Why Do Ethnic Groups Rebel?
New Data and Analysis
By Lars-Erik Cederman, Andreas Wimmer, and Brian Min*
DESPITE its fundamental role in legitimizing the modern state
system, nationalism has rarely been linked to the outbreak of political
violence in the recent literature on ethnic conflict and civil war.
To a large extent, this is because the state is absent from many conventional
theories of ethnic conflict. Indeed, some studies analyze conflict
between ethnic groups under conditions of state failure, thus making
the absence of the state the very core of the causal argument. Others
assume that the state is ethnically neutral and try to relate ethnodemographic
measures, such as fractionalization and polarization, to civil
war. In contrast to these approaches, we analyze the state as an institution
that is captured to different degrees by representatives of particular
ethnic communities, and thus we conceive of ethnic wars as the result
of competing ethnonationalist claims to state power.
While our work relates to a rich research tradition that links the
causes of such conflicts to the mobilization of ethnic minorities, it also
goes beyond this tradition by introducing a new data set that addresses
some of the shortcomings of this tradition. Our analysis is based on
the Ethnic Power Relations data set (epr), which covers all politically
relevant ethnic groups and their access to power around the world from
1946 through 2005. This data set improves significantly on the widely
used Minorities at Risk data set, which restricts its sample to mobilized
*Correspondence should be directed to Lars-Erik Cederman (lcederman@ethz.ch). We would
like to thank the many individuals who helped assemble the data set on which this article relies. We
cannot list all country experts who generously shared their knowledge, but we should like to mention
Dennis Avilés, Yuval Feinstein, Dmitry Gorenburg, Wesley Hiers, Lutz Krebs, Patrick Kuhn, Anoop
Sarbahi, James Scarritt, Manuel Vogt, Judith Vorrath, Jürg Weder, and Christoph Zürcher. Luc Girardin
implemented the software for the online expert survey.The authors acknowledge financial support
from the Swiss National Science Foundation through Grant No. 105511-116795, and from UCLA’s
International Institute. We are grateful for helpful comments by the editors and anonymous referees
of this journal, Kristian Skrede Gleditsch, Simon Hug, and Idean Salehyan, as well as audiences at
eth Zürich, the Graduate Institute of International Studies in Geneva, Hebrew University in Jerusalem,
University of St. Gallen, and Princeton University.
World Politics 62, no. 1 (January 2010), 87–119
Copyright © 2010 Trustees of Princeton University
doi: 10.1017/S0043887109990219
88	 world politics
1
Complementing the current study, Wimmer, Cederman, and Min 2009 show that ethic conflict
results from specific ethnopolitical configurations of power, rather than from ethnic diversity per se.
They identify three particularly conflict-prone configurations: ethnocracies, states with a high number
of power-sharing ethnic elites, and incohesive states with a short history of direct rule by the center.
2
Posen 1993.
3
Hardin 1995; de Figueiredo and Weingast 1999.
minorities and thus largely overlooks the ethnopolitical constellation
of power at the center.
Improved theory and data allow us to show that, contrary to the
expectations held by many scholars of civil wars, competing ethnonationalist
claims over the state constitute the driving force behind
many internal conflicts in the post–World War II era. While we have
analyzed this data set at the country level in another publication,1
we
pursue a more disaggregated, group-level analysis here. We show that
conflict with the government is more likely to erupt (1) the more representatives
of an ethnic group are excluded from state power, especially
if they experienced a loss of power in the recent past, (2) the higher
their mobilizational capacity is, and (3) the more they have experienced
conflict in the past. In view of these findings, we conclude that
ethnonationalist struggles over access to state power are an important
part of the dynamics leading to the outbreak of civil wars.
The article is organized as follows. We first review the relevant conflict
literature and then develop the three main hypotheses that lie at
the core of our theory. The following section introduces the epr data
set. The main results are presented in the next section, followed by a
brief sensitivity analysis and a concluding section.
Existing Approaches to Ethnicity and Conflict:
Security Dilemma, Ethnic Fractionalization, and
Minority Mobilization
In order to understand the conflicts that broke out in Yugoslav and
Soviet successor states as well as in Rwanda during the 1990s, some
researchers have utilized ideas developed to study interstate relations
during the cold war. Most prominently, Posen conceived of ethnic conflict
as a struggle between ethnic groups in the wake of state collapse.2
According to neorealist theory, ethnic groups face a security dilemma
when the Leviathan disappears and react with preemptive violence.
These ideas were subsequently elaborated with the help of rational
choice models.3
As a consequence of the assumption of state breakdown, however,
this research tends to overlook the important role played by state actors
	 why do ethnic groups rebel?	 89
4
Collier and Hoeffler 2004; Fearon and Laitin 2003; for a review, see Kalyvas 2007.
5
Fearon and Laitin 2003. Using different model specifications, other quantitative studies report a
positive impact of ethnic fractionalization on civil war onset; see Sambanis 2001; Hegre and Sambanis
2006; Blimes 2006.
6
For example, Montalvo and Reynal-Querol 2005; cf. Horowitz 1985.
7
For example, Collier and Hoeffler 2004.
8
A number of scholars have called for a more disaggregated approach that focuses on microlevel
mechanisms, for example, Kalyvas 2006; for a review, see Tarrow 2007. Our own approach complements
these studies at a level of analysis that is less detailed yet offers global coverage.
9
Posner 2004.
10
Chandra and Wilkinson 2008.
11
Cederman and Girardin 2007.
12
Brass 1991; Breuilly 1994; Wimmer 2002.
in generating these conflicts in the first place, as the wars in Yugoslavia,
Rwanda, and many other places clearly demonstrate. In the absence of
state agency, political violence may take the form of communal conflict
over land or local political dominance but not of full-fledged civil war.
In recent statistical research on civil wars, many scholars argue that
ethnic grievances are too widespread to explain the rare onset of conflict.
Wars are more likely, so the argument goes, in states that are too
weak to suppress rebellions or where natural resources invite warlords
to enrich themselves by looting.4
Research on the basis of this grievance
hypothesis has undoubtedly helped to clarify the general conditions
that are conducive to civil wars and insurgencies. However, the
grievance hypothesis has not been tested with adequate data; rather,
it has been tested with highly aggregated proxies that do not provide
a direct measure of political inequality along ethnic lines and the resulting
“grievances.” Fearon and Laitin, for example, examine whether
there is a statistical association between measures of a country’s “ethnic
fractionalization” and civil war onset.5
Some scholars have worked with
measures of ethnic polarization that are loosely related to Horowitz’s
theory of ethnic conflict, again without explicit references to the state.6
Others seek to operationalize the concept of ethnic domination but use
a demographic proxy as well.7
We believe that efforts to grasp the propensity for ethnonationalist
conflict with the help of macrolevel indices are problematic.8
First, they
implicitly assume that the ethnic groups listed in the work of anthropologists
and linguists are politically relevant.9
Second, the macrolevel
indices describe a country’s demography,10
which may or may not be
related to the actual constellation of power at the state center.11
Qualitative
studies of ethnic conflict show that in such cases the state is not
an ethnically neutral institution but is an active agent of political exclusion
that generates these conflicts in the first place.12
90	 world politics
A third stream of research starts from this insight and studies the
conditions under which minorities will mobilize against the state and
also the conditions under which such mobilization will turn violent.13
Building on work in the relative deprivation tradition, the Minorities
at Risk (mar) data set established by Gurr and his colleagues remains
the most prominent data source used to evaluate ethnic mobilizations
and violence at the group level.14
Scholars in this tradition have studied
the consequences economic, political, and cultural discrimination (see
below), the settlement patterns that enhance minority mobilization for
conflict,15
domestic diversion mechanisms,16
the dynamics of secessionist
bargaining,17
and third-party intervention,18
as well as the role
of country-level factors, such as government responses to autonomy
claims by ethnic minorities19
and broader international contextual factors
facilitating ethnic mobilization.20
While the mar data set allows for empirical testing of mechanisms
linking group characteristics to conflict propensity, it has its limitations.
We note that the mar-based literature has produced somewhat
conflicting results regarding the question most relevant to this article:
whether or not political disadvantage and discrimination increase the
likelihood of ethnic rebellion. In fact, whereas some studies find that
political disadvantage has an impact on the likelihood of armed rebellion
and secession,21
others find that the degree of political exclusion
has no effect on secessionism.22
The picture is even more mixed as regards
the effect of political discrimination: while Regan and Norton,
as well as Walter, find strong evidence that political discrimination increases
rebellions and secessionist civil wars,23
Fox fails to find any clear
relationship for the subset of ethnoreligious groups,24
and Gurr’s study
of ethnonationalist rebellions in the 1980s even suggests that political
discrimination is associated with less rather than more conflict.25
Olzak
aggregates mar data on the country level and arrives at the conclusion
13
There is also a vast qualitative literature on minority mobilization and ethnonationalist violence;
see Brubaker and Laitin 1998.
14
Gurr 1993a; Gurr 2002.
15
Toft 2003.
16
Tir and Jasinski 2008.
17
Walter 2006a; Jenne 2007.
18
Cetinyan 2002; Saideman 2002.
19
Brancati 2006; Walter 2006b.
20
Olzak 2006.
21
For example, Gurr 1993b; Walter 2006b.
22
Saideman and Ayers 2000.
23
Regan and Norton 2005; Walter 2006b.
24
Fox 2000.
25
Gurr 1993b.
	 why do ethnic groups rebel?	 91
that both formal recognition of ethnic group rights and political discrimination
increase the likelihood of conflict.26
Clearly, some of these discrepancies can be attributed to different
research designs and sample definitions, but we suspect that the
data set’s inherent problems might be responsible for these conflicting
findings as well. The mar data set “hardwires” the degree of power
access into the sample definition by excluding groups in power from
systematic consideration. This reduces the comparative horizon and
thus makes it harder to capture the effects of political exclusions in unambiguous
ways. Moreover, in many countries with dramatic shifts in
power constellations (Chad, Afghanistan, Liberia), the political status
of an ethnic group may change from discriminated minority to ruling
elite from one period to the next. Indeed, studies of ethnonationalism
should treat ethnic groups’ representation within government as a
variable rather than as a constant. Finally, focusing on minorities conflates
the demographic concept of numerical domination with political
exclusion. Accordingly, the mar coding scheme does not fit countries
with ruling minorities or complex coalitions of ethnically defined elites,
as for example in Nigeria, India, or Chad, where ethnic conflict will be
pursued in the name of excluded majorities (rather than minorities) or
ethnic groups that share power (and are thus not “at risk”).27
In sum, much of the recent literature on ethnic conflict and civil
wars fails to get the state’s role right. Many approaches do not take account
of the state as an actor in conflict processes (as in the security dilemma
approach), fail to trace the ethnopolitical power constellations
at the center of state power (as in the minority mobilization school), or
try to capture ethnopolitical discontent through demographic proxies
of diversity. These theoretical, sampling, and measurement problems
hinder the development of precise and testable hypothesis about which
mechanisms connect ethnonationalist politics to political violence. The
following section addresses this task.
Theorizing Ethnonationalist Conflict: Actor
Constellations, Motives, and Collective Action
The classical sociological literature on nationalism offers a good starting
point for understanding the logic of ethnonationalist conflict.
26
Olzak 2006,124.
27
The mar data set tries to address these limitations by including five “advantaged” minorities
that benefit from political discrimination. mar also comprises a series of “communal contenders”
mostly in Africa, that is, groups that share power with others while at the same time mobilizing in
protest or rebellion.
92	 world politics
Nationalism can be defined as a political principle that demands that
the unit of governance and the nation should be congruent. It replaces
the principle of dynastic sovereignty on which agrarian empires were
based.28
Whereas the unit of governance is usually understood to be
the sovereign state, the nation is imagined as a community of common
origin and shared historical destiny.29
In the modern era the introduction of the principles of nationalism—that
ethnic likes should rule over ethnic likes—has led to waves
of political mobilizations and a subsequent transformation of the state
system.30
The ensuing ethnonationalist struggles have taken on a variety
of forms: conflict over access to state power between the leaders
of competing ethnic communities,31
secession from existing states in
order to establish a new state ruled in the name of a particular ethnic
group32
or to join another state controlled by ethnic kin,33
and competition
between new states over mixed territories inhabited by members
of their respective ethnic core groups.34
While this literature helps to explain the broad historical and political
circumstances under which we can expect ethnonationalist mobilization
and conflict, it is much less concerned with identifying the
precise mechanisms through which such ethnonationalist mobilization
turns violent. Only recently have a number of scholars turned to studying
nationalist mobilization and conflict in greater detail.35
We selectively
draw on this scholarship in order to develop our own model of
ethnonationalist conflict.
Our analytical point of departure is Tilly’s polity model, which features
a political system comprising a government and a number of contenders
seeking to maximize their access to executive power.36
Members
of the polity enjoy a privileged position, while those excluded from
direct access to government represent potential challengers (see Figure
1). Adapting Tilly’s model along the lines proposed by Cederman and
Girardin, we assume that polity members and challengers consist of
ethnic groups and their leaders (including politically irrelevant groups
28
Gellner 1983.
29
Anderson 1991.
30
Kedourie 1960; Breuilly 1994; Brubaker 1996; Cederman 1997; Wimmer and Min 2006.
31
For example, Brass 1991; Wimmer 2002.
32
For example, Hechter 2001.
33
For example, Weiner 1971.
34
For example, Brubaker 1996.
35
For example, Beissinger 2002; Cederman 1997, chaps. 7, 8; Hechter 2000; Olzak 2006; Wimmer
2002, chap. 3.
36
Tilly 1978.
	 why do ethnic groups rebel?	 93
such as group 4).37
Given the principle of ethnonational representativity
embodied by the modern nation-state, it can be expected that
challengers seek to avoid the rule of ethnic others by gaining access
to the polity or leaving it in favor of a new polity or an already existing
kin state. Power holders should want to maximize their share of
state power—thus opening the possibility of infighting among powersharing
partners.
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 war38
but
excludes communal conflicts or pogroms in which the state plays less
of an active role. In conflicts that are fought in the name of excluded
groups, rebel movements are composed of mobilized and militarized
37
Cederman and Girardin 2007 refer to “Ethnic Groups in Power” (egips) and “Marginalized
Ethnic Groups” (megs).
38
Sambanis 2004; Kalyvas 2007.
Figure 1
The Polity Model with Included and Excluded Ethnic Groups
Excluded
Group 1
Excluded
Group 3
Excluded
Group 2
Excluded
Group 4
Included
Group 2
Included
Group 1
Included
Group 3
GovernmentPolity
State
94	 world politics
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.
Having identified the main actors and the types of conflicts they may
provoke, we now consider the conditions under which violence is most
likely to erupt. We first study which groups will be more motivated to
support a rebellion against ethnic domination by others. Indeed, many
aggrieved groups have not produced militant rebel organizations. We
therefore have to identify those groups that perceive the government as
particularly illegitimate and are therefore more inclined to support rebellions.
Second, we consider a group’s organizational capacity to challenge
incumbent state power because high levels of motivation alone
will not suffice to produce an armed organization willing to take on the
government army. Finally, rather than constituting historical singularities,
political violence often leaves traces that put nationalist politics
on a contentious track. We therefore need to identify such potential
effects of path dependency. In the following, we consider the causal
mechanisms associated with each of these three mechanisms in turn.
Most Motivated Groups: Excluded, Downgraded,
and Underrepresented
To specify motives and identify the most motivated actors, we draw
on Wimmer’s institutionalist approach to nationalism and ethnic politics,
which assumes that rulers in modern nation-states are no longer
legitimized by the principles of dynastic succession, God’s grace, or
civilizational progress but that they are expected to care for their own,
ethnically defined people.39
Political officeholders thus have institutional
incentives to gain legitimacy by favoring coethnics over others
when it comes to the distribution of public goods and government jobs.
The expectation of ethnic preference works in the other direction as
well, as voters prefer parties led by coethnics, delinquents hope for coethnic
judges, and citizens prefer to be policed by coethnics.
Under the conditions of pervasive ethnic favoritism, political leaders
and followers are driven by the strategic motive to avoid or even
to overturn dominance by ethnic “others.” This motive is simultaneously
material, political, and symbolic: “adequate” or “just” representation
in a central government offers material advantages such as access
39
Wimmer 2002.
	 why do ethnic groups rebel?	 95
to government jobs and services, legal advantages such as the benefits
of full citizenship rights, a fair trial, and protection from arbitrary violence,
and symbolic advantages such as the prestige of belonging to a
“state-owning” ethnic group. In brief, this approach conceives of ethnic
politics as the struggle over control of the state between various ethnically
defined organizations and their constituencies. Especially in weak
states with weakly developed civil societies, the state may be captured
by particular ethnic elites and their constituencies, thus giving rise to
one among other variants of the “weak state–strong society” constellation
analyzed by Midgal and others.40
It follows that groups that lose out in this struggle for state power
are more fertile breeding grounds for organizations that challenge the
government. We postulate a direct relationship between the degree of
state power and the likelihood that an armed rebellion will be instigated
in the name of that particular group.The most excluded groups will thus
be most likely to support armed organizations that challenge the government.
Given nationalist principles of political legitimacy, feelings of
resentment will be widespread and can be channeled into successful collective
action.41
We summarize this reasoning in a first hypothesis:
—H1a. The probability of ethnonationalist conflict increases the more
representatives of an ethnic group are excluded from central executive
power.
The exclusion mechanism also alerts us to the consequences of
changes in power hierarchies. Sociological theories of emotions suggest
that negative emotions are especially likely to be aroused following
loss of power and prestige.42
When the subjects blame others for their
downgrading, anger and resentment increase the readiness to fight in
order to change the situation.43
We postulate a similar mechanism at
the level of collective organizations: leaders of ethnonationalist organizations
will be most likely to resort to violence if they have recently
experienced a loss of relative power. They can channel the resentment
of their constituencies and mobilize to “reverse a reversal.”44
This reasoning
leads to the following hypothesis:
40
Migdal 1988; see also Jackman 1993.
41
Petersen 2002.
42
Kemper 1978.
43
Turner and Stets 2005, 217.
44
Petersen 2002, 173. The mar data set includes an indicator for “lost autonomy” that is associated
with rebellions; see Gurr 2000, 230; Saideman and Ayres 2000; Walter 2006b. Note, however,
that our hypothesis concerns all types of status reversals, including exclusion or demotion that were
not preceded by autonomous status.
96	 world politics
—H1b. The probability of ethnonationalist conflict increases following
a fall in power status that decreases access to central executive power
by representatives of an ethnic group.
Finally, the principle of ethnic representativity of modern nationstates
is also violated if groups in power are “underrepresented” in comparison
with other power-sharing partners. When smaller groups wield
more power than larger groups, representatives of the larger groups can
portray the situation as unfair and stir up fear of ethnic domination
among their constituents.45
Thus, we have the next hypothesis.
—H1c. The probability of ethnonationalist conflict increases if the
ethnic group represented by a power-sharing partner is larger compared
with that of more powerful coalition partners.
Mobilizational Capacity: Group Size
Collective action theory tells us that group motivations are insufficient
to produce political mobilization and violent contestation.46
The motivational
forces described above are thus a necessary, but not a sufficient,
cause for ethnic conflict. Successful mobilization requires both motivation
and organizational capacity.47
While neoclassical collective action theory
in the Olsonian tradition expects free riding in large groups, nationalists
may overcome such dilemmas through intragroup monitoring, by relying
on preexisting social networks, and by mobilizing identity-related cooperation
norms.48
Following resource mobilization theory, we postulate
that larger excluded groups are even more able to challenge a government
because they can draw on their superior numbers to recruit fighters
and have a larger potential resource pool to sustain an organizational
infrastructure.49
As argued by Cederman, Buhaug, and Rød, the political
claims of larger ethnic groups also enjoy more legitimacy: given the
principles of representativity that underlie the nation-state, the exclusion
of large sections of the population from power is more scandalous than
the exclusion of smaller groups, and minority-ruled states (ethnocracies)
are among the least legitimate political regimes in the modern world.50
Based on this argument, we postulate the next hypothesis.
—H2. The probability of ethnonationalist conflict increases with the
ethnic group’s relative demographic size.
45
For a general argument based on intergroup comparison, see Horowitz 1985; and Gurr 2000.
46
McCarthy and Zald 1977; Tilly 1978; Tilly and Tarrow 2006.
47
Gurr 2000.
48
Hechter and Okamoto 2001. Furthermore, free riding may be less problematic in many civil war
situations because of the risks associated with nonparticipation; see Kalyvas and Kocher 2007.
49
McCarthy and Zald 1977; DeNardo 1985.
50
Cederman, Buhaug, and Rød 2009; see also Cederman 1997, chap. 8; Petersen 2002, 51.
	 why do ethnic groups rebel?	 97
History Matters: The Influence of Past Conflict
Ethnonationalist mobilization and contestation are macrohistorical
processes that operate over both short51
and long time spans.52
It may
take decades until perceived humiliation and unfair ethnic status hierarchies
give rise to political mobilization and conflict. Thus, rather
than being an instant and ahistorical phenomenon, nationalist mobilization
takes place in a historical context that might be characterized
by previous episodes of ethnonationalist violence. In extreme cases of
path dependency, actors may find themselves trapped in self-sustaining
cycles of violence.
We postulate that past conflicts influence the likelihood of present
conflict through three mechanisms. First, ethnonationalist activists attempt
to glorify their group’s history through one-sided narratives that
stress their own victories and attribute blame for military losses to traitors,weak-spirited
leaders,or a ruthless enemy.This implies that leaders
might not update their risk assessments and might take up arms again
even when the chances of winning have not improved significantly.53
Second, past experiences of traumatic violence may live on as a part of
oral tradition or they may sometimes be perpetuated in official history
textbooks and public rituals, nourishing calls for revenge.54
Third, prior
exposure to combat means that violence is no longer unthinkable but
constitutes part of the accepted repertoire of action and may help create
organizational structures and identities that can be reactivated at later
points in history or even create a culture of violence.55
We express these
three mechanisms of path dependency in our third main hypothesis.
—H3. The probability of ethnonationalist conflict increases with the
number of prior conflicts fought in the name of the same ethnic group.
Needless to say, this set of hypotheses does not exhaust the links between
ethnicity and internal conflict. Properties of ethnic groups, such
as concentrated settlement patterns,56
cohesive internal structure,57
and
51
Beissinger 2002.
52
Wimmer and Min 2006.
53
Rydgren 2007.
54
Kalyvas 2007.
55
Laitin 1995; Waldman 2004. For alternative explanations of recurrent warfare, see Walter 2004.
Conflict experiences involving other groups may also increase the likelihood of ethnonationalist violence
through diffusion mechanisms; see Lake and Rothchild 1998.
56
Using the greg (Geo-Referencing Ethnic Groups) data set, which is based on a geo-coded version
of the Atlas Narodov Mira, Weidmann 2009 shows that there is a strong link between settlement
concentration and conflict, thus confirming Toft 2003. Data collection that will provide a gis-based
version of epr is under way.
57
Tilly 1978; Gurr 2000.
98	 world politics
availability of external support,58
are known to increase the probability
of civil wars, but investigating these factors requires data that are presently
unavailable in our data set.
Before operationalizing and testing our main hypotheses, we turn to
the data set that underlies this study. Evaluating the theory calls for data
that code access to executive power for representatives of different ethnic
groups, as well as the conflicts in which they have been involved.
Measuring Ethnic Power Relations
The Ethnic Power Relations (epr) data set identifies all politically relevant
ethnic groups and their access to state power in all years from 1946
to 2005.59
In collecting the data, we relied on the expert input of nearly
one hundred students of ethnic politics to assess formal and informal
degrees of political participation and exclusion along ethnic lines.60
In
line with most other data sources on ethnic politics including mar, the
current data set takes ethnic groups as units of observation, rather than
political organizations that claim to speak in their name.61
This data set
improves significantly on previous efforts to code ethnic groups’ access to
power, such as that of Cederman and Girardin, who rely on static measures
of inclusion and limit their sample to Eurasia and North Africa.62
Politically Relevant Ethnic Groups
We define ethnicity as any subjectively experienced sense of commonality
based on the belief in common ancestry and shared culture.63
Differ-
58
The coding of transborder links among groups represented in epr is in progress. Based on the
greg data set, Cederman, Girardin and Gleditsch 2009 find a size-dependent kin effect within Eurasia.
For a more general treatment of transnational factors of civil wars, see Salehyan 2009.
59
The epr data set, which can be downloaded from http://dvn-iq.harvard.edu/dvn.dv/epr, is described
fully with complete coding rules in Min, Cederman, and Wimmer 2009 and in an online appendix.
The data set includes politically relevant ethnic groups in all 155 sovereign states with a population
of at least one million and a surface area of at least five thousand square kilometers as of 2005.
60
The process of contacting and interacting with country experts took almost two years. Once
sufficient coding responses were available, workshops were held to decide on the final coding. We
discussed each coding in light of comments from the experts present, as well as of information from
additional data sources and the accumulating comparative knowledge of the project team itself.
61
The main reason for this methodological decision is that governmental and rebel organizations
are much more prone to change (for example, through processes of fusion, fission, exit, and entry) than
politically relevant ethnic groups, requiring more intricate organization-level data than are currently
available (but see the preliminary efforts currently under way at mar). Recently efforts have been
made to code nonstate actors in civil wars, but in such cases the focus is on rebel organizations and
thus excludes political organizations in nonconflict cases; see, for example, Cunningham, Gleditsch,
and Salehyan 2009.
62
Cederman and Girardin 2007; Buhaug, Cederman, and Rød 2008; though see Cederman, Buhaug,
and Rød 2009 for a global analysis.
63
Weber 1978, 385–98. The following definition of ethnicity relies on Wimmer 2008.
	 why do ethnic groups rebel?	 99
ent markers may be used to indicate such shared ancestry and culture:
common language, similar phenotypical features, adherence to the
same faith, and so on.64
Ethnic groups might be hierarchically nested;
that is, they may comprise several levels of differentiation, not all of
which are necessarily politically relevant. We classify an ethnic group
as politically relevant if at least one political organization claims to represent
it in national politics or if its members are subjected to state-led
political discrimination.65
Discrimination is defined as political exclusion
directly targeted at an ethnic community—thus disregarding indirect
discrimination based, for example, on educational disadvantage or
discrimination in the labour or credit markets. The coding rules allow
for the identification of countries or specific periods in which national
politics was framed in nonethnic terms.66
To capture temporal changes in the set of politically relevant groups
and their access to political power, the sample period from 1946 to 2005
was divided into subperiods, for which separate codings were provided.
The next important step was to code the degree of access to power by
the political leaders representing these various ethnic communities.
Coding Access to Power
The power-access coding is limited to executive power only.67
Depending
on a given country’s power constellations, executive power amounts
to control over the presidency, the cabinet, and senior posts in the administration,
including the army. Experts were encouraged to capture the
most relevant dimension (for example, in a military dictatorship, power
over the army, and in presidential systems, the presidency, and so on). We
were interested primarily in major power shifts rather than day-to-day
reorganizations of cabinets, limiting ourselves to absolute access to power
irrespective of demographic underrepresentation or overrepresentation.
All politically relevant ethnic groups were categorized according
to (1) whether those who claimed to represent a group’s interest held
full control of the executive branch with no meaningful participation
64
We thus include ethnolinguistic, ethnosomatic (or “racial”), and ethnoreligious groups in our
definition of ethnicity, but we exclude tribes and clans that define community in genealogical terms, as
well as regions that do not define commonality on the basis of shared ancestry.
65
It should be noted that political relevance does not equal violent conflict. Endogeneity is not a
problem because only a tiny fraction of years with mobilized or discriminated groups were characterized
by conflict (0.45 percent); cf. Posner 2004.
66
We did not distinguish between degrees of representativity of political actors who claim to speak
in the name of an ethnic group; nor did we code the heterogeneity of political positions voiced by
leaders claiming to represent the same community. Cf. Brubaker 2004.
67
This choice was made in order to limit the coding effort. Legislative politics may also influence
the likelihood of conflict through ethnic political parties and specific types of electoral institutions;
see, for example, Cohen 1997; Brancati 2006; and Birnir 2007.
100	 world politics
by members of any other group, (2) whether they divided power with
members of other groups in a power-sharing regime, or (3) whether
they were excluded altogether from decision-making authority within
the halls of central state power. Within each of these categories, coders
were asked to differentiate between further subtypes.
absolute power
In this case, the political elites that claim to represent an ethnic group
do not significantly share power with other political leaders. There are
two possibilities:
—Monopoly. Elite members hold monopoly power in the executive to
the exclusion of members other ethnic groups.
—Dominance. Elite members of the group hold dominant power in
the executive but there is some limited inclusion of “token” members of
other groups.
power-sharing regimes
By power sharing, we mean any formal or informal arrangement that
divides executive power among leaders who claim to represent particular
ethnic groups. Depending on the relative importance of the positions
controlled by group representatives, there are two possibilities:
—Senior Partner. Representatives participate as senior in a formal or
informal power-sharing arrangement.
—Junior Partner. Representatives participate as junior partners in
government.
exclusion from central power
Finally, if political leaders who claim to represent a particular ethnic
category are excluded from participation in central government, we distinguish
between four possibilities:
—Regional Autonomy. Elite members of the group have no central
power but some influence at the substate level, that is, one level below
the central government. This may be the substate, the provincial, or the
district (though not the local) level, depending on the vertical organization
of the state.
—Separatist Autonomy. A related case is when local governments
controlled by representatives of an ethnic category have declared their
territory to be independent from the central government. This category
differs fundamentally from “regional autonomy” in that group representatives
have often excluded themselves from central state power.
—Powerless. Elite representatives hold no political power at either
the national or the regional level without being explicitly discriminated
against.
	 why do ethnic groups rebel?	 101
—Discrimination. Group members are subjected to active, intentional,
and targeted discrimination, with the intent of excluding them
from both regional and national power. Such active discrimination can
be either formal or informal.68
Conflict Coding
Our coding of conflict is based upon the ucdp/prio Armed Conflicts
Dataset, which defines internal conflict as any armed and organized
confrontation between government troops and rebel organizations or
between army factions that reaches an annual battle death threshold
of twenty-five.69
This definition excludes one-sided conflicts, such as
massacres and genocides, as well as communal riots, pogroms, and
other nonstate conflicts.70
For each conflict, we coded whether actors pursued ethnonationalist
aims and whether they pursued secessionist objectives:Ethnic/Nonethnic
Conflicts were distinguished by the aims of the armed organization on
the one hand and their recruitment and alliance structures on the other
hand. Ethnic aims include self-determination, more influence for one’s
group over government, autonomy, and language and other cultural
rights. Regarding recruitment and alliance structures, we defined ethnic
wars as fought by armed organizations who predominantly recruit
fighters among their leaders’ own ethnic group and who forge alliances
on the basis of ethnic affiliation. For a conflict to be classified as ethnic,
armed organizations must both explicitly pursue ethnonationalist aims
and recruit fighters and forge alliances on the basis of ethnic affiliations.
All ethnic conflicts were then linked to the politically relevant ethnic
group in the name of which an armed organization instigated the
conflict. We looked at the aims and recruitment patterns of each armed
organization separately. In some complex cases (as in Afghanistan,
Burma, Chad, Uganda, Angola, and Zaire), we disaggregated a conflict
into several war fronts when different ethnic claims were made on the
nongovernmental side. This was also necessary when the constellation
of rebel organizations and the ethnic communities they represented
changed dramatically over time.71
Our data set includes 124 ethnic conflicts fought between 1946 and
2005. Given that some of these ethnic conflicts were fought by more
68
We did not include exclusion of noncitizens from power, as long as these noncitizens hold passports
of other states and are effectively able to return to their country of origin.
69
Gleditsch et al. 2002.
70
We drew primarily on version 3-2005b of the acd data set, which provides two levels of conflict
identification, a more general war id number, and a disaggregated sub-id.
71
To avoid endogeneity, the power-access coding always reflects the situation before conflict onset
in years that feature both conflict outbreak and power-status changes.
102	 world politics
than one group, we identify 146 group-level instances of ethnic conflict
onset. Among these, 27 were fought by groups in power, 87 by powerless
or discriminated groups, and 32 by autonomous groups. Half of
these conflicts reached the standard threshold of civil war (that is, more
than one thousand battle deaths in a year). In addition, the data set includes
a distinction between secessionist and nonsecessionist conflict,
to which we will return in our analysis of elite infighting.
Variable Definitions and Data Sources
Drawing on the epr data set, we test all three hypotheses by measuring
power access (H1a,b,c), mobilizational capacity (H2), and prior conflict
(H3) for each politically relevant ethnic group in each year from
1946 to 2005.
The Excluded variable is a dichotomous indicator that indicates
whether representatives of an ethnic category are discriminated against,
are powerless, or have regional or separatist autonomy (H1a). To present
a more nuanced picture, we also use dummy variables referring to
specific epr categories, that is, Junior Partner, Regional Autonomy, Separatist
Autonomy, Powerless, and Discriminated. A Downgraded dummy
variable identifies whether representatives of a group have experienced
a decrease in power status during the previous two years (H1b), based
on the epr power-status categories. In our analysis of elite infighting,
we add the Underrepresented variable, which indicates whether a junior
coalition partner represents a larger ethnic group than a senior partner
(H1c).
Furthermore, we evaluate group size with the logged variable Group
Size, which uses the epr estimate of demographic group size as a proportion
of the total population of the country’s politically relevant ethnic
groups (H2). In order to test the influence of prior conflict, we use
a variable Past Conflict that counts the number of conflicts that have
already been fought in the name of the same ethnic group since the
beginning of the sample period (H3). Finally, Peace Years counts the
number of peaceful years since the last conflict or the beginning of the
sample period.72
In order to control for country-level characteristics, we use two control
variables throughout the article. Based on extensive robustness
analysis, Hegre and Sambanis conclude that variables measuring wealth
and population size have a powerful effect on civil war onset.73
There-
72
Following Beck, Katz, and Tucker 1998, this indicator is tested together with three cubic
splines.
73
Hegre and Sambanis 2006.
	 why do ethnic groups rebel?	 103
fore we include measures of logged gdp per Capita and logged country
Population Size, drawn from the Penn World Tables and World Bank
sources. While several interpretations have been offered of the effects
of income, our theory of ethnic politics expects that richer countries
should have less conflict because they can afford to solve political conflicts
through redistribution.74
Although a country’s population size is
usually significant in country-level studies of civil war onset, it is less
clear what to expect here, especially since its robustness may simply
reflect the fact that, all else equal, conflicts are more likely to occur in
larger countries.
Data Analysis
We are now ready for a first evaluation of our three hypotheses relating
to power access, mobilization capacity, and past conflict. The current
section introduces statistical models that concern groups in all powerstatus
categories, as well as more specific analysis of insurgencies in the
name of excluded groups and infighting among power-sharing partners,
since some of our causal mechanisms relate to specific actor types
only.
Descriptive Analysis
Our data set is structured in a group-country-year format, with 29,519
unique observations, including all politically relevant ethnic groups.
Table 1 offers a first empirical assessment of the power-access hypothesis,
H1. As expected, the frequency of conflict increases roughly with
the degree of exclusion. Excluded groups are much more likely to experience
a rebellion in their name (0.66 percent) compared with groups
in power (0.23 percent). A χ2
-test confirms that this relationship is
significant at p = 0.0001. With the exception of the category Regional
Autonomy, which exhibits relatively less conflict than the power-sharing
categories, the table also reveals that the conflict frequency falls steadily
with increasing access from executive power. Because Separatist Autonomy
differs from the other excluded categories in that the group has
chosen to “exclude itself” from central state power, we list these at the
end of the exclusion category.75
74
Wimmer, Cederman, and Min 2009; cf. Sambanis 2004.
75
Because all separatist groups get involved in violence at some point, the coding of regional
versus separatist autonomy was verified case by case. Removing the separatist cases from the data set
makes no difference to our other findings.
104	 world politics
Going beyond descriptive statistics, we proceed with regression
analysis to test the main hypotheses. Our standard framework consists
of logit models with ethnic conflict onset as a dichotomous dependent
variable. Because the causal logic during wars can be expected to differ
from what it is in peacetime, we drop observations in which a war
is currently ongoing and study only the risk of conflict onset during
peacetime years. To account for the nonindependence of observations
within countries, we present robust standard errors, clustered on the
country unit.
Analysis of All Groups
Our main results are presented in Table 2. Using group-years as the
unit of analysis, our analysis begins by including all politically relevant
groups, except those that enjoyed Monopoly or Dominant status and
by definition could not be associated with conflict against the state (see
the first two columns). In order to operationalize the basic power access
hypothesis, H1a, model 1 uses the Excluded dummy variable to
identify those groups that have no access to central power. Based on
this dichotomization, it is clear that rebellions in the name of excluded
groups are much more likely than conflict in the name of included
groups, thus supporting H1a. The Downgraded measure, which tests
hypothesis H1b, is also strongly confirmed in this analysis. In this case,
the corresponding binary variable has an even stronger effect than the
static exclusion measure. Turning to the test of the mobilization capacity
hypothesis, H2, we also find solid evidence that, ceteris paribus,
Table 1
Ethnic Groups’ Conflict Propensity by Category of Power Access
Group Years Years of Ethnic-Conflict Onset
Included Groups 11,622 27 0.23%
 Monopolya
1,672 n/a n/a
  Dominanta
1,709 n/a n/a
Senior Partner 2,884 9 0.31%
Junior Partner 5,357 18 0.34%
Excluded Groups 17,897 119 0.66%
Regional Autonomy 5,433 13 0.24%
Powerless 7,482 45 0.60%
Discriminated 4,506 42 0.93%
Separatist Autonomy 476 19 3.99%
Total 29,519 146 0.46%
a
No rebellion, by definition.
	 why do ethnic groups rebel?	 105
larger groups are more likely to experience a rebellion fought in their
name. The analysis also confirms our path-dependency hypothesis,
H3. As expected, the impact of Past Conflict is both large and significant.
We also observe that the hazard rate declines during peaceful
periods.
Table 2
Explaining Group-Level Ethnonationalist Conflict
  (1) (2) (3) (4)
All Groups All Groups
Excluded
Groups Only
Elite
Infighting Only
Excluded 1.2121**
(0.2668)
Junior Partner 0.3224 –0.1693
(0.4056) (0.3926)
Only Local Power 0.8321
(0.5392)
Powerless 1.2041** 0.3866
(0.4552) (0.3751)
Discriminated 1.6719** 0.8603*
(0.4303) (0.4237)
Separatist 3.2514** 2.4769**
(0.6793) (0.8423)
Downgraded 1.6419** 1.7590** 1.8211** 1.7526*
(0.3933) (0.3920) (0.4387) (0.8353)
Log (Group Size) 0.2902** 0.3015** 0.3559** –0.1834
(0.0783) (0.0829) (0.0926) (0.2468)
Underrepresented 0.1397**
(0.0321)
Past Conflict 0.8545** 0.6309** 0.6148 0.6477
(0.1727) (0.2387) (0.3245) (0.5621)
Log (gdp per Capita), Lagged –0.3675** –0.3346** –0.3396** –0.2181
(0.1023) (0.0920) (0.1175) (0.1545)
Log (Population), Lagged –0.0046 0.0220 0.0766 –0.1138
(0.0958) (0.0918) (0.1141) (0.1158)
Years since Last Conflict
Onset
–0.1542*
(0.0725)
–0.1398
(0.0748)
–0.1592
(0.0840)
–0.0112
(0.1797)
Peace Years Splines yes yes yes yes
Constant –3.3042** –4.1166** –3.8561* –2.7547
(1.2077) (1.2531) (1.6007) (2.0233)
Observations 24445 24445 16498 7947
Robust standard errors clustered on country in parentheses; ** p<0.01, * p<0.05
106	 world politics
Regarding the control variables, higher levels of gdp per Capita have
a strong, negative influence on conflict onset. However, Population,
which usually has strong effects in country-level analyses, fails to exhibit
any significant or substantive effect.
As a way to compare the marginal effect of our key variables, we
plot the difference that our main variables have on predicted conflict
probabilities, holding other variables at their mean or modal values.
Figure 2 displays the conflict propensities associated with included and
excluded groups as a function of gdp per Capita. For both categories we
observe an income effect, with higher risks of violent rebellion in countries
with lower incomes, in line with previous findings. However, it is
notable that excluded groups across all income levels are three times
more likely to initiate conflict against the state as compared with included
groups that enjoy representation at the center. The substantive
effects of exclusion are especially pronounced in poor countries, where
the already high baseline risk of violence is compounded for excluded
groups. Our explanation is that in poor countries governments lack the
resources necessary to co-opt leaders of protest movements that fuel
resentment because of their exclusion from central government. Since
the central state is all-decisive in poor countries, groups that lack representation
and are marginalized in the distribution of state resources,
government jobs, and public goods may have greater motives to take
up arms.76
While not plotted in the figure, our other key variables also
have very large effects on the risk of conflict: a downgraded power status
and prior conflict history increases the likelihood of a conflict by
5.3 and 2.4 times respectively in comparison with included groups.
In a further effort to investigate H1a, model 2 introduces the more
fine-grained power-access categories. It shows that the conflict-inducing
impact of exclusion from state power represents more of a continuum
than a dichotomy. According to our theory, conflict proneness should
decrease gradually as the access of ethnic leaders to state power improves.
Indeed, this is what we find. The coefficients range from a low and insignificant
0.32 for junior partners up to a highly significant 1.67 for
discriminated groups and 3.25 for separatists. The latter category is especially
conflict prone, not merely because of high motivation but also
because separatism challenges the state to the highest possible degree,
given that modern states are built on the principle of territorial sovereignty.
Again, hypotheses H1b, H2, and H3 are powerfully confirmed
and the effect of the control variables remains virtually unchanged.
76
Cf. Wimmer 2002, chap. 3.
	 why do ethnic groups rebel?	 107
So far, we have considered the relationship of all groups with the
government, but there are reasons to believe that ethnic elites behave
differently, depending on whether or not they are already represented
in the central government. Therefore, we now consider excluded and
included groups separately to identify the relation-specific effects that
are partially concealed in the general analysis.
Analysis of Excluded Groups
Once we restrict our analysis to rebellions in the name of excluded
groups, we are able to specify a clear-cut actor constellation based on a
center-periphery logic. Model 3 inTable 2 presents the empirical results
of this analysis. Given the focus on excluded groups, we find smaller
differences among the power-status categories. However, both Discrimination
and Separatist Autonomy can be statistically distinguished from
the baseline category Regional Autonomy, thus supporting hypothesis
H1a. A recent Downgraded power status among these excluded groups
increases conflict propensity significantly, in line with H1b. Compared
with the general analysis, we find even more unambiguous evidence
that large groups are more likely to experience conflict (H2). The influence
of Prior Conflict remains significant (H3). Finally, the control
variables behave similarly to the results noted in models 1 and 2.
Figure 2
The Effect of GDP per Capita on Ethnic Groups’
Conflict Propensity
a
a
Shaded areas represent 95 percent confidence intervals.
5 10 20 30
GDP per Capita in Lagged 2000 Dollars (’000s)
Log Scale
ProbabilityofEthnicConflictOnset
	 .01
	 .008
	 .006
	 .004
	 .002
0
Excluded
Groups
Included
Groups
108	 world politics
Figure 3 offers a graphic depiction of how mobilizational capacity
influences the risk of conflict for excluded groups. The plots, based on
model 1, show that larger groups are more likely to initiate rebellion
and that the risks are substantially higher for excluded groups that have
had a prior conflict or have recently been downgraded in their power
status. The effects of being downgraded are especially pronounced,
though the effects are imprecisely estimated for larger group sizes.
Analysis of Included Groups
Finally, we shift our attention from excluded to included groups. Here
we consider only cases of power sharing, that is, those that involve senior
and junior partners of governing coalitions, because monopoly and
dominant groups are directly identified with the government and thus
by definition cannot challenge it. Our goal therefore is to understand
why some power-sharing partners initiate conflict against a government
in which they are represented. Following our hypothesis H1b
and H1c, we expect that underrepresented power-sharing partners as
well as those who have experienced a downgrading in their power position
will be more likely to take up arms against other members of the
governing coalition.
Before discussing results, a cautionary remark is in order. As indicated
in Table 1, there are very few cases of conflict instigated in the
name of groups in power, only 27 onsets in 8,241 group-years. This
means that the results of the statistical evaluation are likely to be sensitive
to the coding of this small number of conflict events. With this
caveat in mind, we turn to model 4 in Table 2. The results indicate that
conflicts in the name of junior partner groups are not more frequent
than those fought in the name of senior partners (H1a). However, the
regression analysis suggests that a downgraded power status has an impact
on conflict behavior, thus confirming H1b. We also get strong
results for hypothesis H1c that relate to resentment caused by underrepresentation
within governing coalitions. Group size has no influence
on the dependent variable in contrast to the effect for excluded groups
in models 1–3, suggesting that the mobilizational capacity argument
does not hold for groups that already have access to the organizational
apparatus of government. Furthermore, the path-dependency effect
(H3) cannot be confirmed for the current subsample. Finally, we fail to
find any significant effect for any of the control variables.
Given the rarity of conflict among groups in power, it is appropriate
to examine individual cases. In particular, it should be noted that the
events influencing the coefficients on downgraded power status (H1b)
	 why do ethnic groups rebel?	 109
and underrepresentation (H1c) refer to cases of nonsecessionist conflict
only. The ten secessionist conflicts are not well predicted by our
model. Typically, these secessionist conflicts involved radicalized splinter
groups that were dissatisfied with the power-sharing deal offered by
the government and therefore resorted to violence; among these groups
were the United Liberation Front of Assam,77
the Sikh Sangat organi-
zation,78
or the Albanian uck/kla rebels in Macedonia.79
These cases
illustrate a general limitation of our research design. Different organizations
claim to represent the interests of an ethnic group, and if one
organization participates in government, this does not prevent others
from taking an oppositional stand to mount a rebellion in the name of
“true” group interests. From a theoretical perspective, this outcome is
not surprising because it is, after all, parties and their representatives
that participate in government rather than ethnic groups.
77
Baruah 1994.
78
Chima 1994.
79
Lund 2005.
Figure 3
The Effect of Excluded Ethnic Groups’ Size on Their
Conflict Propensity a
a
Shaded areas represent 95 percent confidence intervals.
0 20 40 60 80 100
Size of Group (%)
0 20 40 60 80 100
Size of Group (%)
ProbabilityofEthnicConflictOnset
ProbabilityofEthnicConflictOnset
Excluded and
Prior Conflict
Excluded and
Downgraded
Excluded Excluded
.15
.1
.05
0
.15
.1
.05
0
110	 world politics
In other cases, the fear of future domination by coalition partners
triggered secessionist bids that turned violent, as in former Yugoslavia
and Bosnia and Herzegovina.80
In principle, the fear of domination
by ethnic others in the future—rather than its actual experience—is
consistent with our theory, although it cannot be captured with existing
indicators without addressing severe endogeneity problems. In a
complementary analysis on the country level, we show that such fears
and the corresponding secessionist conflicts are more likely in states
with a short history of direct rule by the political center.81
The seventeen nonsecessionist instances of elite infighting are
somewhat more numerous but still warrant case-by-case analysis. How
supportive is the evidence in favor of the main hypotheses relating to
a fall in power status (H1b)? We find only two instances of conflicts
within two years of a downgraded power status. After a long period of
dominance by President Eyadéma, who favored his own Kabré group
in Togo’s army and administration, he came under pressure to democratize
and had to agree to power sharing in 1991. Despite their status
as senior partner, he and his entourage in the army became unhappy
with the power-sharing arrangement and launched a coup later that
same year. Meanwhile, the 1984 coup in Cameroon can be seen as a
reaction to the shift of power from the Fulani-dominated regime of
President Ahidjo to the government of Biya, a Beti politician.82
Regarding conflicts driven by underrepresentation within the governing
coalition (H1c),the data point us to three potential cases,namely,
the Sara 1991 rebellion in Chad, the Hutu 1991 uprising in Burundi,
and Shiite participation in the Lebanese civil war in 1975. Rather than
being directly motivated by underrepresentation, the first of these three
conflicts appears to be more related to long-term frustrations felt by
the Sara, who had dominated the state’s early history from independence
in 1960 through 1978,83
and thus is in line with the downgrading
hypothesis (H1b).84
In the other two events, however, the conflict
is indeed related to the underrepresentation mechanism. In 1991 rebels
who claimed to represent the Hutu majority launched an armed
struggle against the Tutsi-dominated power-sharing arrangement.85
In Lebanon, the Shiite political elite were increasingly resentful that
the Maronites were overrepresented in the traditional consociational
80
Kalyvas and Sambanis 2005.
81
Wimmer, Cederman, and Min 2009.
82
Mehler 1993.
83
Decalo 1980.
84
The Downgraded variable does not pick up this case because it is based on a two-year window.
85
Lemarchand 1994.
	 why do ethnic groups rebel?	 111
arrangement guaranteeing them the presidency, especially because the
share of the Shiite population had increased in prior decades.86
According to our data, the influence of past civil conflict plays a role
in some cases of elite infighting. There are obvious instances, such as
the rebellions in Chad that happened against the backdrop of a longlasting
civil war. Nor can there be any doubt that the Hutu defection
from the Tutsi-dominated governing coalition in 1991 was influenced
by memories of past massacres. A similar argument can be made with
respect to the failure of the power-sharing arrangement in Lebanon
in 1975, because Shiite and Sunni leaders had already mobilized their
constituencies in the conflict against the Maronite elite in 1958. In all
three cases the past conflict variable thus picks up meaningful historical
processes.
To sum up, the historical evidence in support of model 4 is strong,
although the results are driven exclusively by nonsecessionist conflicts.
The hypotheses relating to the downgrading and underrepresentation
mechanisms (H1b,c) are tentatively supported by a case-by-case inspection
of the nonsecessionist cases of elite infighting.
Sensitivity Analysis
Table 3 presents robustness checks of our primary findings based on
model 1 of Table 2. We start by considering the robustness of model
1 to the inclusion of dummy variables that divide countries into six
world regions, using the Western states as the reference category (see
model 5). The regional coefficients for North Africa and the Middle
East, sub-Saharan Africa, and Asia are positive and significant at the
p = 0.10 level. More important, we do not detect any major differences
in the effect of our main explanatory variables: exclusion, downgraded,
group size, and conflict history remain powerful even after controlling
for differences across world regions (see hypotheses 1–3). In contrast,
the gdp variable loses some of its impact and is now significant at the
p = 0.053 level.
To further control for local-level factors, we introduce a model with
country fixed effects (see model 6), allowing the constant to vary independently
for each state in which an ethnic conflict has occurred.
This design captures time-invariant country-specific features that are
not captured by our independent variables. The results confirm the robustness
of our main hypotheses, except H3. Whereas the status vari-
86
Makdisi and Sadaka 2005, 61.
112	 world politics
ables and size indicator influence the dependent variable, the impact of
past conflict cannot be statistically confirmed due to the low variance
among the country-observations that experienced conflict. Moreover,
in what is a clear shortcoming, fixed-effects estimation excludes from
the analysis all countries that never experience conflict.87
87
Beck and Katz 2001.
Table 3
Sensitivity Analysis
  (1) (2) (3)
  Region Controls
Country Fixed
Effects
High-Intensity
Onsets Only
Excluded 1.2534** 1.3262** 1.3904**
(0.2652) (0.2960) (0.3780)
Downgraded 1.5911** 1.5821** 1.8696**
(0.4001) (0.3117) (0.4909)
Log (Group Size) 0.2906** 0.2595** 0.3919**
(0.0893) (0.0785) (0.0970)
Past Conflict 0.8388** 0.1138 0.6797*
(0.1662) (0.2378) (0.2692)
Log (gdp per Capita), Lagged –0.2232 0.0228 –0.5069**
(0.1155) (0.2400) (0.1243)
Log (Population), Lagged –0.0335 0.3558 –0.0172
(0.0955) (0.3573) (0.0995)
Years since Last Conflict Onset –0.1582* –0.1432* –0.1888
(0.0731) (0.0624) (0.0969)
North Africa and Middle East 1.2148
(0.6450)
Latin America 0.3654
(0.6992)
Sub–Saharan Africa 1.1917
(0.6549)
Eastern Europe 0.9736
(0.6438)
Asia 1.3391
(0.6914)
Peace Years Splines yes yes yes
Country Fixed Effects no yes no
Decade Fixed Effects no no no
Constant –5.2080** –25.1064 –2.8117*
(1.5947) (0.0000) (1.3791)
Observations 24445 15734 24445
Robust standard errors clustered on country in parentheses; ** p<0.01, * p<0.05
	 why do ethnic groups rebel?	 113
In order to test whether our models are sensitive to the intensity of
conflicts, we replace the coding of the dependent variable based on the
twenty-five battle deaths threshold with one that conforms to the one
thousand battle deaths threshold conventionally used to define civil
wars, while otherwise keeping model 1 unchanged. Model 7 shows that
our results hold in predicting the onset of this smaller set of seventysix
high-intensity conflicts. The variables associated with the main hypotheses
retain their effect, as does the gdp per Capita variable.
Our results are also robust to various changes in the operationalization
of our key variables (results not shown). For example, varying the
two-year window used to calculate the Downgraded variable to one-,
three-, four-, or five-year windows makes no substantive difference for
our results. As discussed in the context of elite infighting, the coding of
our dependent variable allows us to check whether our findings differ
for secessionist and nonsecessionist conflicts. Whereas this is the case
for elite infighting, all the main hypotheses hold robustly for secessionist
and nonsecessionist conflicts fought in the name of excluded
groups.
Finally, our results are also robust to controls for regime type and
time trends (results not shown). Democracies are no less likely than
autocracies to experience ethnic conflict, contrary to the expectations
of democratic civil peace theory. The relationship between anocracy
and ethnic conflict is less clear and depends on whether anocracy is
defined using the standard Polity scores or based on alternative coding
schemes.88
Given the long time frame of our study, we examined the
potential effects of temporal and geopolitical trends by adding a cold
war dummy, a calendar year variable, and decade fixed effects. None of
these variables reduces the statistical or substantive significance of our
main findings.
Conclusion
In this article we have investigated the influence of ethnic power inequality
on civil war and found that exclusion and competition along
ethnic lines are strongly associated with internal conflict. Rather than
continuing to debate whether or not ethnic diversity or polarization
breeds conflict, quantitative researchers might want to try identifying
88
Vreeland 2008 reports that Polity-based codings of regime type are endogenous to civil war and
offers corrected regime variables. In our model, anocracies classified using standard Polity scores are
significantly more conflict prone than other regime types, but the significance disappears when using
the Vreeland coding.
114	 world politics
those ethnic constellations of power that are particularly war prone.
The article contributes to this endeavor by showing that once ethnic
politics is properly conceptualized and measured, it is indeed possible
to identify the conditions under which ethnonationalist civil wars are
more likely to break out. Large ethnic groups that are excluded from
state power or underrepresented in government are much more likely to
challenge the regime’s insiders through violent means. Also in keeping
with our theoretical expectations, a loss of power in recent history or
previous conflict further increases the likelihood of armed conflict.
Readers who are familiar with the qualitative scholarship on ethnic
conflict may not be surprised that representatives of ethnic groups
with less access to power are more likely to challenge the government.
However, this finding is far from trivial. Indeed, the most widely cited
articles in the recent civil war literature contend that ethnic diversity
is unrelated to the likelihood of conflict and conclude that ethnicity
thus does not matter for understanding war and peace.89
Introducing
a stark opposition between matters of motivation and legitimacy
(“grievances”), on the one hand, and factors related to the military and
economic feasibility of insurgencies, on the other hand, these scholars
argue that ethnic grievances are too ubiquitous to explain the rare event
of civil war. Without denying the relevance of feasibility mechanisms,
our findings show that ethnicity should not be discounted as an explanatory
factor in the study of civil wars. We demonstrate empirically
how the logics of contention and mobilization lead ethnically defined
actors who are excluded from state power into armed conflict. Roughly
half of the conflicts fought since the Second World War can be linked
to this dynamic of ethnopolitical struggle for state power.
How do our findings relate to existing research on ethnic conflict? As
suggested above, past quantitative tests of how political exclusion and
discrimination relate to the propensity of violence have produced conflicting
results. We argue that sample selection and data quality problems
are responsible for this inconclusiveness and believe that our new
data set on ethnic power relations (epr) represents significant progress
by enabling more precise coding of political representation in state
governments, including that of dominant majorities or power-sharing
partners. Since the epr data set is time sensitive, it also permits tracing
changes in power access and determining whether such changes affect
the conflict dynamic over time. On the basis of this improved data set,
we are able to establish an unequivocal relationship between the degree
of access to state power and the likelihood of armed rebellion.
89
Fearon and Laitin 2003; Collier and Hoeffler 2004.
	 why do ethnic groups rebel?	 115
While improving on previous studies of ethnic conflict, our research
also has limitations that we intend to address in the future. First, despite
our efforts to capture time-sensitive mechanisms, our focus has been
mostly on structural conditions rather than on dynamic and strategic
processes of contention and mobilization. Second, we have not tried to
explain changes in power structures and the political relevance of different
ethnic categories but have treated these instead as exogenously given.
Third, due to limited data availability, we have refrained from analyzing
the effects of cross-border relationships between ethnic kin, settlement
patterns, and institutional factors relating to party systems and electoral
institutions. Indeed, this study is written as an invitation to fill these
gaps by collecting more precise data on ethnic organizations, their mobilizational
capacities and power resources, and their interactions with
each other as well as with nonethnic organizations and governmental
actors, inside or across state borders. Ideally, such data would make it
possible to develop a fully processual theory that will allow us to analyze
ethnic conflicts as a specific conjuncture in the ongoing struggle
over state power between dynamically evolving sets of actors.
References
Anderson, Benedict. 1991. Imagined Communities: Reflections on the Origin and
Spread of Nationalism. London: Verso.
Baruah, Sanjib. 1994. “The State and Separatist Militancy in Assam: Winning a
Battle and Losing the War?” Asian Survey 34, no. 10: 863–77.
Beck, Nathaniel, and Jonathan N. Katz. 2001. “Throwing Out the Baby with the
Bath Water: A Comment on Green, Kim, and Yoon.” International Organization
55, no. 2 (Spring): 487–95.
Beck, Nathaniel, Jonathan N. Katz, and Richard Tucker. 1998. “Taking Time Seriously:
Time-Series–Cross-Section Analysis with a Binary Dependent Variable.”
American Journal of Political Science 42, no. 4: 1260–88.
Beissinger, Mark R. 2002. Nationalist Mobilization and the Collapse of the Soviet
Union. Cambridge: Cambridge University Press.
Birnir, Jóhanna Kristín. 2007. Ethnicity and Electoral Politics. Cambridge: Cambridge
University Press.
Blimes, Rand. 2006. “The Indirect Effect of Ethnic Heterogeneity on the Likelihood
of Civil War Onset.” Journal of Conflict Resolution 50, no. 4: 536–47.
Brancati, Dawn. 2006. “Decentralization: Fueling the Fire or Dampening the
Flames of Ethnic Conflict and Secessionism.” International Organization 60,
no. 3: 651–85.
Brass, Paul. 1991. Ethnicity and Nationalism: Theory and Comparison. Newbury
Park: Sage.
Breuilly, John. 1994. Nationalism and the State, 2nd ed. Chicago: University of
Chicago Press.
116	 world politics
Brubaker, Rogers. 1996. Nationalism Reframed: Nationhood and the National Question
in the New Europe. Cambridge: Cambridge University Press.
———. 2004. Ethnicity without Groups. Cambridge: Harvard University Press.
Brubaker, Rogers, and David D. Laitin. 1998. “Ethnic and Nationalist Violence.”
Annual Review of Sociology 24: 423–52.
Buhaug, Halvard, Lars-Erik Cederman, and Jan Ketil Rød. 2008. “Disaggregating
Ethno-Nationalist Civil Wars: A Dyadic Test of Exclusion Theory.” International
Organization 62, no. 3 (July): 531–51.
Cederman, Lars-Erik. 1997. Emergent Actors in World Politics: How States and Nations
Develop and Dissolve. Princeton: Princeton University Press.
Cederman, Lars-Erik, Halvard Buhaug, and Jan Ketil Rød. 2009. “Ethno-Nationalist
Dyads and Civil War: A gis-Based Analysis.” Journal of Conflict Resolution
53, no. 4: 496–525.
Cederman, Lars-Erik, and Luc Girardin. 2007. “Beyond Fractionalization: Mapping
Ethnicity onto Nationalist Insurgencies.” American Political Science Review
101, no. 1: 173–85.
Cederman, Lars-Erik, Luc Girardin, and Kristian Skrede Gleditsch. 2009. “Ethnonationalist
Triads: Assessing the Influence of Kin Groups on Civil Wars.”
World Politics 61, no. 3 (July): 403–37.
Cetinyan, Rupen. 2002. “Ethnic Bargaining in the Shadow of Third-Party Intervention.”
International Organization 56, no. 3: 645–77.
Chima, Jugdep S. 1994. “The Punjab Crisis: Governmental Centralization and
Akali-Center Relations.” Asian Survey 34, no. 10: 847–62.
Collier, Paul, and Anke Hoeffler. 2004. “Greed and Grievance in Civil Wars.”
Oxford Economic Papers 56, no. 4: 563–95.
Cohen, Frank S. 1997. “Proportional versus Majoritarian Ethnic Conflict Management
in Democracies.” Comparative Political Studies 30, no. 5: 607–30.
Cunningham, David E., Kristian Skrede Gleditsch, and Idean Salehyan. 2009. “It
Takes Two: A Dyadic Analysis of Civil War Duration and Outcome.” Journal
of Conflict Resolution 53, no. 4: 570–97.
Decalo, Samuel. 1980. “Chad: The Roots of Center-Periphery Strife.” African Affairs
79, no. 317: 490–509.
De Figueiredo, Rui, and Barry Weingast. 1999. “The Rationality of Fear: Political
Opportunism and Ethnic Conflict.” In Barbara Walter and Jack Snyder, eds.,
Civil Wars, Insecurity, and Intervention. New York: Columbia University Press.
DeNardo, James. 1985. Power in Numbers: The Political Strategy of Protest and Rebellion.
Princeton: Princeton University Press.
Fearon, James D., and David D. Laitin. 2003. “Ethnicity, Insurgency, and Civil
War.” American Political Science Review 97, no. 1: 75–90.
Fox, Jonathan. 2000. “The Effects of Religious Discrimination on Ethno-Religious
Protest and Rebellion.” Journal of Conflict Studies 20, no. 1.
Gellner, Ernest. 1983. Nations and Nationalism. Ithaca, N.Y.: Cornell University
Press.
Gleditsch, Nils Petter, Peter Wallensteen, Mikael Eriksson, Margareta Sollenberg,
and Håvard Strand. 2002. “Armed Conflict 1946–2001: A New Dataset.”
Journal of Peace Research 39, no. 5: 615–37.
Gurr, Ted Robert. 1993a. Minorities at Risk: A Global View of Ethnopolitical Conflicts.
Washington, D.C.: United States Institute of Peace Press.
	 why do ethnic groups rebel?	 117
———. 1993b. “Why Minorities Rebel: A Global Analysis of Communal Mobilization
and Conflict since 1945.” International Political Science Review 14, no.
2: 161–201
———. 2000. Peoples versus States: Minorities at Risk in the New Century. Washington,
D.C.: United States Institute of Peace Press.
Hardin, Russell. 1995. One for All: The Logic of Group Conflict. Princeton: Princeton
University Press.
Hechter, Michael. 2001. Containing Nationalism. Oxford: Oxford University Press.
Hechter, Michael, and Dina Okamoto. 2001. “Political Consequences of Minority
Group Formation.” Annual Review of Political Science 4: 189–215.
Hegre, Håvard, Tanja Ellingsen, Scott Gates, and Nils Petter Gleditsch. 2001.
“Toward a Democratic Civil Peace? Democracy, Political Change, and Civil
War, 1816–1992.” American Political Science Review 95, no. 1: 33–48.
Hegre, Håvard, and Nicholas Sambanis. 2006. “Sensitivity Analysis of Empirical
Results on Civil War Onset.” Journal of Conflict Resolution 50, no. 4: 508–35.
Horowitz, Donald L. 1985. Ethnic Groups in Conflict. Berkeley: University of
California Press.
Jackman, Robert W. 1993. Power without Force. Ann Arbor: University of Michigan
Press.
Jenne, Erin K. 2007. Ethnic Bargaining: The Paradox of Minority Empowerment.
Ithaca, N.Y.: Cornell University Press.
Kalyvas, Stathis N. 2006. The Logic of Violence in Civil War. Cambridge: Cambridge
University Press.
———. 2007. “Civil Wars.” In Carles Boix and Susan Carol Stokes, eds., The
Oxford Handbook of Comparative Politics. Oxford: Oxford University Press.
Kalyvas, Stathis N., and Matthew Adam Kocher. 2007. “How ‘Free’ Is Free Riding
in Civil Wars? Violence, Insurgency, and the Collective Action Problem.”
World Politics 59, no. 2 (January): 177–216.
Kalyvas, Stathis N., and Nicholas Sambanis. 2005. “Bosnia’s Civil War: Origins
and Violence Dynamics.” In Paul Collier and Nicholas Sambanis, eds. Understanding
Civil War: Evidence and Analysis, vol. 2. Washington, D.C.: World
Bank.
Kedourie, Elie. 1960. Nationalism. New York: Praeger.
Kemper, Theodore D. 1978. A Social Interaction Theory of Emotions. New York:
John Wiley.
Laitin, David D. 1995. “National Revivals and Violence.” Archives Européennes de
Sociologie 36, no. 1: 3–43.
Lake, David A., and Donald Rothchild, eds. 1998. The International Spread of
Ethnic Conflict: Fear, Diffusion, and Escalation. Princeton: Princeton University
Press.
Lemarchand, René. 1994. Burundi: Ethnic Conflict and Genocide. Cambridge:
Cambridge University Press.
Lund, Michael S. 2005. “Greed and Grievance Diverted: Macedonia Avoided Civil
War, 1990–2001.” In Paul Collier and Nicholas Sambanis, eds., Understanding
Civil War: Evidence and Analysis, vol. 2. Washington, D.C.: World Bank.
McCarthy, John , and Mayer N. Zald. 1977. “Resource Mobilization and Social
Movements: A Partial Theory.” American Journal of Sociology 82, no. 6:
1212–41.
118	 world politics
Makdisi, Samir, and Richard Sadaka. 2005. “The Lebanese Civil War, 1975–90.”
In Paul Collier and Nicholas Sambanis, eds., Understanding Civil War: Evidence
and Analysis, vol. 2. Washington, D.C.: World Bank.
Mehler, Andreas. 1993. Kamerun in der Ära Biya: Bedingungen, erste Schritte und
Blockaden einer demokratischen Transition. Hamburg: Institut für Afrika-Kunde.
Migdal, Joel S. 1988. Strong Societies and Weak States. Princeton: Princeton University
Press.
Min, Brian, Lars-Erik Cederman, and Andreas Wimmer. 2009. “Ethnic Exclusion,
Economic Growth, and Civil War.” Manuscript, ucla.
Montalvo, José G., and Marta Reynal-Querol. 2005. “Ethnic Polarization, Potential
Conflict, and Civil Wars.” American Economic Review 95, no. 3: 796–816.
Olzak, Susan. 2006. The Global Dynamics of Racial and Ethnic Mobilization. Stanford,
Calif.: Stanford University press.
Petersen, Roger. 2002. Understanding Ethnic Violence: Fear, Hatred, and Resentment
in Twentieth-Century Eastern Europe. Cambridge: Cambridge University
Press.
Posen, Barry R. 1993. “The Security Dilemma and Ethnic Conflict.” In Michael
E. Brown, ed., Ethnic Conflict and International Security. Princeton: Princeton
University Press.
Posner, Daniel N. 2004. “Measuring Ethnic Fractionalization in Africa.” American
Journal of Political Science 48, no. 4: 849–63.
Regan, Patrick, and Daniel Norton. 2005. “Greed, Grievance, and Mobilization
in Civil Wars.” Journal of Conflict Resolution 49, no. 3: 319–36.
Rothschild, Joseph. 1981. Ethnopolitics: A Conceptual Framework. New York: Columbia
University Press.
Rydgren, Jens. 2007. “The Power of the Past: A Contribution to a Cognitive Sociology
of Ethnic Conflict.” Sociological Theory 25, no. 3: 225–44.
Saideman, Stephen M. 2002. “Discriminating in International Relations: Analyzing
External Support for Ethnic Groups.” Journal of Peace Research 39, no.
1: 27–50.
Saideman, Stephen M., and R. William Ayres. 2000. “Determining the Causes
of Irredentism: Logit Analyses of Minorities at Risk Data from the 1980s and
1990s.” Journal of Politics 62, no. 4: 1126–44.
Salehyan, Idean. 2009. Rebels without Borders: Transnational Insurgencies in World
Politics. Ithaca, N.Y.: Cornell University Press.
Sambanis, Nicholas. 2001. “Do Ethnic and Nonethnic Civil Wars Have the Same
Causes? A Theoretical and Empirical Inquiry.” Journal of Conflict Resolution
45, no. 3: 259–82.
———. 2004. “What Is Civil War? Conceptual and Empirical Complexities of
an Operational Definition.” Journal of Conflict Resolution 48, no. 3: 814–58.
Tarrow, Sidney. 2007. “Inside Insurgencies: Politics and Violence in an Age of
Civil War.” Perspectives on Politics 5, no. 3: 587–600.
Tilly, Charles. 1978. From Mobilization to Revolution. New York: McGraw-Hill.
Tilly, Charles, and Sidney Tarrow. 2006. Contentious Politics. Boulder, Colo.: Paradigm
Press.
Tir, Jaroslav, and Michael Jasinski. 2008. “Domestic-Level Diversionary Theory
of War: Targeting Ethnic Minorities.” Journal of Conflict Resolution 52, no. 5:
641–64.
	 why do ethnic groups rebel?	 119
Toft, Monica Duffy. 2003. The Geography of Ethnic Violence: Identity, Interests, and
the Indivisibility of Territory. Princeton: Princeton University Press.
Turner, Jonathan H., and Jan E. Stets. 2005. The Sociology of Emotions. Cambridge:
Cambridge University Press.
Vreeland, James Raymond. 2008. “The Effect of Political Regime on Civil War:
Unpacking Anocracy.” Journal of Conflict Resolution 52, no. 3: 401–25.
Waldmann, Peter. 2004. “The Asymmetry between the Dynamics of Violence
and the Dynamics of Peace: The Case of Civil Wars.” In Andreas Wimmer,
Richard Goldstone, Donald L. Horowitz, Ulrike Joras, and Conrad Schetter,
eds., Facing Ethnic Conflicts: Toward a New Realism. Lanham, Md.: Rowman
and Littlefield.
Walter, Barbara. 2004. “Does Conflict Beget Conflict? Explaining Recurring
Civil War.” Journal of Peace Research 41, no. 3: 371–88.
———. 2006a. “Information, Uncertainty, and the Decision to Secede.” International
Organization 60, no. 1:105–35.
———. 2006b. “Building Reputation: Why Governments Fight Some Separatists
but Not Others.” American Journal of Political Science 50, no. 2: 313–30.
Weber, Max. 1978. Economy and Society: An Outline of Interpretative Sociology.
Berkeley: University of California Press.
Weidmann, Nils B. 2009. “Geography as Motivation and Opportunity: Group
Concentration and Ethnic Conflict.” Journal of Conflict Resolution 53, no. 4:
526–43 .
Weiner, Myron. 1971. “The Macedonian Syndrome: An Historical Model of
International Relations and Political Development.” World Politics 23, no. 4
(October): 665–83.
Wimmer, Andreas. 2002. Nationalist Exclusion and Ethnic Conflict: Shadows of
Modernity. Cambridge: Cambridge University Press.
———. 2008. “The Making and Unmaking of Ethnic Boundaries. A Multi-level
Process Theory.” American Journal of Sociology 113, no. 4: 970–1022.
Wimmer, Andreas, Lars-Erik Cederman, and Brian Min. 2009. “Ethnic Politics
and Armed Conflict: A Configurational Analysis.” American Sociological Review
74, no. 2: 316–37.
Wimmer, Andreas, and Brian Min. 2006. “From Empire to Nation-State. Explaining
Wars in the Modern World, 1816–2001.” American Sociological Review
71, no. 6: 867–97.
Young, Crawford. 1976. The Politics of Cultural Pluralism. Madison: University of
Wisconsin Press.