneous set of cases, and
it appears that none of the standard suggested predictors of civil war explain
much of the variation here (see also Buhaug 2006). The conventional indicators
of vertical inequalities in economic and political privileges, the Gini coefficients
of income dispersion and democracy, exhibit generally inconsistent and mostly
trivial effects on the risk of civil war across all three conflict types. There are a
number of possible interpretations that may be applicable here. Following on
our previous analysis, one might argue that social inequalities at most are only
weakly and inconsistently related to general civil conflict risk unless they follow
important ethnic cleavages. This may, of course, be due to the low quality of
the available data, and it is possible that more comprehensive and sophisticated
data will give stronger results for the role of general social inequality in civil
war (see, e.g., Boix 2008). However, this may also reflect another important
component of agency, namely how mobilization does not follow directly from
objective inequality or grievances per se, but requires entrepreneurs and orga-
nizations. This calls for further research into when specific organizations, such
as Marxist ones, have been able to successfully mobilize or not in the face
of large social inequalities. Although mobilization may be easier when groups
can be mobilized along ethnic lines, organizations still play a mediating role
between groups and resort to conflict, and we turn to this topic in the next
chapter.
9

Conclusions for Theory and Policy

We are now ready to take stock of the findings of this book. The overall message
is clear: group-level inequality and grievances matter for conflict. In contrast,
much of the contemporary literature on within-state conflict has tended to
brush aside ethnic grievances in favor of materialist interpretations that high-
light individual economic incentives, natural resources, and state weakness,
while overlooking the fundamental importance of group-level mechanisms.
Once properly conceptualized and operationalized, horizontal inequality can
be shown to have a strong impact on the outbreak and duration of civil war.
In this concluding chapter we first summarize our main findings and then
discuss their theoretical significance, before turning to a discussion of past and
future trends, as well as the policy relevance of our findings.

Summary of the Main Empirical Results

Throughout the book, we have highlighted access to state power as a pivotal
factor strongly influencing both the risk of conflict and its duration. Our empir-
ical investigations indicate that economic horizontal inequality contributes
powerfully to the risk of civil war outbreak as well. Table 9.1 summarizes
these findings by listing all hypotheses for which we have found solid empirical
support.
Our first set of results pertains to political horizontal inequality. Perhaps the
most important corroborated claim of the entire book is that excluded groups
are more at risk of conflict (H4.1) together with its dynamic version relating
to how downgrading increases the prospects of violent conflict (H4.2). We
also found that the relative size of excluded groups has a positive influence
on conflict risk (H4.3), as does the number of previous conflicts (H4.4). This
set of relationships is exactly what you would expect given our theoretical
framework and its postulated grievance mechanisms.
205
206 Inequality, Grievances, and Civil War

table 9.1. A Summary of the Confirmed Hypotheses of This Book

Confirmed group-level hypotheses about conflict onset

H4.1. Ethnic groups that suffer from limited access to state power are more likely to
experience conflict than those that enjoy full access.
H4.2. Groups that experienced recent loss of state power are especially likely to
engage in internal conflict.
H4.3. Large excluded ethnic groups are more likely to experience conflict than are
smaller excluded groups.
H4.4. Ethnic groups that have experiencd conflict in the past are more likely to
experience conflict onset than those that have not.
H5.1. Relatively poor ethnic groups are more likely to experience civil war than those
that are closer to the country average.
H5.3. Economic inequality makes less wealthy, excluded groups more likely to see
civil war.
H6.3. The probability of conflict follows an inverted U-shape for the relative size of
the TEK group in the secondary dyad.
H6.4. The conflict-dampening effect occurs for relatively large TEK groups that are
included, and not for those that are excluded.
Confirmed country-level hypotheses about conflict onset
H7.1. Countries where large groups suffer from limited power access are more likely
to experience conflict than those with lower levels of exclusion.
H7.2. Countries where the least wealthy groups fall far below the national average are
more likely to experience conflict than those that are more equal.
Confirmed organization-level hypotheses about conflict duration and outcomes
H8.1. Rebel organizations with greater strength relative to the government will tend to
fight shorter conflicts.
H8.2. Rebel organizations that control territory tend to fight longer conflicts.
H8.3. Rebel organizations with a link to an excluded ethnic group will see longer
conflicts with the state.
H8.4. Conflicts with the state where rebel organizations have a link to an excluded
ethnic group are less likely to end in a decisive outcome.
H8.5. Rebel organizations that receive support from TEK groups are likely to fight
longer conflicts with the state.
H8.6. Rebel organizations that receive support from TEK groups controlling another
state are likely to fight longer conflicts with the state.

However, it is not only political asymmetries among ethnic groups that

matter, but also economic horizontal inequalities. Indeed, the poorer groups
are compared to the country average, the higher the likelihood of civil war
outbreak (H5.1). We have also established that this effect appears to worsen
the prospects of sustained peace in cases that are already characterized by
political inequality (H5.3). In this important sense, the conflict-inducing impact
of economic horizontal inequality appears to hinge on there being some level
of political grievances to start with. Thus, the combined effect of political and
economic inequalities is not so much additive as conditional in the sense that the
Conclusions for Theory and Policy 207

impact of the latter depends on the level of the former. In contrast, the evidence
for the claim that wealthier groups are overrepresented in the conflict statistics
is considerably weaker and cannot be statistically ascertained in general (see
H5.2). However, we note that the results suggest a strong positive effect for
wealthier and excluded groups, although the evidence here is somewhat less
consistent, possibly due to the low number of such cases in the observed data.
Relaxing the closed-polity assumption, or that conflict risk can be satisfac-
torily assessed by considering domestic factors and conditions alone, we also
studied the influence of transborder ethnic kin (TEK) and found that the risk of
violent conflict is the highest in situations where the TEK group is of roughly
equal size compared to the incumbent group(s) in the country in question
(H6.3). However, there are important differences as regards the TEK group’s
own power status. Contrary to common fears about the dangers of irreden-
tism, our results show that if transnational communities happen to control
their respective states, this influence has a conflict-dampening impact for large
groups (H6.4). Where the TEK group is stateless, however, one cannot count
on such a reduction of conflict risk (H6.5).
Once we shift the focus from factors characterizing groups to those that
influence the conflict risks of entire countries, we find very similar effects.
We summarize the main results as the corroboration of two main hypotheses,
each one corresponding to a dimension of horizontal inequality. Extending
the claim of H4.1 to the country level, we found that countries with large
excluded and discriminated groups are especially prone to governmental ethnic
conflict (H7.1). Similarly, countries where some ethnic groups are severely
disadvantaged in economic terms are significantly overrepresented among those
experiencing territorial conflict (H7.2). In contrast, there is little evidence of
greater vertical inequality increasing the risk of conflict, whether along political
or economic lines.
Finally, we were able to derive a series of important results explaining why
some rebel organizations fight civil wars that last longer than others, and
why conflicts end in a particular way. Consistent with previous studies, it
appears that rebel organizations that are relatively weak and are in control of
their own territory are especially prone to endure longer conflict (H8.1 and
H8.2, respectively). Returning to the main theme of the book, we detect a
powerful influence of power status on conflict duration: excluded groups fight
longer wars than those that enjoy privileged power access (H8.4). Such fights
are also less likely to be decisive (H8.3). Finally, our analysis shows that rebel
organizations that receive support from TEK groups will be involved in longer
wars (H8.5), especially if the TEK groups control another state (H8.6).

Theoretical Consequences and Extensions

In view of the strong and coherent effect of both political and economic hor-
izontal inequalities on internal conflict processes, which persists even when
controlling for alternative explanations, we conclude that it is premature to
208 Inequality, Grievances, and Civil War

write off grievances as causes of civil wars. Far from denying the importance of
materialist interpretations, we argue that these often interact with grievance-
related mechanisms. For example, our perspective is broadly compatible with
Fearon and Laitin’s (2003) emphasis on state weakness as a key factor increas-
ing the risk of conflict. However, opposed to their account, we insist on mea-
suring this weakness in relation to the strength of ethnonationalist challengers
rather than treating it as merely a “technological” matter. Along the same
lines, Hartzell, Hoddie, and Rothchild (2001) argue that state weakness is typ-
ically associated with state-led exclusion of entire ethnic groups from access
to power. Far from being ethnically neutral, then, the lack of state capacity
is intimately intertwined with issues relating to ethnicity and nationalism, and
thus also to the emergence of ethnonationalist grievances.
We suspect that similar interactions between grievances and opportunities
can be found with respect to the role of relative location, terrain, and natural
resources as well. Indeed, it is not only the effectiveness of policing and military
power that declines as one distances oneself from the center of government, but
also the state’s ethno-cultural penetration (Cederman 2008). Prominent studies
in historical sociology demonstrate that reactive identity formation, opposing
what is seen as the center’s illegitimate incursion and exploitation, operates
predominantly in geographically peripheral areas that have not been success-
fully assimilated by the power center (Rokkan in Flora 1999; Hechter 2001).
Likewise, natural resources are not merely fought over by greedy individuals,
but can also serve as a powerful source of grievances to the extent the state
elites exploit them without sharing the riches with ethnically distinct local pop-
ulations claiming them as their own property (e.g., Aspinall 2007). All these
examples underline the futility of simplified bunching together and labeling of
“proxies” as pertaining to this or the other side of abstract debates involving
greed, opportunities, or grievances.
At this point, it is time to go back to the four weaknesses of the contem-
porary civil war literature that we listed in Chapter 1. There we identified
deficits in mainstream, quantitative research as regards intermediate aggrega-
tion levels, motivational explanations, the underdeveloped role of the state,
and theoretically relevant data resources.

Intermediate Disaggregation Rather Than Just Individual- or

Country-Level Analysis
We have noted that much of the recent scholarship on internal conflict opts
for micro-level research designs that highlight the role of individuals in conflict
processes. Of course, this trend should be welcomed since it often strengthens
causal inference by relying on a battery of methods including extensive archival
research, interviews, surveys, and natural experiments (Kalyvas 2008b). How-
ever, something has been lost in the quest for internal validity. By leaping from
country-level analysis to individual-level investigations, an entire class of phe-
nomena involving group-level mechanisms has been left understudied. Despite
Conclusions for Theory and Policy 209

the importance of private grudges and individual self-interest, we have argued

that a large and important class of civil wars unfolds primarily in constel-
lations featuring ethnic groups and governments. Social psychological theory
offers compelling reasons why individuals identify with groups and act on their
behalf (e.g., Turner 1987). Moreover, there is a vast empirical literature that
proposes a number of conflict-inducing mechanisms at the group level. In par-
ticular, we have shown that Gurr and his colleagues have pioneered this type of
investigation (Gurr 1993a; 2000b). Indeed, on the basis of our own data sets,
we find patterns that strongly confirm the theoretical group-level postulates of
this book.
This does not mean that we believe that groups constitute coherent
“essences” or are the only relevant actors in conflict processes. Quite on the
contrary, we have chosen to treat ethnic groups as our main units of analysis for
pragmatic reasons, and we think that our empirical results confirm the useful-
ness of this choice. Nevertheless, ethnic identities vary from those that involve
strong levels of identification and solidarity that justify talk of “groupness,” to
situations that are so fluid that ethnic groups cannot be said to exist at all. Yet,
we part company from Brubaker (2004) as regards his wholesale rejection of
identities and groups as approximations of a large set of phenomena. It would
seem that tractability reasons speak for separating analytically blurred clusters
of ethnic identities into recognizable units as long as one is conscious of the
simplification underlying such analysis. However, as soon as the attention shifts
from conflict onset to properties of conflicts, such as duration and outcome, it
becomes possible to choose potentially more appropriate units of analysis. In
Chapter 8, for example, we study how such patterns depend on the properties
of rebel organizations and their relations to ethnic groups.
The limitations of exogenous and unitary ethnic group definitions are obvi-
ous, even if they are allowed to vary over time as is the case with the EPR-ETH
data set. First, some important aspects of conflict processes hinge on internal
tensions and strategic interactions among communities, subgroups, and organi-
zations that are linked to the same ethnic category (e.g., Wucherpfennig 2011).
Second, conflict processes can be seen as emanating from the very change of
ethnic and nationalist boundaries (Cederman 1997). Statistical research based
on panel data is inherently limited when it comes to handling such processes,
especially in the case of mergers and splits of units (Abbott 2001). These limita-
tions leave plenty of room for alternative methodologies, including historical,
longitudinal studies and computational models that allow for a much more
flexible way of handling identities and boundaries (Cederman 2002).

Motivational Rather than Merely Cognitive Mechanisms

The field of ethnicity and nationalism studies appears to have undergone a
“cognitive turn” that privileges “cold” and “bloodless” accounts of conflict at
the expense of those that accord emotions a central behavioral role (Brubaker
2009). Without denying the relevance of cognitive explanations, our focus on
210 Inequality, Grievances, and Civil War

grievances deviates from this trend at least to the extent that it brings atten-
tion to emotional reactions to inequality, including anger and resentment (see
Chapter 3). For example, attempts have been made to emulate conflict situa-
tions in laboratories by cuing subjects with faces and symbols, but it remains
doubtful whether the findings from such cognitive exercises can ever be prop-
erly generalized to real-world settings where life or death is at stake (e.g.,
Habyarimana et al. 2009). In fact, there is strong evidence, even within exper-
imental settings, pointing to the essential role of emotional mechanisms in
collective action processes. Indeed, prominent research in experimental eco-
nomics has provided compelling reasons to believe that emotional reactions to
unfair resource distributions help groups overcome collective-action dilemmas
(Fehr and Gächter 2000).
Apart from brief references to selective cases in the empirical chapters, our
book does not offer any direct evidence that emotions are involved, which is also
difficult to demonstrate. Instead, we have relied mostly on indirect measures.
However, we believe that grievance-based explanations gain credibility in light
of these findings. In particular, the fact that loss of power is strongly associated
with conflict outbreaks for groups could be seen as a clear sign that emotions,
such as resentment and even outrage, may operate (see H4.2). In any case,
it is hard to see how a purely cognitive approach could close the causal gap
between horizontal inequality and conflict. We very much hope that future
research will continue to investigate conflict mechanisms involving emotions
explicitly. Major progress has been made by systematic investigations using
case studies and other qualitative methods (e.g., Petersen 2002; 2011).

Ethnonationalism Rather than Merely Ethnicity

Running like a unifying theme throughout the chapters of this book, another
critical line of reasoning targets the tendency of quantitative conflict researchers
to reduce ethnicity to demographic aspects of “societies.” Measures such as
fractionalization and polarization pay no attention to the actual power rela-
tions that motivate challengers and incumbents in civil wars. After all, civil
wars by definition do not merely confront individuals or groups with each
other, but primarily feature the state and a specific incumbent government
against one or more organized opposition actors. Thus, the state should be
seen as both as a protagonist and a prize worth fighting over in internal conflict
processes.
The reason for this oversight has a lot to do with the shift away from
nationalism to ethnicity in recent conflict research (Cederman 2013). Limiting
the focus to measurable aspects of ethnic politics within the “container” of the
state has caused many researchers to lose sight of the institutional aspects of
ethnicity, especially those associated with state organs and policies. Bunching
ethnic and nationalist phenomena together into highly aggregated, primarily
ethno-demographic indices that tap into societal and demographic properties
such as diversity and polarity, many of these scholars have been keen on proving
Conclusions for Theory and Policy 211

that there is nothing special about ethnicity and that it has no systematic or
statistically discernible impact on collective violence.
Following in the footsteps of seminal studies by Horowitz (1985), Brass
(1991), Hechter (2001), and Mann (2005), our analysis vindicates the value of
adopting an institutional perspective on the link between ethnicity and conflict.
Our analysis has also been directly inspired by the quantitative studies produced
by Gurr and other scholars working with Minorities at Risk (MAR) data. As
demonstrated in Chapter 2, this important stream of research in many ways
anticipates our own theoretical framework by highlighting the relationship
between ethnic groups and government. These studies have generally found that
exclusive and discriminatory state policies tend to provoke grievances, thereby
increasing the risk of political violence. Much of the grievance skepticism in
the literature, including that of Fearon and Laitin (2003) and Collier and
Hoeffler (2004), has targeted this research tradition. More recently, however,
a number of scholars, who have followed up and improved on the earlier work
on inequality and grievances, have found strong evidence of such effects (see,
e.g., Regan and Norton 2005; Østby 2008b; Stewart 2008b; Østby et al. 2009;
Goldstone et al. 2010). By further improving the data situation and offering
a more comprehensive perspective than that of separate articles, the fact that
this book confirms and further develops this recent wave of scholarship casts
additional doubt on the attempts to downplay grievance-based explanations of
civil war.
In fact, Fearon and Laitin now seem ready to at least partially abandon their
previous, sweeping rejection of ethnic grievances as explanations of civil war.
Indeed, their recent analysis of sons-of-the-soil conflicts features explicit frus-
trations on the part of cornered peripheral populations, who feel provoked by
migratory flows typically consisting of members of the dominant ethnic group
in the state (Fearon and Laitin 2011). Nevertheless, they treat this empirical
pattern as a specific phenomenon rather than as a special case of grievance
mechanisms. Moreover, they remain reluctant to open their theorizing to gen-
eral patterns involving inequalities and grievances as causes of civil war onset,
partly due to data problems associated with selection bias in the MAR data,
and partly because of problems relating to endogeneity, as discussed at the end
of Chapter 4. Our own duration analysis, presented in Chapter 8, suggests
that sons-of-the-soil conflicts can be safely subsumed under the more general
heading of political exclusion. Furthermore, our most recent research based on
an instrumental-variable approach going back to the power access within the
colonial empires indicates that there is indeed reverse causation, but this leads
us to underestimate, rather than overestimate, the conflict-inducing impact of
exclusion (see Wucherpfennig et al. 2012).
As illustrated by these observations, there is plenty of room for elaboration:
future research will have to propose more innovative research designs that help
us account for endogeneity while probing more deeply into the role of the
state’s suborganizations and political institutions as well as its activities and
212 Inequality, Grievances, and Civil War

policies in areas such as policing and military affairs, law enforcement, tax
collection, education, linguistic and religious rights, and cultural symbolism.

Theoretically Relevant Data and Measures, Rather

than the Standard Toolbox
This book would not have been possible without a major investment in data
collection. Rather than relying on existing data sources and measures, we have
been involved in multiyear efforts to build and combine data sets on ethnic
groups, their wealth, access to power and links to groups in neighboring coun-
tries, as well as their relations to rebel organizations. By addressing group-level
properties, we believe that our measures are more directly linked to specific,
conflict-related theoretical claims than most standard indices in the quantita-
tive conflict researcher’s “standard toolbox.” Furthermore, they improve on
existing data sources through their comprehensiveness in terms of sampling.
Rather than restricting the sample to “minorities at risk,” the EPR-ETH data
project casts a wider net that also catches all privileged groups. Furthermore,
our data offer systematic information about economic inequalities at the group
level that go beyond what previous studies have been able to offer. Ultimately,
these improvements allow us to draw firmer conclusions about the influence of
inequalities, and indirectly grievances, on conflict.
In order to gather and combine the necessary information about hundreds
of groups, organizations, and states around the world, we have relied on online
surveys, geographic information systems, and relational databases. These com-
putationally demanding procedures are made available to other researchers
through a new integrated data portal called Geographic Research On War,
Unified Platform, or GROWup for short.1 This system integrates all relevant
data used in this book and will also provide updates of our analyses as new
data become available in the future.
Obviously, the empirical analysis of this book by no means exhausts what
we want to know about ethnonationalist conflict. Indeed, the wish list of addi-
tional data sources is long. Future versions of the EPR-ETH Data set will offer
information about different dimensions of ethnicity, including linguistic and
religious categories. It would also be desirable to combine the spatial wealth
estimates of Chapter 5 with information drawn from surveys and satellite data
on nightlights emission. As mentioned above, more detailed information on
institutional aspects of state organizations and their policies would also be
very useful, as would more detailed information on the links between eth-
nic groups on the one hand and political parties and social movements on the
other hand. Here we have only focused on civil wars that confront governments
with nonstate challengers. Fortunately, systematic data are already available
on other types of political violence, such as one-sided violence (Eck and Hult-
man 2007), communal (or nonstate) conflicts (Fjelde and von Uexkull 2012),

1 See http://growup.ethz.ch.
Conclusions for Theory and Policy 213

interstate wars (Gleditsch 2004; Braithwaite 2010; Sarkees and Wayman

2010), and terrorism (Enders, Sandler, and Gaibulloev 2011). It is harder to
find data that allow us to trace nonviolent conflict processes before they gen-
erate violence or entirely nonviolent processes (Chenoweth and Stephan 2011;
Svensson and Lindgren 2011).

Inequality and Grievances in Time and Space

So far, we have discussed our results, as well as their theoretical significance
and limitations, while assuming that they are universally valid both in time
and space. However, such assumptions are most certainly too strong, as they
conceal important contextual variation in terms of historical trends and geo-
graphic areas. Although it is beyond the scope of this book to model such
spatiotemporal processes explicitly, we will here consider how different levels
of inequality may affect the likelihood of conflict in terms of both the time
period and the specific world region in question. This will help us identify the
regions and countries most at risk when it comes to future outbreak of civil
war.
Even though our analysis in Chapter 7 fails to detect any straightforward
connection between political vertical inequality and internal conflict, we start
by considering the evolution of governments’ openness since WWII.2 It is
well known that the world has undergone a powerful democratization pro-
cess at least since the 1970s (Gleditsch 2002; Gleditsch and Ward 2006). These
regional “waves of democratization” started with the fall of the dictatorships
of Southern Europe, followed by a series of democratic transitions in Latin
American polities in the 1980s, culminating with the victory of democracy
after the fall of the communist dictatorships in the early 1990s (Huntington
1991). In addition, Asia and even Sub-Saharan Africa have exhibited steady,
if less dramatic, shifts toward democratic rule since the 1980s. Until 2009,
which marks the end of our data set, the glaring exception from this trend was
constituted by the Middle East and North Africa, a region that has recorded
hardly any progress in this respect since the end of WWII.
As an illustration of these trends, we use data from the Polity Data set (Gurr
1993a) to plot the average level of democracy for six world regions on a scale
from −10 to 10 (see Figure 9.1). The picture is striking: the three waves of
democracy referred to above are clearly visible, especially the most dramatic
one pertaining to “Eastern Europe” (which also includes the former Soviet
Union). The Western world region, which includes Western Europe, the United
States, Canada, Japan, Australia, and New Zeeland, exhibits very high levels

2 Theories developed to account for the democratic peace imply that mutual democracy has a
pacifying on interstate relations (see Chan 1997). However, the relationship between democracy
and international conflict has long been acknowledged to be more complex (see, e.g., Gleditsch,
Hegre, and Strand 2009).
214

figure 9.1. Average democracy levels by world region.

Conclusions for Theory and Policy 215

of democracy already after WWII. With the exception of the Middle East and
North Africa, there is strong convergence of higher levels of democracy toward
the end of the sample period. Obviously, the revolutions of 2011, commonly
referred to as the “Arab Spring,” may well change the status of this region as
a laggard in terms of democratization, but at the moment of writing it is still
too early to tell how fast, or whether, the global trend toward full democracy
will include the countries of that region.
Whereas the general evolution toward democratization is by now well estab-
lished as a major historical pattern, a lack of data on ethnic groups’ access to
power has prevented scholars from detecting a similar development in terms
of ethno-political inclusion. Yet, this knowledge gap is of critical importance,
especially in the light of this book’s argument that horizontal, rather than ver-
tical, inequality causes civil wars. Fortunately, the EPR-ETH Data set allows
us to shed light on this important question. In parallel to the previous graph,
Figure 9.2 traces the average level of ethnonationalist exclusion for the same
world regions. Here exclusion is measured as the proportion of the population
excluded from executive influence because the power access of their respective
ethnic group is blocked.
Just as the degree of democracy has increased over the last couple of decades,
a similar improving trend can be detected in terms of the power access of ethnic
groups. The graph again reveals major differences across the globe. Whereas
the Western states excluded a very small fraction of their population already in
1946 and have remained at that low level, other world regions have recorded
a considerable decline in terms of exclusion since the late 1999s. As with
democracy, however, there is still a major gap between the West and the least
exclusive non-Western areas. Possibly as a reflection of normative pressure
with global reach, the latter appear to be converging on an exclusion level of
about 10–15% but could well continue to decline. Again, the Arab world and
North Africa stand out as having recorded only modest improvements, such as
the changes following after the US invasion and fall of Saddam Hussein in Iraq
in 2003, but the level of exclusion remains about twice as high as most other
parts of the world. The difference compared to the Western states is even more
dramatic.
The findings of Chapter 4 indicate that discriminated groups are even more
prone to experience violence than excluded ones, a finding that is reflected at
the country level (see Chapter 7). We therefore turn to the development of
ethnic discrimination by world region. The population that is discriminated
represents a strict subset of our exclusion measure. Figure 9.3 plots regional
averages of ethnically discriminated segments of each country’s population. In
this case, the Middle East and North Africa stand out even more clearly. In
2009, the level of discrimination for this part of the world amounted to around
15%, making it an order of magnitude more discriminatory than other world
regions. The opposite pattern can be detected for Western countries, where the
level of discrimination approaches zero over the years. In between, there is a
216

figure 9.2. Ethnically excluded population by world region.

 figure 9.3. Ethnically discriminated population by world region.
217
218 Inequality, Grievances, and Civil War

converging trend to low levels below 5%, including a major decline in the case
of Sub-Saharan Africa.3
Since the origins of exclusion and a full explanation of these patterns falls
outside the scope of our book, we can only speculate about the reasons for
the overall decline of exclusion and discrimination during the last couple of
decades. According to Gurr (2000a), there are some signs that an international
regime of conflict prevention and resolution may be emerging. Indeed, Gurr
claims that since the mid-1990s, Western governments and international orga-
nizations have done much more to curb violence in troubled areas, and as a
consequence, the incidence of new ethnic conflict has dropped significantly.
What does the nascent international regime consist of? Gurr (2000a) men-
tions four main pacifying developments that have characterized the post-Cold
War period after its first turbulent half decade. First, increased attention has
been paid to group rights rather than merely to individual human rights. In
this respect the Organization of Security and Cooperation in Europe (OSCE)
and the Council of Europe have played key roles. Second, ethnic pluralism
and power sharing have become the expected modes of governance rather than
confrontation and domination. Third, instead of resorting to an all-or-nothing
strategy, parties in secessionist disputes now as a rule content themselves with
compromise solutions, such as cultural autonomy. Finally, a key ingredient in
the emerging regime depends on multilateral pressures orchestrated by interna-
tional organizations, and a willingness to deploy more forceful policy instru-
ments rather than letting regions descend into violence. Although this trend is
not perfect, the international intervention in Libya in 2011 is a case in point.
Ideally, coercive prevention should suffice to deter political leaders who may
be tempted to escalate conflicts for ideological or opportunistic reasons.
Apart from the inherently normative attractiveness of this increasing pattern
of ethnic inclusiveness, this book shows that the trend has potentially major
pacifying effects as regards internal conflict. Despite these encouraging signs of
progress, however, the contemporary world still exhibits considerable diversity
as regards levels of ethnic exclusion and discrimination. In order to assess the
prospects of conflict in these world regions, Figures 9.4 and 9.5 introduce maps
of the share of the excluded and discriminated populations respectively for the
year of 2009. The former figure tells us that the most important clusters of
exclusive rule can be found in Sub-Saharan Africa, the Middle East, and North
Africa, as well as in large parts of Asia. The exclusion levels are especially high
in the Sudan, Rwanda, the Democratic Republic of Congo, Congo Brazzaville,
and Angola. In the Middle East, Israel, Jordan, and Syria also exhibit extremely
high levels. In the ranked societies of the Western Hemisphere, the difficulties
experienced by native populations trying to participate in national politics are

3 Our findings on discrimination confirm Asal and Pate’s analysis in Marshall and Gurr (2005).
Using data from Minorities at Risk, they report a strong declining trend in ethnic political
discrimination around the world.
219

figure 9.4. Share of excluded population by country in 2009.

220

figure 9.5. Share of discriminated population by country in 2009.

Conclusions for Theory and Policy 221

table 9.2. Top-Ten Countries with Largest Excluded and Discriminated Shares
of the Population in 2009

Share of Excluded Share of Discriminated

Country Population Country Population
Syria 87.6% Bahrain 70%
Rwanda 84.8% Jordan 59.2%
DR of Congo 80.2% Latvia 37.6%
Sudan 75.4% Israel 31.3%
Bhutan 75% Estonia 29.9%
Congo 74.2% Bhutan 25%
Angola 70.5% Saudi Arabia 21.5%
Bahrain 70% Turkey 19.9%
Jordan 59.2% Iran 16.8%
Guatemala 52% Bangladesh 11.1%

on ample display. Unsurprisingly, the Western countries are almost entirely

free of exclusive policies.
When we turn to discriminated populations, the Middle East stands out.
Israel, Jordan, Saudi Arabia, Bahrain, Turkey, and Iran belong to this category
of states. In addition, two Baltic democracies, namely Estonia and Latvia, are
also listed as exercising discrimination since they make political participation
difficult for their Russian minorities by severely restricting naturalization along
ethnic lines.
It is instructive to list the countries with the highest level of exclusion and dis-
crimination. Introducing the first of these “exclusive” clubs, Table 9.2 reveals
that Syria’s ethnic minority rule reaches a staggering 87.6% of the population,
closely followed by Rwanda with 84.8% and the Democratic Republic of the
Congo at 80.2%. With the exception of Bhutan (75%) and Guatemala (52%),
all of the top-ten cases can be found in Africa or the Middle East.
Interestingly, the African cases disappear from the list when we turn to
outright discrimination (see the rightmost columns of Table 9.2). The Middle
East, together with two cases in Eastern Europe and Asia, correspond to this
category. At the very top, we find Bahrain and Jordan with 70% and 59% of
the population, followed by Latvia, Israel, and Estonia (37.6%, 31.3%, and
29.9%, respectively).
Apparently democracy does not immunize countries against discrimination
of large parts of the population. Whereas Guatemala is the only democratic
country among the top-ten of the exclusion list, we find as many as four democ-
racies among the most discriminatory regimes in 2009, namely Latvia, Israel,
Estonia, and Turkey. It should be noted, however, that discrimination takes
very different forms in these cases. As mentioned above, the two Baltic states
were classified as discriminatory because of their restrictive citizenship rules,
which in practice deprive the Russian minorities of even basic political rights
222 Inequality, Grievances, and Civil War

(Kelley 2004). In the case of Israel, long-standing and increasingly harsh dis-
crimination in the occupied territories effectively blocks the Palestinians from
taking part in the demos (Pappe 2004; Lustick 2008). Based on assimilatory
principles in contrast to the Israeli approach, ethnic domination in the Turkish
case does not prevent Kurdish individuals from participating even at the elite
level as long as they are willing to assimilate to the dominant culture. However,
up to 2009, the Kurds were exposed to discriminatory policies that prevented
any Kurdish group-level influence over the executive.
What do these observations on the most egregious cases of exclusion and
discrimination tell us about the risk of civil violence? In fact, three of the top-
ten exclusion cases involve ethnonationalist conflict, namely Rwanda, which
even experienced an outbreak that year, as well as the Sudan and Angola.
Likewise, we find three conflict cases among the states that discriminate the
largest populations. Within two years after these observations, the turbulence
in the Middle East and North Africa triggered major conflict in Syria and
Bahrain, both characterized by minority rule. Furthermore, tensions rose in
Jordan for the same reasons, and even Saudi Arabia experienced trouble with
its Shia minority in the east of the country. At the same time, the long-lasting
conflicts in Israel and Iran showed no signs of being solved. In contrast, the
Baltic states have remained peaceful, possibly in part thanks to the deterrent
effect of their large transborder ethnic kin, as explained in Chapter 6. Going
back to the trends displayed in Figures 9.2 and 9.3, the high levels of exclusion
and discrimination in the Middle East and North Africa, if enduring, can be
expected to generate conflict for years to come. Despite encouraging recent
reductions of exclusion and discrimination in other parts of the world, such as
Eastern Europe and the former Soviet Union, Asia, Sub-Saharan Africa, and
Latin America, there is still considerable potential for ethnonationalist warfare
in these world regions. In contrast, the Western World has been almost totally
liberated from ethnonationalist conflict. After the settling of the armed conflict
in Northern Ireland and promising steps toward a more comprehensive political
settlement for the Basque region in Spain, few major risks of civil war remain
among those countries.4

Policy Consequences
We ended Chapter 1 by offering some preliminary remarks on the policy rel-
evance of our grievance-based perspective. Having now presented empirical

4 Of course, there are still nasty cases of discrimination of the Roma populations, but their scattered
settlement patterns make organized mobilization and civil conflict unlikely. It is conceivable that
the economic crisis that started in 2008 could reverse some of this progress. There have been
occasional ethnonationalist populist stirrings, such as the Hungarian government’s irredentist
gestures toward Hungarian populations in Slovakia, but domestic support for such moves has
remained limited so far and the dense web of international institutions and norms within the
Western world have generally helped keep such tendencies in check.
Conclusions for Theory and Policy 223

evidence in its favor, we return to the issue of what can be done to prevent and
end civil wars. It is clear that those scholars who dismiss grievances will focus
on different policy priorities than we do. In a pioneering article on theory and
policy, Andrew Mack (2002, p. 522) explains how theoretical inferences have
consequences for political decision making:
If grievances have nothing to do with the onset of war, then seeking to assuage them
via preventive diplomacy, conflict resolution and confidence-building strategies will do
nothing to reduce the risk of armed conflicts. If Collier & Hoeffler and Fearon & Laitin
are correct, and what counts is not grievance but the relative capabilities of rebels versus
the state, then strategies of “peace through strength,” repression and deterrence would
appear to be optimal prevention strategies.

Whether the explanatory approach diagnoses ruthless rebels or weak states

as the main cause of conflict, most of the grievance critics can be expected
to recommend a strengthening of the state’s enforcement power. Ballentine
and Nitzschke (2005, p. 4) summarize the prescriptive consequences of the
“loot-seeking” paradigm:
The idea that civil war is driven by rebel greed was particularly appealing to some
policy-makers, discouraged by the complexity and seeming intractability of “ethnic”
and religious conflicts of the early 1990s. If many contemporary conflicts are driven
by contests over economic resources, the “resource wars” should be more amenable to
resolution than conflicts over such indivisible identity issues as ethnicity, religion, or
ideology. The greed-thesis shaped politics as well as policy, as corrupt and repressive
leaders in conflict countries found it a useful argument to deflect attention from their
own wrong-doings by putting the blame for their countries’ misery on “greedy rebels.”

According to this view, what is needed is not compromise and diplomacy, but
effective policing (van Creveld 1999). Summing up the experiences of Western
interventions after the end of the Cold War, Mueller (2004, p. 139) argues
that such efforts are relatively straightforward and effective in most cases:
“The intimidating, opportunistic thugs have been successful mainly because
they are the biggest bullies on the block. However, like most bullies (and
sadists and torturers), they tend not to be particularly interested in engaging a
formidable opponent.” The disorganized and undisciplined thugs are no match
for the professional armed forces of the developed countries. In sum, Mueller’s
conclusion for policy is strikingly opportunistic:
The record suggests, then, that it is possible to use policing wars to order the new
world – or at least to eliminate many of the criminal regimes and to pacify many of the
criminalized civil wars that are the main sources of unnatural death and deprivation in
the world (p. 140).

Rationalist scholars have also argued in favor of outside intervention as a way

to impose order on warring parties and pacify conflict zones. Anarchical condi-
tions imply that governments and their challengers face difficulties in credibly
committing to disarmament and political compromise (Fearon 1998; Walter
224 Inequality, Grievances, and Civil War

2009a). Provided they are forceful and sustained enough, third parties can pro-
vide the necessary credibility for agreements to stick. For example, Fearon and
Laitin (2004) analyze the conditions under which “neo-trusteeships” will be
able to stabilize postconflict situations.
From a more radical vantage point, some scholars assert that negotiated
agreements often prevent strong states from emerging and even argue that the
best option is to “give war a chance” (Luttwak 1999). In ethnonationalist con-
flicts, this could in practice amount to endorsing ethnic cleansing followed by
ethnic dominance within each “purified” ethnic unit (Kaufmann 1996; 1998),
or at least a more skeptical attitude toward compromise solutions secured
through agreements (Toft 2010).
Without belittling the urgency of establishing order and strengthening law
enforcement as a way to handle post-conflict situations, we believe that too
much weight has been put on these functions at the expense of encouraging
and pressuring conflict parties to accept compromises and political concessions.
Nationality problems will not go away until the underlying issue has been
resolved. In the absence of an equitable solution, resentment, and even hatred,
may fester from generation to generation. In particular, in today’s world of
mass mobilization along ethnonationalist lines, it is very unlikely that “decisive
victories” will guarantee long-term stability. Indeed, our results in Chapter 8
indicate that excluded groups are more likely to fight longer and less decisive
civil wars than included ones (see H8.1 and H8.4, respectively). Furthermore,
there is a danger that the wrong incentive structure for potential rebels or
ruthless state rulers will result from excessive reliance on symptom treatment,
such as external reinforcement, or on noninterference in the hope that this will
create order through ethnic dominance. Instead, the best option is to address
the root causes of the problem before violence erupts. Our book indicates the
most obvious way to do so: if ethnic exclusion and inequality lead to conflict,
then ethnic inclusion and a fairer distribution of public goods will offer the
best prospects for conflict prevention.
Of course, power sharing is never an easy option and often fails, even without
a history of prior conflict. It has often been remarked that most power-sharing
experiments end in recurrent violence (Fearon and Laitin 2007; Toft 2010).5
However, the argument that a conflict resolution method often fails is not a
valid objection against it if conflict would have broken out sooner, or would
be even more likely in its absence. Put differently, since power sharing is not
randomly assigned, and in fact probably more likely in difficult cases where
tensions are high or conflict has already occurred, then we cannot conclude
that it causes more harm than good on the basis of simple static correlations.
Failing to understand this logic of selection would lead us to identify hospitals

5 Other scholars think that power sharing in its various guises worsens the prospects of peace by
deepening cleavages and fueling mutual hostility (Rothchild and Roeder 2005; see also Roessler
2011).
Conclusions for Theory and Policy 225

as a threat against the patients’ health, since more people are ill in hospitals
than elsewhere.
Once we move beyond such naive analyses, things look quite different.
Explicitly considering the conditions under which power sharing is invoked,
Wucherpfennig (2011) models it as a result of the choices of strategic actors
rather than as an exogenously occurring, random condition. Based on EPR
data, he demonstrates that groups that are “upgraded” through improved
access to governmental institutions are less likely to experience recurrence of
conflict than those that are not accorded such concessions.
Of course, power sharing within the central executive is not the only way that
ethnic inclusion can be implemented, as illustrated by Rothchild’s (1997) exten-
sive analysis of different approaches to the management of ethnic conflict in
Africa. In some fortunate cases, elites’ early choice of cooperative strategies put
the development of their countries on a pacific path that allows “polyarchical”
regimes to emerge. Such systems offer extensive opportunities to participate
in political activities, both in parties and in civil society, regardless of one’s
ethnic identity (see also Horowitz 1985). In other cases, ethnic inclusion may
rely extensively on political decentralization and ethnonationalist autonomy
short of secession, although ethnic federalism has also been associated with the
outbreak of ethnic conflict by accelerating centrifugal tendencies within decen-
tralized regimes (see, e.g., Bunce 1999; Deiwiks et al. 2012).6 However, as
expected by Gurr’s (2000a) analysis of declining ethnic warfare from the mid-
1990s, the different strategies of ethnic inclusion appear to be more effective
than non-cooperative policies based on ethnonationalist domination. Summing
up his empirical investigation, Rothchild (1997, p. 19) concludes that

wherever Africa’s ruling elites have encouraged inclusiveness or diffused power by means
of electoral fine-tuning or territorial decentralization or have applied positive principles
of proportionality and reciprocity in four key areas – political coalitions in central
government politics, elite recruitment, public resource allocation, and group rights and
protections – they have managed to reduce the intensity of state-ethnic conflicts.

In some difficult conflict cases, however, hostility and mutual suspicion may
be so acute that power-sharing deals are simply not feasible and the only
hope of stopping a conflict hinges on separation of the involved populations.
Arguably, this is why Kosovo was allowed to secede and why a “two-state
solution” to the Israeli-Palestinian conflict remains the predominant recom-
mendation among scholars and politicians. However, in general, there are good
reasons to be cautious in recommending such geopolitical “lobotomy,” includ-
ing the lack of a guarantee that the ensuing situation will remove all nationality

6 Because ethno-federal institutions may lead to long-term fragmentation and even conflict, it is
safer to introduce federalism in such a way that federal boundaries do not coincide with ethnic
ones (Deiwiks 2011). However, it may be difficult to satisfy secessionist activists with less than
partial sovereignty over ethnic matters, such as linguistic policies and religious institutions.
226 Inequality, Grievances, and Civil War

problems and prevent the parties from starting an interstate war (Sambanis and
Schulhofer-Wohl 2009). Most importantly, careless encouragement of seces-
sionism threatens to cause a problem of moral hazard that may undermine the
stability of territorial borders and could ultimately reverse the gains made in
terms of outlawing interstate warfare (Fearon 2004a). Instead, what is desper-
ately needed is an international legal framework that stipulates clear conditions
under which partitioning could be allowed to proceed.
Thus, wherever possible, challengers of the current political order should
be encouraged to participate in the regime’s politics by being granted gov-
ernmental concessions through different types of power-sharing arrangements.
In some cases this means sharing responsibility within the executive, and in
others it entails offering minority rights and safeguards, and possibly regional
autonomy in federal arrangements.
Here we cannot aspire to offer anything like a complete assessment of what
specific types of institutions and policies stand the best chance of pacifying and
possibly even transcending ethnonationalist conflict, despite the topic’s obvious
importance to policy making. Our goal has been much more modest and basic:
we have attempted to offer a better understanding of the causes that trigger
and prolong civil wars by removing arbitrary restrictions on what constitutes a
good explanation. As we have shown, much of the extremely influential quan-
titative literature covering such conflicts has tended to brush aside grievances
as matter of theoretical assumption and questionable empirical evidence. By
improving data and measurements, we have been able to detect strong effects
of both political and economic horizontal inequalities, thus strengthening the
case of grievance-based interpretations. It is our hope that this book will con-
tribute to a more lively and balanced debate about the causes and cures of civil
wars, thus leaving more room for accounts that explicitly consider collective
emotional reactions to maltreatment by oppressive and ethnically monopolistic
governments.
Index

Abkhazians, 68, 85 Shiite Arabs in Bahrain, 88

ACD2EPR Data Set, 69, 70, 149, 191, 192, Shiite Arabs in Iraq, 83, 87, 88, 90, 137,
195 160
Aceh, 83, 84 Shiite Arabs in Saudi Arabia, 222
Acehnese, 18, 83, 84, 116 Sunni Arabs in Bahrain, 88
Afar, 85 Sunni Arabs in Iraq, 87, 88, 90, 137
African National Congress (ANC), 186 Sunni Arabs in Syria, 88
Afrikaners, 87 Aspinall, Edward, 43, 83, 208
Alawites, 63, 88 Assad, Bashar, 63, 88
Albanian National Liberation Army (NLA) in autonomy, 48, 59, 61, 62, 68, 71, 80, 82, 84,
Macedonia, 138 85, 86, 100, 113, 137, 158, 166, 167,
Albanians 187, 218, 225, 226. See also power
in Macedonia, 85, 120, 138 status
in Serbia/Kosovo, 4, 80, 102, 104, 113, 116, Ayres, R. William, 126, 135, 139, 173
120, 137
Macedonia, 137 Bahrain, 88, 164, 221, 222
al-Hashimi, Tariq, 87 Bakke, Kristin, 28, 160
alien rule, 27, 59, 99 Bakongo, 84, 87
al-Maliki, Nouri, 87 Baldwin, Kate, 100, 117
al-Qaeda, 192 Barbalet, Jack M., 64
Americo-Liberians, 67, 83, 182 Barth, Fredrik, 13
Amhara, 82 Basques, 98, 115, 222
ancient hatred, 26, 64. See also emotions, Bates, Robert H., 35
hatred Beissinger, Mark, 35
Anderson, Benedict, 33, 38, 43, 147 Belarus, 140
anger. See emotions Bhutan, 88, 221
Angola, 84, 218, 222 Black September, 88
Apartheid, 63, 83, 87, 90, 127, 186 Blainey, Geoffrey, 127, 174
Arab Spring, 88, 164, 215 blame, attribution of, 37, 41, 99
Arabs, 81, 215 Blattman, Christopher, 2, 3, 17, 89
in Israel, 81, 114 Bodo, 84
in Mali, 83 Boix, Carles, 93, 95, 146, 167
in Sudan, 82, 89, 115 Bosnia, 2, 68, 80, 138, 139

251
252 Index

Brass, Paul, 27, 38, 40, 41, 61, 211 Congo, Democratic Republic of, 121, 138,
Brubaker, Rogers, 22, 38, 80, 123, 178, 140, 218
209 Connor, Walker, 58
Burundi, 182 conquest, 14
Coser, Lewis A., 23, 38
Cabindan Mayombe, 84 Côte d’Ivoire, 83, 98, 116
Calhoun, Craig, 23, 34, 57 Council of Europe, 218
Cameroon, 87 Cramer, Christopher, 21, 94
Campbell, Donald T., 39, 40 Croatia, 2, 80, 86, 100, 102, 117, 139
categorization, 38 Croats
causation, 176. See also mechanisms in Bosnia, 85
reverse, 52, 90, 211. See also endogeneity in Yugoslavia, 86, 104, 115
spurious, 53 cultural division of labor, 13, 33, 96, 98, 115
Cetinyan, Rupen, 122, 126 Cunningham, David E., 69, 121, 149, 171,
Chad, 85, 87 173, 177, 183, 184, 188, 189, 190, 191,
Chandra, Kanchan, 19, 21 193, 195, 202
Chechens, 4, 85, 111, 112, 113, 145, Cunningham, Kathleen Gallagher, 118
152
Chiapas, 114 Darden, Keith A., 38
Chinese Darfur, 114
outside China, 101, 128, 140 Deiwiks, Christa, 102, 225
civil war democracy
data on. See UCDP/PRIO Armed Conflict data on, 70, 150, 193. See also SIP, Scalar
Data Set Index of Polities
data on ethnonationalist. See ACD2EPR effect on conflict, 74, 156
Data Set relation to discrimination, 221
defined, 41, 59 relation to exclusion, 74, 151
ethnic governmental, 150 relation to vertical political inequality, 147
ethnic territorial, 150 trends in world regions, 213
history of, 64, 70, 77, 131, 154, 176 Derluguian, Georgi M., 112
nonethnic, 150 Deutsch, Karl W., 13
Clausewitz, Carl von, 58, 180 discrimination
cognitive mechanisms. See mechanisms, as part of political relevance, 66
cognitive comparison of world regions, 218–22
Coleman, James S., 36 defined, 68
collective action, 12, 15, 42, 44 effect on conflict onset, 71, 74, 75, 78–82,
dilemma of, 24, 46, 146 83, 86, 87, 89
frames, 42, 83 MAR-based studies of, 29, 61
free riding in, 44, 45, 186 measure of max level, 150
grievance driven, 46, 204 time trend in world regions, 215–8
overcoming dilemma of, 45, 177 Djerma-Songhai, 88
rationalist theory of, 44 Djibouti, 85
revolutionary, 49 Doe, Samuel, 83, 182
role of emotions in, 46, 47–48, 210 domination, 33. See power status, dominance;
role of ethnicity in, 177 minority rule
Collier, Paul, 1, 15, 16, 17, 18, 19, 20, 25, 28, ethnic, 82, 83, 88, 222, 225
51, 52, 83, 86, 93, 125, 130, 148, 175, ethnic, demographic indicator of, 20
177, 178, 185, 190, 195, 211, 223 downgrading, 41. See also power status, loss of
colonialism, 33, 58, 63, 87, 91, 115 effect on conflict onset, 62, 72, 74, 76,
internal, 83, 96, 99, 114, 191, 211 86–7, 107, 131, 205
conflict. See civil war measure of, 69
Congo (Brazzaville), 84, 87, 88, 218 Dudayev, Dzokhar, 113
Index 253

dyad ethnic fractionalization. See also ethnic

between governments and organizations, diversity
173, 183 indices of, 6, 15, 20, 21, 26, 27, 29, 143,
between included and excluded groups, 146 147, 155, 156, 158, 178, 210
interaction between primary and secondary, ethnic groups
135 advanced, 94, 96, 97
power within, 184 as unitary actors, 22–4, 172, 178, 209
primary, 70, 123, 124, 125, 141 backward, 94, 96, 97, 113, 115
primary, effect on conflict, 131 comparison, 40, 117
secondary, 123, 124, 125, 131, 141 data of settlement areas. See GeoEPR-ETH
secondary, effect on conflict, 125–9, 131 Data Set
data on. See EPR-ETH Data Set
East Timorese, 89 defined, 23
Eelam People’s Revolutionary Liberation downgraded, 41. See downgrading; power
Front (EPRLF), 173 status, loss of power
ELF, ethnolinguistic fractionalization index. excluded vs. included, 58, 71, 74
See ethnic fractionalization, indices of incumbent, 58. See ethnic groups, excluded
emotions, 6, 13, 24, 25, 26, 41, 43, 46, 80, vs. included
178, 209, 210 marginalized. See ethnic groups, excluded
ancient hatred. See ancient hatred vs. included
anger, 3, 12, 41, 46, 60, 182, 210 political relevance of, 66
fear, 46, 64, 176 self-excluded, 71
hatred, 14, 19, 25, 64, 81, 171, 172, 178, settlement areas of, 4, 60, 84, 94, 98, 102,
180, 204, 224 117, 122
moral shocks, 46 size of, 62–4, 69, 72, 74, 77, 87, 91, 107,
pleasure of agency, 47 122, 125, 126, 131, 135, 146, 150,
resentment, 3, 25, 41, 46, 48, 61, 64, 81, 158
83, 182, 187, 210, 224 ethnic identities
Emotions, 26, 46 endogeneity of, 178
endogeneity, 36, 52, 53, 78, 89, 90, 91, 92, fluidity of, 19, 178, 180
116. See also causation, reverse; selection freezing of, 23
bais irrelevance as concept, 22
of grievances, 50 Ethnic identities
enduring rivalries, 64 as focal points, 25
EPR-ETH (Ethnic Power Relations) Data Set, ethnic polarization, 20, 26, 27, 147, 178, 185,
4, 66–8 187, 188, 210
Eriksson, Mikael, 69 ethnic security dilemma. See security dilemma,
Eritrean Islamic Jihad Movement (EJIM), ethnic
187 ethnicity. See also ethnonationalism
Estonia, 68, 124, 139, 140, 221 apolitical conceptions of, 26, 27, 57, 65,
Ethiopia, 82, 85, 89, 138 172, 195
ethnic conflict, 6. See also civil war as arbitrary coordination device, 18
banality of, 2, 18, 178 as boundary marker, 96
classical theories of, 12–15 as cause of conflict, 26
cognitive explanations of, 25 as cause of conflict duration, 180
duration of, 172 ascriptive nature of, 187
essentialist explanations of, 21 defined, 23
myth of, 18 politicized, 27, 119, 187, 208
stateless theories of, 27 stickiness of, 181
ethnic diversity, 1, 16, 21, 26, 51, 57, 146–8, transborder reach of, 119
172, 177, 178, 210. See ethnic ethno-demographic diversity. See ethnic
fractionalization, indices of diversity
254 Index

ethnonationalism, 50, 80, 81, 83, 90, 121 GIS, Geographic Information Systems, i, 4,
vs. ethnicity, 6, 27, 92, 210–2 122, 212
ethnonationalist triad, 141 Giuliano, Elise, 111, 112, 115
Ewe, 84 Gleditsch, Nils Petter, 69, 130, 149, 171, 191,
exclusion 213
border-transgressing conflict effect, 141 Goertz, Gary, 64, 145
categories of, 68 Goldstone, Jack A., 16, 28, 45, 49, 62, 93, 211
comparison of world regions, 218–22 Goodwin, Jeff, 41, 42, 46, 49, 50, 51, 62, 182
effect on conflict, 90–1 greed
effect on duration, 191 and grievance, 113
ethnonationalist, 35 as conflict motive, 2, 15, 25, 52, 83, 208,
measure of max level, 150 223
occupational, 13 vs. grievance, 15, 208
political, 1, 3, 7, 32, 33, 34, 41, 50, 98, 107, vs. grievances, 4
114, 119, 138, 172, 193–203. See also grievances
horizontal inequalities, political classical conflict theories, 11–15
political, priority over economic, 99 conflict mechanisms, 35–51
time trend in world regions, 215 critique of skepticism, 20–7
difficulty of measuring, 3, 29
fear. See emotions skepticism in conflict literature, 1–2, 20
Fearon, James D., 2, 3, 6, 15–17, 19, 20, 21, GROWup , Geographic Research On War:
25, 26, 28, 29, 51, 52, 65, 89, 90, 93, unified platform, 4, 212
123, 128, 145, 147, 148, 151, 154, 164, Guatemala, 221
177, 178, 179, 180, 184, 185, 186, 195, Guinea-Bissau, 88
196, 208, 211, 223, 224, 226 Gurr, Ted Robert, 3, 11, 12, 13, 14–15, 16,
Fehr, Ernst, 210 27, 28, 29, 32, 37, 38, 47, 49, 61, 65, 66,
Forsberg, Erika, 85, 183 70, 82, 93, 96, 118, 144, 173, 182, 209,
fractionalization. See ethnic fractionalization 211, 213, 218, 225
framing, 24, 37, 44, 45, 91, 99, 113, 115, Guyana, 88
117
France, 33, 40, 62, 91 Haile Selassie, 82
Free Aceh Movement (GAM), 83 Hale, Henry E., 25, 39, 111, 115
freezing hypothesis, 34 hatred, 64. See also ancient hatred; emotions,
French Revolution, 33, 40, 57, 62 hatred
Fur, 1, 104, 115 Hausa
in Niger, 88
Gamson, William, 42 Hausa-Fulani
Gates, Scott, 3, 18, 70, 98, 150, 184, 185, in Nigeria, 82, 86
191, 193 Hechter, Michael, 13, 15, 45, 46, 47, 96, 98,
G-Econ Data Set, 101, 102, 107, 116 99, 114, 115, 186, 208, 211
Gellner, Ernest, 13, 33, 34, 35, 38, 43, 58, 60, Hedström, Peter, 36
62, 147, 181 Hegre, Håvard, 72, 98, 121, 154, 158, 174,
GeoEPR-ETH (Data Set), 4. See also 213
EPR-ETH Data Set Herrera, Yoshiko M., 40, 99, 111
Georgia, 68, 85, 140 Hoeffler, Anke, 15, 16, 17, 18, 19, 20, 25, 28,
Gilpin, Robert, 33, 57 51, 52, 83, 86, 93, 130, 147, 148, 175,
Gini coefficient, 6, 20, 148, 151, 152, 153, 177, 178, 185, 211, 223
155, 156, 161, 166, 167. See also vertical horizontal inequalities, 3
inequalities country-level measures, 150
Gio, 83 cultural, 32
Girardin, Luc, 21, 29, 58, 66, 119, 121, 122, defined, 31–2
123, 126, 129, 141 economic, 32
Index 255

economic, effect on conflict, 116, 190, 206 Jasper, James M., 24, 42, 46
economic, measures of, 106 Jenne, Erin K., 122, 126
measures of, 6 Jordan, 88, 218, 221, 222
origins of, 33–5, 40 Justice and Development Party, 137
political, 32. See also exclusion, political
political, data on, 65–9. See EPR-ETH Data Kachins, 82
Set Kalyvas, Stathis N., 2, 3, 17, 18, 21, 22, 23,
political, effect on conflict, 90–1, 205 24, 26, 45, 50, 59, 64, 176, 177, 178,
social, 32 180, 184, 208
Horowitz, Donald L., 2, 3, 13–14, 15, 24, 25, Karen, 1, 82, 89, 184, 192
27, 32, 38, 39, 45, 60, 96, 97, 99, 115, Karen National Union, 171, 192
122, 126, 129, 147, 176, 180, 185, 211, Kaufman, Stuart J., 43, 176, 177, 180
225 Kaufmann, Chaim D., 23, 176, 177, 178, 180,
Hroch, Miroslav, 38 224
Huber, John D., 100, 117 Kazakhstan, 140
Hug, Simon, 28, 29 Kemper, Theodore D., 41
Humphreys, Macartan, 17, 18, 83, 116, 138 King, Charles, 18, 119, 125
Huntingon, Samuel P., 12, 213 Kosovo, 80, 81, 86, 113, 116, 121, 138, 225
Hunziker, Philipp, 91, 211 Kosovo Liberation Army (UÇK), 81, 121, 138
Hussein, Saddam, 67, 83, 87, 88, 90, 124, Krahn, 83
137, 160, 215 Kreutz, Joakim, 192
Hutu, 85 Kurdish Democratic Party (KDP), 137
Kurdish Democratic Party of Iran (KDPI), 137
Igbo, 82, 86 Kurdistan Workers’ Party (PKK), 80, 136, 137
Ijaws, 82, 116 Kurds, 1, 5, 34, 80, 87, 136, 137, 140, 141,
India, 65, 84, 89, 153, 158, 173, 195 222
indiscriminate violence, 17, 49, 50, 52. See in Iran, 89
also repression in Iraq, 85, 89, 137
individualism, 6, 13, 16, 20, 24, 44 in Turkey, 68, 79, 80, 124, 129, 136, 222
critique of, 31, 32, 45, 47
in indicators, 6. See also ELF; Gini Laitin, David D., 6, 15–17, 19, 20, 21, 25, 26,
coefficient 28, 43, 51, 52, 89, 93, 139, 145, 147,
Indonesia, 18, 84, 89, 140 151, 154, 178, 179, 185, 186, 195, 208,
inequality. See horizontal inequalities; vertical 211, 223, 224
inequalities Langer, Arnim, 33, 98, 101, 116
comparing effect of horizontal and vertical, Lari/Bakongo, 84
154–60 Latvia, 68, 221
predictive performance of horizontal vs. Lemarchand, René, 85
vertical, 161–4 Liberation Tigers of Tamil Eelam (LTTE),
injustice, 37, 47, 48, 50, 51, 62, 91, 113, 117. 173
See also framing; blame, attribution Liberia, 67, 83, 87, 182
evaluation of, 40–1 Lichbach, Mark Irving, 12, 17, 45
frames, 42 Lustick, Ian S., 222
internal colonialism. See colonialism, internal
Iran, 5, 89, 136, 137, 221, 222 Mack, Andrew, 174, 223
Iraq, 5, 67, 83, 85, 87, 88, 89, 124, 136, 137, Malay-Muslims
160, 167 in Thailand, 114
irredentism, 121, 123, 125, 126, 135, 139, 207 Malaysia, 140
irrelevance-of-grievances claim, 20, 36, 44. See Malesevic, Sinisa, 33
also grievances, skepticism in conflict Mali, 18, 83, 116, 130, 138
literature Mann, Michael, 33, 35, 63, 64, 211
Israel, 7, 68, 81, 90, 114, 153, 218, 221 Mano, 83
256 Index

MAR, Minorities at Risk Data Set, 3, 14–15, Moro, 84

29, 61, 66, 96, 122, 211 motivational mechanisms. See mechanisms,
Marx, Karl, 13 motivational
Marxist theory, 12, 31, 33, 38, 94, 96 Mueller, John E., 2, 18, 19, 20, 26, 177, 178,
Mason, T. David, 42, 45, 173 223
Mbochi, 87, 88 Muller, Edward N., 12
McAdam, Doug, 19, 48 Myanmar, 1, 82, 85, 89, 104, 140, 171, 184,
mechanisms 192
actor-specific, 4
alternative, 4, 53 Naga, 84, 89
cognitive, 6, 13, 25, 209–10 Namibia, 151, 152
conflict-generating, 35–51, 209 nation, 34, 57. See also nationalism
emotional, 25, 26 relation to state, 58, 60, 62
grievance, 30, 205, 208, 211 National Liberation Front of Angola (FNLA),
group-level, 6, 208 84
inequality-generating, 31 national self-determination, 33, 34, 39, 40, 59,
motivational, 6, 209–10 149
observationally equivalent, 52, 53 nationalism, 3, 4, 22, 27, 34, 35, 38, 40, 42,
resource-based, 20 79, 113. See also nation; ethnonationalism
retaining fighters, 185 as macro-historical context, 57
Melander, Erik, 204 as source of categorical inequality, 34
Mengistu Haile Mariam, 82 as template for rebellion, 43
Mercer, Jonathan, 26, 46 defined, 58
Mexico, 114 exclusionary, 35
MFDC, Mouvement des forces démocratiques link to ethnicity, 208
de la Casamence, 195 modular nature of, 43
Miguel, Edward, 2, 3, 17, 89 principles of, 33, 35, 59
Milošević, Slobodan, 80, 86, 113, 139 shift away from in literature, 210
Min, Brian, 27, 63, 66, 85 structural theories of, 43
Minorities at Risk Data Set. See MAR violation of principles, 181
minority rule, 63, 83, 85, 87, 88, 127, 167, nationalist systems change, 57, 58
182, 221, 222 NATO, 81
Mizo, 84 Ndadaye, Melchior, 85
mobilization Nepal, 88, 98, 191
as path-dependent process, 34 new wars
collective, 18 theories of, 18, 26, 27
conflict-generating, 44–8, 83 Niger, 88, 89, 116, 138
ethnic focal points in, 25 Nigeria, 18, 82, 87, 88
ethnonationalist, 23, 224 Nordhaus, William D., 94, 101, 102, 103,
fueled by horizontal inequalities, 98 104
governmental response to, 49 NSA, Non-State Actors Data Set, 69, 149,
grievance-based, 17, 24 191. See also UCDP ACD Data Set
group-level, 32
induced by horizontal inequalities, 111, Oberschall, Athony, 12, 19, 46, 48, 186
112, 115 Ogaden, 82, 138
of the masses, 33, 34, 35, 37, 128 Olson, Mancur, 24, 44, 46, 47
opportunity-based, 12 Olzak, Susan, 13, 35, 66
resource, 14, 24, 95, 167, 172. See social Opp, Karl-Dieter, 24, 45
movement theory opportunities
Moldova, 140 vs. grievances, 4, 15, 18, 25, 63, 177,
Moore, Barrington Jr., 12, 42 208
moral shocks. See emotions opportunity structures, 24, 95, 179, 185
Index 257

Organization of Security and Cooperation in relation to ethnic groups, 172–4, 176, 178,
Europe (OSCE), 218 184, 193
Oroma, 85 relation to excluded groups, 198
Østby, Gudrun, 94, 97, 98, 100, 211 relation to TEK, 189–90, 198
resilience, 186–8
Palestinians, 1, 7, 34, 81, 90, 114, 222 strength of, 193
in Israel, 68 territorial control, 193
in Jordan, 88 Regan, Patrick M., 11, 28, 66, 121, 189, 211
Pappe, Ilan, 81, 114, 222 relative deprivation theory, 11, 12, 13, 14, 15,
Papuans, 84, 89 19, 28, 30, 47, 144
Paraguay, 171 repression, 1, 17, 32, 44, 48, 50, 59, 61, 89,
Patriotic Union of Kurdistan, 137 95, 112, 114, 118, 127, 147, 148, 184,
Petersen, Roger, 7, 26, 35, 41, 43, 46, 62, 64, 223
80, 182, 187, 210 resentment. See emotions
Philippines, 84, 114, 158 resource mobilization. See mobilization,
Polletta, Francesca, 42, 43, 46 resource
Posen, Barry R., 26, 64, 176, 179 reverse causation. See causation, reverse
Posner, Daniel N., 21, 25, 27 Rød, Jan Ketil, 63, 66, 97, 102
power sharing, 5, 41, 59, 62, 67, 71, 82, 85, Roessler, Philipp G., 35, 90, 181, 188, 224
218. See also power status, power sharing Rokkan, Stein, 34, 208
regimes Ross, Michael L., 18, 145
as conflict-resolution method, 224–5 Rothchild, Donald, 204, 208, 224, 225
critique of, 6, 204 Rothchild, Joseph, 61, 62, 181, 182, 187,
power status 188
data on. See EPR-ETH Data Set Russia, 4, 65, 85, 94, 99, 111, 112, 113, 124,
discrimination, 68 125, 139, 140, 141, 145, 152, 159, 160,
dominance, 67 163
effect on conflict duration, 207 Russians
excluded categories, 68 in Estonia, 139, 140
excluded vs. included, 71 outside Russia, 68, 112, 120, 122, 128, 139,
junior partner, 68 140
loss of, 76. See downgrading Rwanda, 2, 121, 138, 167, 176, 218, 221,
monopoly, 67 222
of ethnic groups, 61, 62, 66, 67, 188 Rwandan Patriotic Front, 138
of TEK groups, 123, 141, 181, 207
power sharing regimes, 67 Saideman, Stephen M., 28, 66, 122, 126, 139,
powerless, 68 173
regional autonomy, 68 Salehyan, Idean, 69, 88, 119, 121, 127, 128,
senior partner, 67 149, 171, 173, 183, 184, 188, 189, 191,
separatist autonomy, 68 193, 195
undivided power, 67 Sambanis, Nicholas, 16, 17, 18, 21, 59, 72,
prediction 93, 101, 121, 147, 158, 176, 226
out-of-sample, 88, 161–4 Sara, 85
primary dyad Saudi Arabia, 99, 116, 221, 222
primary. See dyad, primary Scott, James C., 12, 82, 84, 95
secession, 48, 49, 59, 61, 82, 84, 85, 86, 96,
rebel organizations 98, 99, 113, 115, 138, 158, 167, 186,
as firms, 177 187, 218, 226
as unit of analysis, 177, 209 secondary dyad. See dyad, secondary
data on, 69, 191. See ACD2EPR Data Set; security dilemma
NSA Data Set ethnic, 26, 64, 176, 177, 179
recruitment and retention, 185–6 Selbin, Eric, 43
258 Index

selection bias, 27, 28, 66, 211. See also Tamil Eelam Liberation Organization (TELO),
endogeneity 173
Serbia, 4, 80, 102, 138, 139 Tamils, 1, 173
Serbs, 100 Tarrow, Sidney, 17, 45
in Bosnia, 68, 138, 139 Taylor, Charles, 83, 182
in Croatia, 80 Teso, 88
Shan, 82 Thailand, 114
Shlaim, Avi, 81 Tibet, 84
Sikh, 85 Tilly, Charles, 12, 14, 17, 19, 30–2, 34, 45,
Simmel, Georg, 23, 38 47, 58, 96, 101, 116, 181, 187
SIP, Scalar Index of Polities, 70, 150, 151, Toft, Monica, 173, 176, 188, 224
155, 193 transborder ethnic kin (TEK), 119
Skocpol, Theda, 12, 49 data on, 130
Slovenes, 86, 98, 102, 103, 104, 115 effect on conflict, 140–2
Slovenia, 80, 86, 100, 102, 117 effect on conflict duration, 190
Smith, Antony D., 43 in conflict literature, 123
Snyder, Jack, 176, 179 included vs. excluded, 124
social identity theory, 25, 39 Treisman, Daniel S., 99, 115
social movement theory, 45. See also Trinidad and Tobago, 88
mobilization Tripuri, 84
Sollenberg, Margareta, 69 Tuareg, 83, 116, 138
Somalia, 38, 138 Tuđman, Franjo, 80
sons-of-the-soil conflicts, 16, 84, 179, 180, Tullock, Gordon, 24, 177
195, 211 Turkey, 1, 5, 62, 79, 80, 124, 129, 136, 137,
as special case of exclusion, 16, 196, 211 221
Sornette, Didier, 57, 58 Turner, John C., 13, 25, 39, 209
South Africa, 63, 83, 116, 127, 163, 164, 167, Tutsi, 85, 138, 140
186
South Ossetians, 85 ubiquity-of-grievances claim, 19, 36, 39, 51.
sovereignty See also grievances, skepticism in conflict
full or partial, 59 literature
popular, 33, 34 UCDP Conflict Termination Data Set,
territorial, 57, 84 192
Spain, 98, 115, 222 UCDP Non-State Actors Data Set, 149
Sri Lanka, 1, 173, 187 UCDP/PRIO Armed Conflict Data Set, 69,
state. See also sovereignty 130, 149, 150, 161, 164, 165, 171, 173,
ethnic neutrality of, 20, 26–7 191
repression by. See repression UFLA, United Liberation Front of Asom
state-to-nation deficit, 58, 60. See also nation, (Assam), 195
relation to state Uganda, 83, 88
Stewart, Frances, 3, 13, 31, 32, 37, 43, 93, 94, Ukraine, 140
96, 97, 98, 101, 104, 116, 211 United Kingdom, 2, 81, 91, 114, 153
Strand, Håvard, 69, 213 United States, 33, 40, 87, 90, 215
Stroessner, Alfredo, 171
Sudan, 1, 18, 82, 84, 89, 99, 104, 218, 222 Van Evera, Stephen, 120, 122
Swedberg, Richard, 36 Vanhanen, Tatu, 70, 147, 151
Swiss French, 68 vertical inequalities
Swiss German, 68 country-level measures, 150–1
Swiss Italians, 68, 129 defined in contrast to horizontal, 32
Switzerland, 68 economic, classical literature on, 94–6
Syria, 63, 88, 164, 218, 221, 222 Vietnam, 140

Tajfel, Henri, 13, 25, 39 Wallensteen, Peter, 69, 149

Index 259

Walter, Barbara F., 66, 85, 122, 128, 173, Wimmer, Andreas, 27, 63, 66, 74, 85,
175, 177, 178, 179, 182, 184, 187, 188, 181
189, 204, 224 Wood, Elisabeth J., 3, 17, 47–8, 50, 93
Ward, Michael D., 28, 149, 160, 213 Woodward, Susan L., 113, 115
Warren, T. Camber, 57, 58 Wucherpfennig, Julian, 91, 102, 149, 180,
weakest link 183, 195, 209, 211, 225
principle of, 145, 146, 166, 202
Weber, Max, 31 Yoruba, 82
Weidmann, Nils B., 98, 102, 117 Yugoslavia, 68, 80, 98, 100, 103, 113, 115,
Weiner, Myron, 16, 43, 84, 121, 123, 135, 117, 121, 122, 176, 178
141, 179, 195
Weinstein, Jeremy, 3, 17, 45, 179, 185 Zimbabwe, 83, 87, 152
Wilkinson, Steven I., 21, 60 Zionism, 81, 114. See also ethonationalism
Williams, Robin M., 33, 39, 40, 41, 64, 182 Zürcher, Christoph, 85, 113