Square Pegs in Round Holes: Inequalities, Grievances, and Civil War

Author(s): Halvard Buhaug, Lars-Erik Cederman and Kristian Skrede Gleditsch

Source: International Studies Quarterly , June 2014, Vol. 58, No. 2 (June 2014), pp. 418-
431
Published by: Wiley on behalf of The International Studies Association

Stable URL: https://www.jstor.org/stable/24017836

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

The International Studies Association and Wiley are collaborating with JSTOR to digitize,
preserve and extend access to International Studies Quarterly

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
International Studies Quarterly (2014) 58, 418-431

Square Pegs in Round Holes: Inequalities, Grievances, and

Civil War1
Halvard Buhaug

Peace Research Institute Oslo

Lars-Erik Cederman

ETH Zurich

AND

Rristian Skrede Gleditsch

University of Essex and PRIO

Much of the recent research on civil war treats explanations rooted in political and economic griev
able suspicion and claims that there is little empirical evidence of any relationship between ethnicit
political violence. We argue that common indicators used in previous research, such as the ethno-lin
tion (ELF) and the Gini coefficient for income dispersion, fail to capture fundamental aspects of po
economic inequality that can motivate conflict. Drawing on insights from group-level research, we dev
level indices that directly reflect inequalities among ethnic groups, including political discrimination a
tials along ethnic lines. Our analysis reveals that these theoretically informed country profiles are mu
of civil war onset than conventional inequality indicators, even when we control for a number of altern
tially related to grievances or opportunities for conflict.

Despite decades of scientific debate and numerous Weidmanncross and Gleditsch 2011; 0stby 2008).2 We argue
national studies, the link between inequality that andtheinternal
contradictory findings of the civil war literature
conflict remains persistently contested and frustratingly
to a large extent stem from the use of empirical measures
unclear. This assessment remains as valid today of inequality
as in theand grievances that lack strong theoretical
late 1980s, when Lichbach (1989) published a compre justification and of assumptions of causal homogeneity
hensive but inconclusive review of the literature. Whether that fail to distinguish between different types of internal
framed as a Marxist proposition (Boswell and Dixon
conflict. In trying to "push square pegs through round
holes," scholars of civil war have failed to adequately op
1993) or a psychologically inspired thesis along the lines
of "relative deprivation" (Gurr 1970), the idea that
erationalize both the independent and dependent vari
ables in the grievance-conflict nexus.
inequality triggers civil war and other forms of political
violence has drawn plenty of criticism (for example, TillyTo overcome these limitations, we propose replacing
1978; Skocpol 1979). More recently, many other promi conventional individualist measures of grievances with
nent studies of civil war fail to uncover any systematic new indicators that more clearly tap plausible political
relationship and reject the influence of inequality and economic inequalities at the group level, thus shift
together with other grievance-related explanations moreing the analytical focus from so-called vertical inequality
generally (for example, Collier and Hoeffler 2004; Fearonto horizontal inequality. This study is certainly not the
and Laitin 2003; though see also Boix 2008; Cederman, first to advocate that a group perspective can inform our
understanding of civil war (see, for example, Gurr 1993;
Regan and Norton 2005; Stewart 2008), and a number of
Halvard Buhaug is Research Professor at the Peace Research Institute Oslo
recent empirical studies have examined the effect of
and Professor of Political Science at the Norwegian University of Science and
Technology. His research covers geographic dimensions of armed conflict and
ethno-political and economic inequalities on civil war
security implications of climate change. Recent publications include articles inonset (for example, 0stby 2008; Hegre, 0stby and
Global Environmental Change, International Organization, International Security, Raleigh 2009; 0stby, Nordas and R0d 2009; Cederman
Journal of Conflict Resolution, Journal of Peace Research, and PNAS. et al. 2011). However, many of these studies are restricted
1 Authors' notes'. This research builds on several joint research projects in geographic scope, and most consider groups or other
between PRIO, ETH, and University of Essex, and we acknowledge financial subnational entities as the units of analysis. As such, they
support from the European Science Foundation, the Research Council of do not lend themselves easily to comparison with the
Norway, Swiss National Science Foundation, and the UK Economic and Social
Research Council. We are grateful for valuable input from colleagues and par
ticipants at conferences where earlier drafts have been presented. We thank 2 Many studies of inequality and conflict, such as Muller and Seligson
the editor and two anonymous reviewers for constructive comments. Replica(1987), examine forms of political violence much more encompassing than
tion data and supplementary information can be obtained from www.prio.no/civil war, including deaths due to events such as riots, and often consider
data. counts of the number of events rather than binary conflict measures.

Buhaug, Halvard, Lars-Erik Cederman, and Krisdan Skrede Gleditsch. (2013) Square Pegs in Round Holes: Inequalities, Grievances, and Civil War. International Studies Quarterly
doi: 10.1111/ isqu.12068
©2013 International Studies Association

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
 Halvard Buhaug, Lars-Erik Cederman and Kristian Skrede Gleditsch 419

country-oriented civil war literaturethe

atexplanatory
large nor allow
power of such grievance-based account
of political
comparison with the risk of civil war for countries violence (see also Muller 1972; Oberschall
without
ethnic cleavages. 1978; Skocpol 1979). In particular, these critics argu
Our study is the first to combine that an explicit
frustrations group
are simply too common to plausibly
focus in theory building and data generation with propo
account for outbreaks of violence, especially since protest
sitions and an empirical framework that canidentify
be easily thwarted
specificby powerful governments. Ther
country profiles associated with elevated conflict risk.
fore, explanations In
of collective political violence need t
so doing, we are able to capture thegauge political
nonstateunderpin
challengers' access to material and orga
nings of social grievances, whether nizational
relatedresources
to ethnicrather than interpreting their motiva
tions, which
exclusion from national politics or systematic this line of reasoning deems to be largely
differences
in economic opportunities and privileges between
irrelevant. ethnic
More recent quantitative research on civil war
groups. The subsequent statistical analysis
tends todraws on new
reach similar conclusions, although this liter
georeferenced economic and ethno-political
ture consists data that
mostly of cross-national comparative pane
also help remedy severe missing datastudies
problems
of civil character
wars involving the state as opposed to the
izing most earlier comparative research focusonon inequality
broader forms andof political instability or dynamics
of escalation
conflict. Crucially, the country-level approach in earlier
allows us tosociological research (cf. new stu
systematically compare our group-based inequality mea
ies on micro-dynamics of civil war, for example, Kalyva
2006; Tarrow
sures with standard indicators of vertical 2007).3
inequality.
Unlike group-level analyses, we can consider non-ethnic
How do researchers contributing to the empirical
as well as ethnic civil wars. Finally, we can evaluate
country-level the
literature on civil war attempt to captur
grievances?
scale sensitivity of findings from subnational Without
studies pretending to exhaust all possibil
and
to what extent these can be scaled up and
ties, we canreplicated at
divide the arguments into two main dimen
the country level. sions, namely those that focus on ethno-political and
We find strong evidence that horizontal
economic inequality and
grievances, respectively.
ethno-political discrimination matter. Countries with one
or more ethnic group (s) radically poorer than the Ethno-Political Grievances
national average and countries with large groups discrimi
nated from national politics have a significantly higher The difficulty of measuring grievances directly has led
risk of armed anti-governmental opposition. Moreover, many researchers to investigate how structural features
we find that horizontal economic inequality is primarily such as societal divisions can generate violent conflict.
associated with separatist attempts whereas widespread Although different types of cleavages, including class
ethno-political discrimination appears to motivate chal based ones, can theoretically be linked to conflict onset,
lengers targeting central governmental power. In con the most obvious alternative is to focus on ethnic distinc
trast, traditional proxies for individual-level grievances in tions because of their ascriptive and highly visible nature
a society, such as the Gini coefficient of income disparity as well as their clear importance in many actual conflic
and various fractionalization indices, have either no or (Horowitz 1985; Sambanis 2001).4 However, argument
much weaker impacts on the risk of civil war. We also linking ethnicity to conflict are usually not associated wi
show that our more theoretically informed grievance a distinctive set of causal mechanisms, and many rema
measures yield better out-of-sample predictions than quite do vague. Political economists have long suspected th
conventional models of civil war. Sensitivity tests reveal ethnic diversity leads to instability and unrest. In a classi
that these results cannot be dismissed as artifacts of a nar cal study, Rabushka and Shepsle (1972) contend that et
row conceptualization of conflict or inequality or a result nic pluralism is usually incompatible with democratic
of a specific sampling strategy. stability. More recently, a series of studies suggest th
We proceed as follows. We first review the literature ethnically
on diverse societies harbor difficult-to-solve conte
inequality, grievances, and civil war, with particulartion attenderiving from diverging preferences and differentia
tion to common arguments for dismissing the role skills of
and habits (for overviews, see Alesina and La Ferr
grievances in conflict. We discuss the difference between ara 2005; Kanbur, Rajaram and Varshney 2010). Drawing
vertical and horizontal inequality and argue that political on sociobiological reasoning about ethnic groups, Vanha
and economic inequalities that coincide with group nencleav
(1999) reaches a similar conclusion on ethnicity and
ages are much more likely to lead to violent mobilization nepotism. Based on an extensive cross-national sample,
than interpersonal inequalities unrelated to social he struc
finds that significant ethnic divisions tend to produce
tures. The following two sections provide a detailed violent conflict. More broadly, Sambanis (2001) and Fea
dis
cussion of our empirical measures and a presentation ron and of Laitin (2003) associate ethnic diversity with a lar
the empirical analysis, before we conclude. ger class of arguments outlining the role of ethnic and
nationalist grievances in conflict processes, whether pro
foundly primordialist like Vanhanen's argument, or
Inequality, Grievances, and Political Violence
explicitly modernist along the lines of Gellner (1983),
The role of grievances in conflict research attracted Anderson
criti (1991), and other prominent theorists of
cal scrutiny long before Collier and Hoeffler (2004) nationalism.
pit
ted "grievances" against "greed" as explanations of Ethnic civil fractionalization is the most common choice of
war. In contemporary conflict research, grievances are to test arguments linking ethnicity to conflict.
indicator
normally associated with relative deprivation theory,
which postulates that frustration in response to failed 3 Our characterization here does not apply to all quantitative civil wa
material expectations tend to produce violence through scholarship, and we identify and discuss some notable exceptions below.
psychological mechanisms (Gurr 1970; see also Davies
4 Other potentially relevant social cleavages that we do not pursue furthe
1962; Huntington 1968; Muller and Seligson 1987).
hereTilly
due to lack of good data include caste, clans, and narrow religious mov
(1978) and other resource mobilization theorists question
ments (for example, Mormonism and Salafism).

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
 420 Inequalities, Grievances, and Civil War

Fractionalization indices are operationalized in accor wealth comparisons. Of course, Marxist interpretations o
dance with Herfindahl's formula, which can be inter political violence as direct consequences of class conflict
preted as the probability that two randomly selected constitute the locus classicus (see, for example, Bosswell
individuals in a population belong to different groups. and Dixon 1993). Beyond this ideologically explicit theo
Initially introduced by Easterly and Levine (1997) inrizing, a a long-standing tradition of studies in comparative
study of economic development, the so-called ethno politics and sociology focus on peasant rebellions target
linguistic fractionalization indices (ELF) are usually com ing radically asymmetric land distribution in the develop
puted with data from the old Soviet ethnographic Atlas ing world (for example, Russett 1964; Moore 1966; Scott
Narodav Mira. However, fractionalization indices can in 1976). For example, in an influential study, Booth (1991
principle be computed with other group definitions that argues that persistent inequality and exploitation of peas
reflect alternative and/or separate dimensions of ethnic ants by rich landowners in Central America triggered rev
ity, including language and religion (see, for example,olutionary challenges to incumbent regimes in the 1970s
Alesina, Devleeschauwer, Easterly, Kurlat and Wacziarg and 1980s. Focusing on conflict during this period, Booth
2003; Fearon 2003). (1991: 34) claims that
Some researchers have suggested alternative curvilinear
relationships between diversity and conflict, where the economic development trends worsened the region's
risk of conflict will be lower at very high or low levels of historically extreme maldistribution of wealth and
fractionalization (for example, Sambanis 2001; Collier income, intensifying grievances among negatively
and Hoeffler 2004). Others have argued that it is not so affected class groups. ... Such problems led the
aggrieved to demand change and sparked growing
much diversity that increases the risk of conflict but
opposition to incumbent regimes by political parties,
polarization, especially a situation where two large ethnic
labor unions, religious community organizers, and rev
groups face one another (Forsberg 2008; Horowitz 1985;
olutionary groups. Violent repression of opposition
Montalvo and Reynal-Querol 2005; 0stby 2008). Although demands for reform ... not only failed to suppress
these arguments are clearly distinct and suggest different mobilization for change but actually helped forge revo
empirical measures, they are essentially pure diversity lutionary coalitions that fought for control of the state.
measures that look at the demographic size of groups
rather than their political status as the origin of insecurity In this account, we can identify a distinctive causal
and conflict.
chain starting with persistent inequality leading to griev
The conflict literature has so far failed to yield a clear ances among the peasant population fueling demands for
picture with regard to the effect of ethnicity on civil con political change and redistribution. Denied such reforms,
flict. Whereas some authors find evidence of a positive and possibly even encountering state-led repression, the
effect of ethnic diversity on conflict, including Sambanis aggrieved will see little choice but to rebel.
(2001) and Hegre and Sambanis (2006), other influential Because it is exceedingly difficult to measure griev
studies find no effect at all (for example, Fearon and ances directly in a large-N cross-country setting, most rele
Laitin 2003; Collier and Hoeffler 2004). Since fractional vant statistical studies rely on structural indicators of
ization often has been seen as a general proxy for ethnoindividual or household income inequality. The most
political grievances, many researchers have concluded widely applied such measure is the Gini coefficient,
that ethnic grievances have little or no explanatoryreflecting the extent to which the observed income distri
power. Surveying up the recent literature, Laitin (2007: bution differs from an equal distribution, with higher
25) argues that values indicating greater inequality.5 Using this indicator
as a proxy for economic grievances, the most prominent
ethnic grievances are commonly felt and latent; the studies of civil war find no evidence of a link between
factors that make these grievances vital and manifest
economic inequality and conflict. While acknowledging
differentiate the violent from the nonviolent cases. Ex
some possible data problems, these scholars interpret this
ante measures of grievance levels are not good predic
tors of the transformation of latent grievances into
non-finding as a confirmation that grievances are largely
manifest ones. And it is the factor that turns latent irrelevant for explaining civil war (Fearon and Laitin
grievances into violent action that should be consid
2003; Collier and Hoeffler 2004).
ered as explanatory for that violence. Boix (2008) refines the standard argument about
inequality and conflict by considering the impact of fac
Despite such attempts to dismiss grievances astor irrele
mobility. According to his logic, conflict is likely only
in those cases where inequality relates to immobile
vant for explaining conflict, our first hypothesis expresses
resources since wealthy elites are unable to move their
the general expectation from individual-level arguments
relating to ethno-political grievances: wealth abroad should political change threaten their
assets. Relying on structural measures of landownership
rather
Hypothesis 1: The probability of civil war increases with than comparisons of income levels directly, Boix
ethnic
diversity. reports strong support for a link between differences in
wealth and conflict. Likewise, influential formal politico
economic models that take classes or social interests as
Economic Grievances
actors, such as Acemoglou and Robinson (2006), pos
late a strong relationship between income distribution
The classical formulation of relative deprivation inspired
by Davies (1962) assumes that conflict-inducing frustra preferences for redistribution, and incentives for viole
revolution.
tions stem from a gap between actual outcomes and aspi
rations (Gurr 1970). However, relative deprivation can
also be defined in relation to differences to wealthier
5 Other studies, such as Muller and Seligson (1987), have relied on alter
members of a society. Income inequality is the most obvi
native measures of income distributions such as the share of income held by
ous way to measure grievances based on interpersonal
the poorest or wealthiest percentiles. These measures are also based entirely
on the observed income distribution for individuals or households.

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
 Halyard Buhaug, Lars-Erik Cederman and Kristian Skrede Gleditsch 421

Ethno-Political Grievances
The following hypothesis captures the preceding argu
ments:
Arguments hinging on ethnic diversity, measured
through fractionalizadon and other individual-based
Hypothesis 2: The probability of civil war increases with eco
indices, fail to capture group-level grievances and are
nomic inequality among individuals.
thus poor proxies for most established theories of eth
nic conflict and nationalism. Instead of focusing on
The conventional literature that pitches explanations
of civil war outbreak either at the individual level or more merely ethno-demographic properties, it makes more
sense to articulate an explicitly political account that
generally at the level of entire societies says little about
characterizes the relationship between the ethnic group
substate actors and structures operating between these
(s) in power and those that are excluded from access
two levels, such as ethnic groups and organizations. This
to executive power (Cederman and Girardin 2007).
lacuna may explain the divergent findings and lack of
Importantly, tapping the political configuration of eth
support for a relationship between grievances and civil
nicity implicitly also entails a temporal dynamic, since
war. We now turn to theories that specifically highlight
hold on national power and other political privileges in
the group-level perspective.
a society—unlike relative group sizes—can change over
short time periods.
The French Revolution initiated a new era in world
Linking Group-level Inequality and Grievances to Civil
War politics that made nationalism the dominant sour
political legitimacy. The limited social intrusivene
As we have seen, the most prevalent proxies for griev
pre-nationalist states meant that borders could be
ances depend on individualist principles and are insensi
adjusted primarily according to the geopolitical demands,
tive to other social cleavages or group structures.
but this flexibility was undermined in a system emphasiz
However, civil wars are not primarily fought between indi
ing that cultural and political borders should coincide
viduals, but between governments and organized nonstate
(Gellner 1983). Fierce competition broke out in areas
groups. According to Stewart (2008: 11):
characterized by intersecting ethnic and political bound
the majority of internal conflicts are organized group aries once the state became the coveted prize of aspiring
conflicts—they are neither exclusively nor primarily a national movements. By excluding entire ethnic groups
matter of individuals committing acts of violence from power, incumbent elites were able to hoard power
against others. What is most often involved is group and limit the distribution of the spoils to the in-group.
mobilization of people with particular shared identi Yet, despite the immediate advantages accruing to the
ties or goals to attack others in the name of the favored group, such exclusionary policies are likely to
group. trigger conflict as grievances grow among the powerless
and discriminated parts of the population (Gurr 1993;
In order to capture this important distinction, Stewart Cederman, Wimmer and Min 2010).
contrasts vertical inequality (VI) among individuals (VI) This process requires a fair amount of political mobili
with the notion of horizontal inequality (HI) across zation and leadership in order for a sense of moral out
groups. More specifically, His are defined as "inequalities rage to spread in the concerned population. Indeed,
in economic, social or political dimensions or cultural sta emotional commitment is clearly not enough, because
tus between culturally defined groups" (p. 3). Of the four weak movements may be effectively crushed by powerful
dimensions conceptualized by Stewart, we will focus on governments. Thus, only rebel organizations that control
the economic and political aspects of horizontal inequal sufficient material and organizational resources are able
ity, which can be contrasted directly to vertical income to challenge the state through violent means (Tilly 1978;
inequality as a measure of economic grievances, and eth Regan and Norton 2005). Contrary to the beliefs of the
nic fractionalization as an indicator for ethno-polidcal resource mobilization school, however, it does not auto
grievances. matically follow that the effect of grievances is swamped
Of course, the cohesion of ethnic groups cannot be by power differentials. Instead, we postulate that the
taken for granted across the board (Brubaker 1996)— stronger the emotional power of the grievances in the
defection may occur in many cases (Kalyvas 2006) and first place, the more readily the rebels will be able to
identities sometimes shift as a result of conflict (Gurr overcome collective-action dilemmas blocking armed
1993)—but social psychological theory offers strong rea resistance (Goldstone 2001; Emirbayer and Goldberg
sons to believe that individuals often identify through2005). Since grievances in turn depend on the severity of
groups (see Tajfel and Turner 1979). Rather than relying initial horizontal inequality, we arrive at the following
on direct personal relations, the massive scale of social hypothesis that measures horizontal inequality in terms of
systems in the modern world leaves actors little choice political discrimination:
but to rely on categorization to simplify reality (Gellner
1964). Mass media, education, and other identity-confer Hypothesis 3: The probability of civil war increases with politi
ring mechanisms allow political institutions to foster cal discrimination.
collective identities that often are associated with consid
Note that this hypothesis highlights the degree of
erable emotional commitment. Political ideologies, espe
discrimination in a country rather than merely focusing
cially those appealing to nationalist values, can engender
on the size of the excluded population (cf. Wimmer,
a strong sense of solidarity. In such cases, individual pref
Cederman and Min 2009). If the emotion-based mecha
erences are trumped by collective motivations, implying
nism outlined above holds, we should be able to detect
that the individual acts on behalf of the group and is will
an especially strong link between discrimination, viewed
ing to make major sacrifices in the name of collective
as a subset of exclusionary policies, and conflict onset.
identities and abstract ideological principles (Anderson
As our analysis is pitched at the level of entire coun
1991).

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
422 Inequalities, Grievances, and Civil War

tries, which in many cases feature

tion and ahorizontal
large inequalities
number of
from disaggregated eco
excluded groups, discriminatednomic data.
groups should be espe
cially important potential rebels, even
We are now inif suchtogroups
a position formulate our last hypothesis:
are often numerically small and carry little weight in
conventional aggregate measuresHypothesis 4: The probability
of ethnic of civil war
diversity at increases with eco
the country level. nomic horizontal inequality.
Below, we explore additional aspects of the ethno-polit
This section has shown that the recent literature pro
ical environment, including the claim that a recent down
vides ample evidence that political and economic hori
grading of ethnic groups' power status is particularly
zontal inequality increase conflict risk, but so far, thes
conducive to conflict, as well as the possibility that politi
results either have been limited to parts of the world due
cal power-sharing arrangements can undermine stability
to data problems or relate to subnational units (groups
and peace (for example, Jarstad and Nilsson 2008).
or geographic areas) with little consideration of country
level dynamics. Likewise, earlier attempts to link ethnic
Economic Grievances
grievances to civil war at the country level ignore the
political configuration of ethnicity, whereas disaggregated
By now it should be clear that vertical inequality, mea
(group-level) studies of ethno-political marginalization
sured as the Gini coefficient, cannot fully capture all rele
are limited to considering ethnic conflict and must by
vant dimensions of societal disparity. In a powerful
design exclude countries where ethnicity carries no polit
critique of such individual-level conceptions of inequality cal relevance.
that bears strong resemblance to Stewart's notion of hori
This study provides the first truly global cross-country
zontal inequality, Tilly (1999, 2007) advances a "rela
assessment of how intergroup inequalities in economic
tional" perspective that explains how durable inequality
and political privileges are associated with civil war out
results from categorical differences. In Tilly's (2007: 9)
break. In so doing, we will also assess whether the hypoth
words, "a view of inequality as outcomes of individual-by
eses advanced in disaggregated studies hold at the
individual competition according to widely shared stan
country level and what types of aggregate indicators are
dards of merit, worthiness, or privilege obscures the sig
best suited to capture the theoretical arguments associ
nificance of organized distinctions and interactions
ated with horizontal inequality. Once such indicators
among members of different social categories." Rather
have been found, we can compare their performance
than being a mere reflection of differences in skills or
directly to established country-level correlates of civil
changes in the supply of resources, then, inequality thus
war and assess their contribution to our ability to predict
conceived can be seen as an outcome of "politics of
conflict out of sample. We now turn to these tasks.
exclusion" whereby political elites restrict distribution
conditional on social categories or groups.
The explicit role played by political agency points Methods and Measurements
directly to how wealth discrepancies may trigger political
Our four hypotheses are evaluated empirically throug
violence. The road from inequality to conflict leads via
country-level regression analysis of civil war involvemen
grievances, which can be seen as reactions to perceived
among all members of the international system, 196
injustice. Objective resource asymmetries are known to
2005 (see Gleditsch and Ward 1999). This seeming dep
emerge in many ways, including through colonialism and
internal domination (Williams 2003: 106-107), but do ture from the recent trend toward disaggregating ci
war (Cederman and Gleditsch 2009) might seem count
not themselves suffice to produce grievances. Members of
intuitive as the logic underlying our theoretical fram
disadvantaged groups must first be made conscious of
work explicitly refers to social groups within countr
their predicament through explicit intergroup compari
However, our country-level approach here is complim
son and convinced that the unequal distribution of
tary and not inherently incompatible with a disaggregate
wealth is not merely unjust, but also to be blamed on the
state's incumbent elite (Gamson 1992). focus. First, as we explain in further detail below, o
operational measures of horizontal inequality build o
Again, we expect other factors to influence the likeli
the notion of the "weakest link" whereby conflict risk
hood of conflict, including most importantly the power
considered a function of the relative discrepancy in
of the nonstate challenger vis-a-vis the incumbent state
wealth or privileges between the national average and
(for example, Butler and Gates 2009; Buhaug 2010).
most marginalized group in society. Second, many soc
However, as argued in connection with Hypothesis 3, if
science phenomena and correlations are scale depen
the causal process is mediated by a grievance mecha
dent, in the sense that apparent patterns in the data at
nism, the extent of structural inequality in a society
one resolution or level of analysis may disappear or
should have a discernible impact on the outbreak of vio
lent conflict. change at other scales.6 Generating country-aggregated
indicators from group-specific data thus permits evaluat
In contrast to the at best mixed results of the large-N
ing the scale dependence of earlier findings (for exam
studies focusing on vertical inequality, Horowitz (1985)
ple, Buhaug, Cederman and Rod 2008). Moreover, and
forcefully argues that both "backward" and "advanced"
importantly, it also facilitates comparing the performance
groups are overrepresented as conflict groups and pro
of intergroup inequality with standard (vertical) inequal
vides case studies to support his claims. Likewise, Stewart
ity measures at the country level. Countries are arguably
(2008) reports on a series of case studies that strongly
the most relevant units of observation for risk profiles as
support the importance of horizontal inequalities. Using
survey data from Africa, 0stby (2008) and Hegre et al.
(2009) have also been able to find confirming evidence 6 For example, population size and oil dependence are found to increase
for the thesis at the group level. More recently, Ceder a country's baseline civil war risk (Fearon and Laitin 2003), but it does not
man et al. (2011) provide further support using spatial necessarily follow that subnational conflict risk is highest in the most densely
methods for deriving global estimates of wealth distribu populated or oil abundant regions of a country (Buhaug and Rod 2006).

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
 Halyard Buhaug, Lars-Erik Cederman and Kristian Skrede Gleditsch 423

mann,input
well as forecasting, as available projected Girardin, Cederman
data on and Wimmer 2011).9 We then
core features such as economic development, democrati
identified the richest and poorest group in each country,
zation, and demographic changes almost fromexclusively per
which we constructed country-level inequality indica
tain to countries (cf. Goldstone, Bates,tors that capture
Epstein, the relative gap between the mean
Gurr,
Lusdk, Marshall, Ulfelder and Woodward national
2010;
income
Hegre,
and the income level for the poorest
Karlsen, Mokleiv Nygard, Urdal and Strand 2013).
and richest group, respectively:10
Data on civil war onset and ethnic group involvement
NHI: Negative horizontal inequality = country-level
are derived from the Non-State Actor data set (Cunning
GDP per capita/mean per capita income for poorest
ham, Gleditsch and Salehyan 2009), which in turn is
based on the UCDP/PRIO Armed Conflict Dataset, group.

henceforth ACD (Gleditsch, Wallensteen, Eriksson, Sol PHI: Positive horizontal inequality = mean per capita
income for richest group/country-level GDP per
lenberg and Strand 2002; Themner and Wallensteen
capita.
2011). The link from organizations to the ethnic groups
in the Ethnic Power Relations data (Cederman et al. Since the G-Econ data represent the year 1990 and are
time invariant, our economic inequality variables are sta
2010) was established through the ACD2EPR coding tic. This is unfortunate as we know that almost all coun
project. We use the most inclusive definition of civil war,
tries experienced considerable economic growth durin
counting all conflicts between a state and one or more
the sample period, and growth rates vary between cas
rebel groups that generated at least 25 battle-related
and over time as well. Yet, this limitation may be acce
deaths in a calendar year. We use two alternative depen
able if we do not have strong reason to believe that th
dent variables (DVs). The first is a standard binary indica
spatial distribution of poverty and wealth changes signifi
tor, where civil war onset is coded in the initial year of a
cantly over time also within countries. Lack of good
new armed intrastate conflict and after a lull in fighting
subnational data on economic activity hinders a compr
in excess of two calendar years (183 observations). In
hensive assessment of this issue, but much research su
addition, we use a four-category onset indicator that sepa
gests that social inequalities are often persistent, implyin
rates between onsets of different conflict types (no onset
that static georeferenced income data may be less pro
is the reference group, coded zero):
lematic than intuitively assumed (see, for example, Til
1. Ethnic territorial conflict, 55 observations; 1999; Stewart and Langer 2008). For India, one of a
2. Ethnic governmental conflict, 42 observations; and handful of countries with reliable time-series data on eco
3. Non-ethnic conflict, 86 observations.7 nomic activity at a subnational level, we have nearly ide
tical economic growth rates for all states during the las
The classification of territorial and governmental con
30 years, according to statistics from the Reserve Bank of
flict is based on the ACD incompatibility indicator. Fur
India (see supplementary information for details). A sim
thermore, conflicts are considered ethnic if a rebel group
lar stationary ordering of poor versus rich regions is ev
makes claims on behalf of a specific ethnic community
and recruitment is based on ethnic affiliation. For both
dent in France (Combes, Lafourcade, Thisse and Toutain
2011), and other countries such as the Yugoslav Federa
variants of the dependent variable, subsequent years of
tion and the Sudan also seem to be well represented by
conflict activity are coded as zero except where a new
the G-Econ data (Lang 1975; Buhaug, Gleditsch, Holter
conflict breaks out.8 For sensitivity tests, we also use Fea
mann, 0stby and Tollefsen 2011).11
ron and Laidn's (2003) civil war data, which we classify in
A potentially more challenging problem is the possibi
the same conflict subcategories based on their identifica
tion of ethnic/non-ethnic and center/exit wars. ity of reverse causality, as relative poverty at the grou
level may reflect past conflict. In other work (Buhaug
We consider a number of potential proxies for ethnic
et al. 2011), we have explored the relationship between
grievances and inequality. The models presented below
feature standard, individual-based measures of ethnic and
local economic activity and conflict in detail and con
ducted various sensitivity tests such as limiting the analysis
economic diversity: Fearon and Laitin's (2003) ethnic
to the post-1990 period and accounting for (or exclud
fractionalization index (ELF) and a Gini index of income
ing) areas that have hosted armed conflict in the past
dispersion (World Income Inequality Database, WIID).
These tests failed to reveal strong indication of an endo
To minimize missing data problems in the WIID data, we
enous relationship between conflict and income.12 From
apply linear interpolation between data points and
a policy perspective, a predictive link between economi
extended the time series by copying the earliest/latest
marginalization and elevated conflict risk is in itself of
known value to earlier/later years by country.
considerable importance, regardless of the underlying
Measures of horizontal economic inequality were gen
reasons why economic inequality arose in the first place
erated through a number of steps. First, we calculated
group-level data on wealth for all ethnic groups in all
countries by joining the G-Econ gridded data set on eco 9 See Buhaug et al. (2008) for further documentation on how group-spe
nomic activity (Nordhaus 2006) with the GeoEPR data setestimates can be constructed from spatial data by means of geographi
cific
on ethnic group settlements (Wucherpfennig, Weid
information systems (GIS) software.
10 In ethnically homogenous countries (for example, North Korea) and
countries where ethnicity has no distinct spatial dimension (for example
Rwanda),
7 There is little value in disaggregating outcome category 3 with respect tothese measures take on the value 1.
11 Note that the problem of restricted temporal domain applies with equa
incompatibility as virtually all non-ethnic civil wars fall in the governmental
conflict category. force to the common proxies for vertical inequality. Most measures of ethn
fractionalization are based on Soviet data from the 1960s, whereas Gini mea
8 Our definition of defining civil war implies that a country may host sev
sures ofor
eral distinct armed conflicts at the same time, involving distinct groups income inequality often depend on heavy interpolation and extrapo
lation.
incompatibilities (examples include Ethiopia, India, Myanmar, and Yugosla
12 We do not dispute that major wars can have devastating and long-last
via). Recoding observations with ongoing conflict as missing (that is, consider
ing civil war countries not at risk of facing another challenger)ing impacts
does not on the local and national economy, but few of the armed intra
substantively affect the results presented here. state conflicts included in this analysis reach this magnitude.

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
424 Inequalities, Grievances, and Civil War

To our knowledge, the G-Econ-based A indicators

rus constitute
the only available data of intergroup inequality with a
global coverage.
Our second intergroup grievance indicator captures
systematic inequality in ethno-political opportunities,
based on the Ethnic Power Relations data (Wimmer et al.
2009). The EPR project identifies political status for all
politically relevant ethnic groups worldwide for all years thi „
since 1946. In this study, we focus on political discrimina 11^*0
arg

tion as a potential source of ethnic grievance. At the .

yug
B

saf
country level, we consider the demographic size of . sud

the largest discriminated ethnic group (LDG) relative to chn ■ ■ "

per b „
■ «

" ■ NAM
the joint size of the discriminated group and the group
1 /
zim
, ( »» ■ ■
* "" - *' ■ U.k.ai'lA.to L>«/* ■>
(s) in power. This variable is naturally bounded within i

the interval [0, 1]. We further include two dummy 20 40 60

GINI
80

variables to control for additional aspects of the ethno

political context. The first indicator flags whether one or B rwa
more ethnic group(s) in the country lost political status ■

during the preceding year (downgrade).1' Second, we irq

identify country years where the political system is

founded on a division of executive power between leaders jor ang
of different ethnic groups (power-sharing).
Figure 1 compares our group-based indices with con o

ventional measures of ethnic and economic dispersion. 9^

bcdi
Evidently, economic marginalization of ethnic minorities
„

isr
lIt ken
may be substantial even in countries with seemingly egali kuw
est

tarian wealth structures (for example, Russia). Similarly, sau ■ zim afg l®r uga
discrimination of large ethnic groups is found in rela tur

* " ■
tively homogenous (polarized) as well as very heteroge "
« /

■* * * ■ *
neous societies. We also note that many of the "*** ** " m K— ma * m torn ■

Observations with high intergroup economic/political .4 .6

inequality scores (vertical axes) have a recent history of ELF

intrastate conflict.
Fig 1. Vertical Versus Horizontal Inequality, Year 2000
In addition to the various inequality and dispersion
measures, we consider a number of control variables that
conceivably may be correlated with both horizontal
inequality and conflict: logged GDP per capita (Heston, HI Model," retains all right-hand-side regressors, but
Summers and Aten 2009), democracy (Gates, Hegre, replaces the standard civil war onset variable with the
Jones and Strand 2006), and logged population size (Hes four-category outcome variable that distinguishes between
ton, Summers and Aten 2009). In addition, to account different types of conflict. The results are displayed in
Table 1.
for possible serial dependence and a different risk pat
tern for countries already involved in intrastate fighting,In line with some earlier research, Model 1 indicates
we include a civil war lag indicator.14 All controls thatare ethnic diversity is positively correlated with civil war
lagged by 1 year to minimize bias from possible reverse onset. The estimated effect is quite large in substantive
causality. terms and statistically significant. All other factors held at
median values, the predicted risk of civil is nearly three
times higher when ELF is at the 95th percentile
Regression Analysis
(ELF = 0.86) compared to the predicted value at the 5th
We estimate a series of binary and multinomial logit percentile (ELF = 0.03). Vertical income inequality, in
regressions to assess the hypotheses. We start with the contrast, appears unrelated to civil war, in line with most
conventional binary civil war onset indicator as the earlier studies. We also note that national political config
dependent variable. The first model, which we refer to asuration is largely unrelated to the likelihood of civil war.
the "VI Model," is a standard model of civil war onset There is some indication of a parabolic effect of democ
that contains the ELF and Gini proxies for vertical racy with semi-democracies being more conflict prone
ethno-political and economic grievances plus controls.(results not shown), although the results do not reach
The "HI Model" additionally includes group-based (hori statistically significance by conventional criteria (p > 0.1).
zontal) measures of ethno-political discrimination and Consistent with Hegre and Sambanis (2006), we find that
economic inequality. The third model, or the "Extended population size and level of development have significant
positive and negative impacts on the risk of conflict in
the conventional VI Model.
13 The EPR data set classifies politically relevant ethnic groups into one of In the HI Model, we introduce the new horizontal
seven possible categories according to their extent of access to central state grievance proxies as well as controls for power-sharing
power: monopoly, dominant, senior partner, junior partner, regional auton
among ethnic groups and downgrading of their power
omy, powerless, and discriminated. Any shift downwards on this hierarchical
ladder implies political downgrading.
status (see Model 2). We immediately note that the effe
14 The results do not change if we replace the lagged conflict incidence of ethnic diversity drops by about 15% while individu
dummy with Beck, Katz and Tucker's (1998) non-parametric approach to time income inequality remains insignificant. More impor
dependence conditional on years at peace or time since independence. tantly, we now find that ethnic politics matter. In agree

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
 Halvard Buhaug, Lars-Erik Cederman and Kristian Skrede Gleditsch 425

Table 1. Determinants of Civil War Onset, 1960-2005

(3) Extended HI model

(I) VI model (2) HI model
All civil wars All civil wars Eth. terr. Eth. gov. Non-eth.

ELF 1.148** 0.974* 1.713 1.623 0.394

(0.424) (0.428) (0.977) (0.892) (0.521)
Gini -0.005 -0.004 -0.039 -0.029 0.024*
(0.010) (0.010) (0.025) (0.025) (0.011)
LDG 1.288** -0.219 3.476** 0.666
(0.346) (0.830) (0.626) (0.588)
PHI -0.045 -0.036 -0.810 0.045
(0.175) (0.252) (0.857) (0.246)
NHI 0.321** 0.497** -0.082 0.201
(0.119) (0.161) (0.388) (0.215)
Downgrade 0.860** 1.391** 0.944 0.422
(0.255) (0.418) (0.526) (0.448)
Power-sharing -0.029 -0.769 0.862* 0.062
(0.221) (0.484) (0.438) (0.314)
Democracy 0.176 0.350 1.374* -0.091 -0.157
(0.319) (0.345) (0.607) (0.819) (0.452)
Population 0.249** 0.234** 0.408** 0.067 0.167
(0.069) (0.079) (0.123) (0.184) (0.101)
GDP per capita -0.382** -0.432** -0.773* -0.405 -0.188
(0.140) (0.147) (0.329) (0.305) (0.194)
Civil War lag 0.161 -0.026 0.193 -1.022 0.168
(0.279) (0.298) (0.466) (0.992) (0.357)
Constant -5.968** -6.311** -8.511** -4.729* -7.216**
(0.782) (0.850) (1.729) (2.043) (1.095)
Pseudo-fl2 .06 .08 .11
Observations 5,219 5,219 5,219

(Notes. Binary (1 & 2) and multinomial (3) logit coefficients with standard errors clustered on countries in parentheses. LDG = largest discriminated group;
PHI = positive horizontal inequality; NHI = negative horizontal inequality. **p < .01, * p < .05.)

ment with Hypothesis 3, regimes founded on political dis only pertain to a particular conflict type or run in oppo
crimination of sizable ethnic groups are disproportion site direction across distinct types of conflicts (Sambanis
ately involved in civil war, and the magnitude of the 2004).
effect is on par with that of ELF. Moreover, the HI Model In the Extended HI Model, we estimate the effects of
supports our expectation that countries with economi the grievance proxies specifically for ethnic separatist
cally highly marginalized groups are more conflict prone wars (outcome 1), ethnic governmental wars (2), and
(see Hypothesis 4). We find little evidence that relative non-ethnic wars (3), almost all of which are governmental
wealth increases conflict risk, thus contrasting the near (see Model 3).16 The results are striking. ELF no longer
symmetric U-shaped association between economic exhibits a significant effect with conventional levels of
inequality and conflict at the ethnic group level reportedconfidence on any positive outcome (although the mar
by Cederman et al. 2011. Although we note that the latter ginal impact for the point estimate on both ethnic con
study is limited to ethnic conflicts in the post-Cold Warflict types—ignoring the large standard errors—remains
period, it remains to be determined whether the dis quite high). At the same time, the positive coefficient for
crepancy is also in part an aggregation effect. Interest ethno-political discrimination nearly triples for ethnic
ingly, including horizontal inequality also improves the governmental conflicts compared with the basic HI
performance of GDP per capita by some margin.15 TheModel, while it remains insignificant for other conflict
other covariates are largely unaffected by the inclusion of types. Indeed, the elasticity of discrimination with respect
the group-based grievance variables. to DV outcome 2 amounts to a factor of five (estimated
Models 1-2 offer empirical support for our claim that civil war risk increases from less than 0.003 to 0.013 with
intergroup inequalities matter more for civil war risk than a shift from p5 to p95, all other variables held at their
vertical disparities. Yet, not all conflicts are the same; median values). This result supports the expectation tha
prior research has shown that territorial (that is, primarily populous, politically discriminated ethnic groups are
separatist) and governmental (that is, primarily revolu more likely to seek to overthrow the ruling regime o
tionary) conflicts differ on several dimensions (Buhaug otherwise alter the political system through violent means
2006). Similarly, conflicts may be categorized as either if necessary. Moreover, countries with one or more ver
ethnic or non-ethnic (the latter sometimes being referred poor ethnic groups—which typically make up only a fra
to as ideological; see Sambanis 2001). Aggregating all civil tion of the country population—are more likely to se
wars could thus mask important differences in effects that conflicts that aim for separation from the core or

The marginal impact of (negative) intergroup economic inequality 16 Cf. Wimmer et al. (2009), who distinguish between secessionist and
should be interpreted with some care, however, as the parameter estimate non-secessionist conflicts. We prefer the distinction between territorial and
shrinks significantly when the most unequal societies (Argentina, Russia, and governmental conflicts since it circumvents the highly heterogeneous category
Thailand in some years) are removed from the sample. of non-secessionist conflicts.

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
426 Inequalities, Grievances, and Civil War

demand greater levels of autonomy fromrather

data on intergroup discrepancies, are better able to
that attempt
ing to capture governmental power.distinguish
This between conflict
result and non-conflict observations
actually
than standard vertical
becomes stronger if we drop the outliers. Lastly,measures
weoffind
ethnic and economic
diversity. Calculations
some evidence for class-based mobilization in thatofhigher marginal effect for individua
variables demonstrate
individual income inequality is positively associated that this
with difference is not only si
nificantcivil
the risk of non-ethnic, revolutionary in statistical
war terms
(see but
alsoalso quite large in substan
Goldstone 2001). tive terms. Next, we compare the predictive performance
Overall, our analysis shows that conventional explana of the HI Model and a reduced version of the VI Model.
tory variables of civil war are much better at accountingGiven the widespread interest in the disappointing out-o
for territorial than governmental conflict. In fact, ethnic sample prediction ability of conventional country-leve
governmental conflicts are explained largely by a discrimstatistical models of civil war and the aspirations of for
inatory political system and power-sharing. Whereas less casting conflict through efforts such as the Political Insta
than one-third of all observations in our sample have a bility Task Force project (see Goldstone et al. 2010;
system of ethno-political power-sharing, the share is above Ward, Greenhill and Bakke 2010), we wish to consider
60% for cases with ethnic governmental conflict whether more theoretically grounded measures and disag
outbreaks. The latter finding suggests that consociational gregated information can improve on standard models
regimes are particularly prone to factional fighting relying on conventional country-level indicators. We use
over control of the executive. Yet, there may also be the a observations for the 1960-1999 period to train the
selection effect at play here, whereby countries withmodels and then use the estimated probabilities for coun
higher perceived interethnic competition are more likely tries in 1999 to predict civil war onset out of sample,
to establish a system of institutional power-sharing. It within
is the next decade, 2000-2009. To facilitate direct
remarkable that the Gini coefficient is the only covariate comparison, we exclude the VI indicators from the HI
that obtains moderate statistical significance for non-eth model (unlike Model 2) and estimate both models on
nic conflicts in the Extended HI Model. This may partly the exact same sample of observations.18 The selection of
reflect greater heterogeneity among the so-called "ideo control variables is identical to the models presented
logical" civil wars that cannot be accounted for with con above.
ventional explanatory variables. A first, simple test would be to apply a binary classifica
The results from Table 1 provide suggestive evidence tion scheme and compare predicted outcomes for the
that grievances and inequalities matter for violent con two models with data on actual outbreaks of civil war. We
flict, although not in the simple, individualist manner first aggregate the yearly probabilities for 1999 into risk
implied by the demographic and rather apolitical arguof conflict over the subsequent decade p*w and then
ments that have been operationalized with the ELF and convert the continuous prediction scores into a binary
Gini indices. Instead, supporting our group-based reason predicted onset/no onset outcome by using p* = .5 as
ing, political discrimination and economic marginalizathe threshold criterion for predicted conflicts. As shown
in Table 2, by this procedure the VI model predicts civi
tion of ethnic groups both exhibit positive and statistically
war onset in 14 of the 130 sample countries during th
significant effects on the risk of civil war. Inequity in polit
ical participation and power might in principle be associ first decade of the new millennium (p* > .5). Four of
ated with armed conflict of any kind; however, whenever these predictions were accurate, whereas there were
access to these privileges is determined by ethnic affilia another 22 civil wars not predicted by the model. The
tion and sizable groups of society are subject to systematicremaining 104 out-of-sample observations have p* < .5
discrimination, the odds of mobilization and conflict aim and hence are classified as no onset. 94 of these predic
ing at restructuring the political system increase. Large tions were true, while ten non-war observations are
politically discriminated groups constitute a larger threatmissed (false positives). The HI model fares better; it suc
to the ruling regime than small, peripheral minorities cessfully predicts eight civil war onsets—twice as many as
and are more likely to succeed in capturing and maintainthe VI model—while the number of false onsets drops to
ing state control. The African National Congress' strugglenine. At the same time, it correctly identifies 95 non
against the Apartheid regime in South Africa and the Shionset countries, whereas the number of false negatives
ite resistance to Saddam Hussein's Ba'athist minority gov (that is, missed civil wars) is 18. In other words, using
ernment constitute two relevant examples here. p* — .5 as the classification criterion, the HI model cor
Conversely, countries with large intergroup discrepancies rectly identifies 44% of all civil war onsets and 91% of
in wealth and economic opportunities are more likely the to non-onsets during the subsequent decade; the corre
face separatist challenges. The Chechen wars of 1994 sponding figures for the VI model are 15% and 90%,
and 1999 serve as near-ideal-type cases of this dynamic, respectively.
occurring in Russia's least developed part of the country The appropriate prediction threshold c depends on
(Hale and Taagepera 2002). Other relevant examples our relative costs for identifying true events and false pos
include the Kurdish nationalist struggle in Turkey and itives. A more comprehensive comparison of the models'
the Albanian uprising in Kosovo in 1999. forecasting capability across a range of different threshold
values is provided by a receiver operating characteristic
(ROC) plot. ROC curves visualize the rate of true posi
Out-of-Sample Predictions
So far, we have shown that our new country-level indica
tors of ethnic and ethno-political inequalities, derived 18 A number of countries are dropped due to missing data on the
indicator. The results of the out-of-sample assessment do not change
allow each model to be estimated on (and generate predictions for) the
17 As the income inequality measures are normalized by the average valid sample.
national income per capita, high inequality values by design are driven by 19 The probability of conflict over the decade is defined from the ann
small minority groups, which have little impact on the calculation of national probabilities p by p* = 1— (1— p)10 , that is, as the complement of the p
mean income. bility that an observation will not see conflict in any of the 10 years.

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
 Halyard Buhaug, Lars-Erik Cederman and Kristian Skrede Gleditsch 427

Table 2. Classification Table for Out-of-Sample Prediction, 2000

2009

VI model prediction HJ model prediction

Observed No onset Onset No onset Onset

No civil war onset 94 10 95 9

Civil war onset 22 4 18 8

(Notes. The table shows out-of-sample binary

sification criterion for coding civil war onset,
-1999.)

.4 .6
VI model

Fig 3. Comparison of Out-of-Sample Predictions for HI and VI Mod

els, 2000-2009

false-positive prediction (that is, false as there was no new

conflict onset after 2000). Conversely, some countries
seem more conflict prone when judged by their ELF and
Gini indicators than by ethno-political discrimination and
income deviation for the poorest group. Tanzania and
the Philippines are both considered to be about 50%
more at risk in the VI model, but neither experienced a
civil war outbreak in the prediction period. The least
likely case that actually saw civil war out of sample is the
False positive
United States (p*m = .17), which is perhaps a question
VI model HI model able classification of the 9/11 attacks in the Uppsala
data.21 Ethiopia is the most likely candidate for conflict
Fig 2. Receiver Operating Characteristic that did not see a new
Curves forcivil
VIwarandoutbreak
HI (p*ui = .77). Of
Model Predictions, 2000-2009 the 18 false negatives for the HI model (triangles seen in
the lower left quadrant of Figure 3), a majority either
endured a civil war at the outset of the prediction period
tives against the rate of false positives across the full (six countries) or had not experienced conflict in at least
range of possible cutoff points c for a binary variable a decade (six cases)—each of which condition lowers the
p* > c (see Hosmer and Lemeshow 2000). The better a a priori probability of a new civil war onset.
model predicts, the more steeply the curve rises and the
larger the area under the curve (AUC, expressed as share
of the total area of the plot). As seen in Figure 2, the Sensitivity Analysis
ROC curve is higher for the HI model almost across theAlthough the results presented thus far are encouraging
board and the AUC score is notably larger than that ofand point toward a clear conclusion, we consider a num
the VI model.20 Evidently, the predictions from the ber of additional robustness tests to address likely chal
model with group-based indicators of horizontal inequal lenges. One possible concern relates to the inclusive
ity perform better in identifying the countries that seenature of the ACD data, which cover all armed intrastate
civil war onset out of sample than the predictions from a
conflicts with at least 25 annual casualties. This compara
model that consider standard grievance proxies based ontively low fatality threshold allows recording multiple con
individual-level ethnic/economic dispersion indices. flicts in the same country at the same time. It could be
Figure 3 provides a complementary assessment of the that our results are driven by a number of low-intensive
discrepancy in predictions between the VI and HI mod conflicts and that the reported relationship between
els. For most countries, the predictions of the two models
inequality and conflict is not representative for more
are similar, but there are some notable exceptions. The severe and (arguably) more politically relevant major civil
estimated risk of conflict in Russia from 2000 to 2009, for
wars.22 In Table 3, we replace the ACD conflict data with
example, is twice as high in the horizontal or group-based
model compared with the vertical inequality model,
reflecting the comparatively large between-group eco 21 Somewhat controversially, the UCDP/PRIO data project treats the 9/11
nomic and political inequalities in the country. Similar terrorist attacks as a civil war over the control of the US government. This
conflict would not have fulfilled the data set's inclusion criteria for an armed
patterns are found for Rwanda and Yugoslavia as well,
intrastate conflict had al-Qaeda limited itself to striking civilian targets, but is
although in the latter case, the HI model returned a
included since the Pentagon qualifies as a government target.
22 For example, Fearon (2010) speculates that the low fatality threshold of
20 The predictive power of the VI model exceeds the HI model only in the a UCDP/PRIO data leads to an overrepresentation of conflict observations
narrow band where the true positive rate is very high (>0.75) and the false in populous and highly ethnically fractionalized countries. However, we note
positive rate is also quite high. Since the left part of curve—where the ratethat of at least in our data set, the ELF score for conflict countries is actually
marginally lower when all UCDP/PRIO intrastate conflicts are considered
true vs. false positives is the highest—is the most relevant, it is difficult to see
the higher performance of the VI model in this area as strong support for than when the sample is restricted to Fearon and Laitin's (2003) major civil
that model. wars.

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
 428 Inequalities, Grievances, and Civil War

Table 3. Alternative Civil War Data, 1960-1999 as democratic and inclusive societies (LDG at 5th percen
tile), all else held constant. The marginal impact of high
(5) Extended HI Model
(4) HI Model negative economic inequality is comparable, increasing
All cw Eth. terr. Non-eth.
the estimated civil war risk more than twofold with a cor
Eth. gov.
responding shift in values for the NHI indicator. We
ELF 0.183 1.834* -0.154 -1.316 interpret this as additional empirical support for out
(0.562) (0.909) (0.825) (0.944) claim that politically relevant ethnic inequality trigge
Gini 0.007 0.001 0.005 0.022
conflict, whereby the (extent of) unequal distribution
(0.013) (0.020) (0.022) (0.024) economic and political privileges between ethnic group
LDG 1.501** -0.767 3.654** -0.300
has a consistent and positive impact on the probability
(0.473) (0.978) (0.749) (0.982) rebellion.
PHI -0.179 -0.068 -0.070 -1.451
Next, we reassess the specific expectations on how
(0.247) (0.289) (0.475) (1.712)
NHI 0.526** 0.549** 0.456 0.370
ethno-political and economic grievances might relate to
various types of civil war. Again, the results are encourag
(0.148) (0.153) (0.414) (0.596)
0.332 0.279 0.601 0.112
ing in the sense that our key HI indicators replicate the
Downgrade
(0.458) (0.653) (0.810) (0.942)
pattern found in Table 1. The effect of intergroup ethnic
Power-sharing 0.317 -0.349 0.871 1.066* and political disparities is evident even in the limited
(0.308) (0.467) (0.554) (0.510) sample of severe civil wars. Economic marginalization of
Democracy 0.423 0.159 0.835 0.337 one or more ethnic minorities significantly increases the
(0.492) (0.639) (0.752) (1.081) risk of separatist conflict, but not other forms of orga
Population 0.207* 0.461** -0.266 0.252 nized, state-based violence. In contrast, political discrimi
(0.084) (0.117) (0.187) (0.146) nation of the most likely contenders for state power (that
GDP per capita -0.580** -0.405 -0.878** -0.444 is, the largest groups not in government) increases the
(0.186) (0.231) (0.291) (0.415) likelihood of governmental ethnic conflict by a factor of
Civil war lag -0.631 -0.619 -32.060** -0.094 four, but has no systematic bearing on territorial or non
(0.344) (0.583) (0.453) (0.607) ethnic wars.25 Interestingly, our group-based controls for
Constant -6.808** -10.550** -4.151* -7.200** downgrading and power-sharing lose much of their
(1.121) (1.588) (1.980) (2.634) impact on major civil wars. While this might indicate that
Pseudo-i? .06 .11
certain ethno-political constellations and events might
Observations 4,433 4,433
carry greater potential for escalating conflict to all-out
(Notes. Logit and mlogit coefficients with standard errors clustered on coun
wars than others, we are reluctant to put too much
tries in parentheses. LDG, largest discriminated group; PHI, positive horizon emphasis on the observed differences since these out
tal inequality; NHI, negative horizontal inequality. Civil war data from Fearon comes are quite rare. Except for the positive and signifi
and Laitin (2003). **p < .01, *p < .05.) cant association between ethnic fractionalization and
ethnic territorial wars, Model 5 reveals only trivial
for the vertical inequality measures, adding fur
weight to our argument that it is the sociopolitical c
Fearon and Laitin's (2003) civil war data. This data set is uration of ethnicity, rather than diversity per s
limited to armed conflicts that generated at least 1,000 causes civil war.
deaths in total, with a yearly average of at least 100 deaths
Space constraints prevent a thorough presentation of
and with at least 100 killed on each side. Model 4 is a re
all the additional sensitivity tests carried out to examine
estimation of Model 2, whereas Model 5 is identical to
the robustness of the results, and we refer to our supple
Model 3 except for choice of DV.23 mentary information for a complete documentation of
We immediately note the weak and insignificant effectthese tests. Among other things, we sequentially replaced
of ethnic diversity in Model 4, which stands in contrast to
each of the four sets of inequality indicators with alterna
its substantial impact in Model 2. Evidently, countries
tive measures. As an alternative measure of demographic
with many ethnic groups are more likely to be challengeddiversity, we used Montalvo and Reynal-Querol's (2005)
by smaller (and almost always peripheral) insurgencies,
ethnic polarization index (RQ) and replaced the Gini
but these are unlikely to escalate to large civil wars.24 Ver
index of income disparity with Boix' (2008) proxies for
tical income inequality, too, appears irrelevant for major
immobile economic assets. Neither of these changes sub
civil war risk, replicating the result for the Gini index
stantively affected the behavior of the group-based
reported above. Inequality in wealth and political rights,
inequality indicators. Furthermore, we replaced the weak
when measured between groups rather than individuals, est link-inspired measure of ethno-political discrimination
however, still matters. Regimes consisting of relatively(LDG) with a derivative of Cederman and Girardin's
small political elites, with widespread discrimination (2007)
of N* index, based on the EPR data and reflecting
large ethnic groups, are systematically and substantively
extent of political discrimination (instead of exclusion
overrepresented in the conflict sample. These ethnocra more generally). As expected, the N* index replicated
cies (LDG at 95th percentile) are twice as conflict prone
our earlier finding; extensive ethno-political discrimina
tion is associated with ethnic governmental conflict, but
23 In order to separate between different types of civil war, we reliednotonwith other forms of civil war. Finally, we replaced the
relative
Fearon and Laitin's original classification of ethnic wars as well as their dis wealth/poverty indices (PHI, NHI) with a unified
tinction between "center" (that is, governmental) and "exit" (territorial) civil
wars. These models contain fewer observations as Fearon and Laitin's civil war
between-group inequality measure, analogous to the Gini
data only run through 1999. In models not shown, we also use a beta version
of Fearon's (2010) updated civil war data for an extended sample period,
1960-2008. The results do not change.
4 A similar pattern is reported by Buhaug (2006) and Hegre 25
andCalculations
Sam based on a shift in discrimination from the 5th to the 95th
banis (2006). percentile value, holding all other factors in Model 5 at median values.

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
 Halyard Buhaug, Lars-Erik Cederman and Rristian Skredf. Gleditsch 429

index (see Stewart, Brown and Mancini 2005) and based nomic horizontal inequality by comparing the relative
on the G-Econ data set.26 This indicator performed less wealth of the poorest and most affluent groups in rela
well than the directed indices and failed to return a statis
tion to the country average.
Once the conflict types have been properly unpacked,
tically significant coefficient on any type of civil war out
come. Evidently, large negative discrepancies froma the
clear picture emerges. First, we find that the presence
of ethnic groups that are much poorer than the country
country average income level (that is, where one of a few
groups is comparably poor, and remaining groups as area whole increases the risk of territorial conflict. Since
relatively equal) entail larger risks than large positivemost of the very poor groups are quite small, demandin
deviations (where a small elite is wealthy and most otherself-determination is more viable than attempting to over
groups are equally poor). throw the ruling elite. Second, our results indicate tha
large discriminated groups boost the probability of gov
Additional sensitivity tests involved investigating a possi
ble interaction effect between economic and ethno-politi ernmental civil wars, in part because of the evident dis
cal marginalization (cf. Stewart 2008; Cederman et connectal. between demographic power and political
privileges. In contrast, conventional proxies for ethnic
2011), without finding consistent evidence that the effect
of either HI type is conditional on the other. Regressiondiversity and vertical economic inequality lose much of
models limited to the post-Cold War period also yield their
lar effect when horizontal inequalities are accounted
for. These findings are robust to a series of sensitivity
gely similar results. We estimated logistic and linear fixed
tests. Of course, it could be that further improved data
effects regression to correct for possible unobserved time
independent factors that might correlate with civiland warmore sophisticated indicators of individual-level
onset. Moreover, we explored the sensitivity of our inequality
find could lend support to conventional grievance
ings to changes in model specifications, sample inclusionarguments. However, our study casts doubt on the too
criteria, and outlier manipulation. These tests further often unreflective reliance on standard proxies, such as
increased our confidence in the importance of ethno ELF and the Gini coefficient, in the absence of explicit
political and economic grievances for understanding conflict mechanisms. It would seem that, despite their
where and when civil wars break out. popularity in the econometric literature, the arguments
that attempt to link ethnic diversity with the outbreak of
Conclusion civil war have been especially poorly articulated and insuf
ficiently anchored in specific conflict-inducing mecha
Despite widespread agreement among practitioners nisms. and
laymen that material and political inequalities matter The findings
for presented here are of significant policy
popular unrest and civil war, several prominentrelevance, scholarsfor at least two reasons. First, our group-based
in the empirical civil war literature dismiss this indices
link of
byhorizontal inequalities make us better able to
referring to the alleged ubiquity of grievancespredict that hin
the locus and timing of future civil wars than do
ders separating between cases of peace and war. conventional
Our measures of ethnic fractionalization and
findings suggest that there are good reasons to income be skepti disparity (for example, Goldstone et al. 2010
cal of this claim. Theoretical misspecification and, conse the inherently dynamic feature of ethno-po
particular,
quently, poor validity of applied measurements calare
discrimination
a in many countries demonstrates a
major reason why earlier research has failed to tematic
detect pattern of covariation with civil war outbr
robust relationships between measures of societal Second, our results verify earlier findings on the im
inequalities and civil war. Instead of considering interpertance of ethno-nationalist politics for translating socie
sonal differences in opportunities and privileges as theinequalities into political violence. This perspective
main causes of grievance in a society, we have shown that us that conflicts will remain extremely difficult to res
political and socioeconomic disparities increase the riskand, if resolved, are likely to recur, as long as the u
of civil war primarily when they overlap with ethnic lying problems of political exclusion or horizontal
cleavages. nomic inequality continue to fester. Thus, including
While several recent disaggregated studies have beenempowering previously discriminated populations,
able to tease out such effects at the subnational level, this reducing inequality along ethnic lines through a f
manuscript is the first to propose global country-leveldistribution of public goods and privileges are more lik
measures of both economic and political horizontal to promote peace and stability than short-sighted
inequality that allow us to compare the effect of such attempts to "strengthen the state" by supporting illegiti
mechanisms to standard indicators of vertical inequality mate and exclusionary regimes in the name of "stability."
on non-ethnic as well as ethnic wars. As a way to over
come the information loss associated with aggregation References
from substate- to state-level analysis, our new indicators
operate according to the principle of the weakest link: Acemoglu, Daron, and James A. Robinson. (2006) Economic Origins o
Which parts of the chain are most likely to trigger civil Dictatorship and Democracy, New York: Cambridge University Press.
Alesina, Alberto, and Eliana La Ferrara (2005) Ethnic Diversity and
war onset? Thus, we operationalize ethno-political griev
Economic Performance. Journal of Economic Literature43(3): 762-800.
ances in terms of the size of the largest discriminated Alesina, Alberto, Arnaud Devleeschauwer, William Easterly,
group within a country rather than considering the total Sergio Kurlat, and Romain Wacziarg (2003) Fractionalization.
excluded population. Furthermore, we measure eco Journal of Economic Growth 8(2): 155-194.
Anderson, Benedict (1991) Imagined Communities: Reflections on the
Origin and Spread of Nationalism. London: Verso.
26 In other tests not shown, we considered Baldwin and Huber's (2010)
Baldwin, Kate, and John D. Huber (2010) Economic versus Cultural
between-group inequality (BGI) data and 0stby's (2008) horizontal economic Differences: Forms of Ethnic Diversity and Public Goods Provision.
inequality measure—both generated from various Demographic and Health American Political Science Review 104(4): 644-662.
Surveys (DHS). However, the limited, non-random coverage of those data
Beck, Nathaniel, Jonathan N. Katz, and Richard Tucker (1998)
implies that the results would not be directly comparable to the reported find
Taking Time Seriously: Time-Series—Cross-Section Analysis with a
ings and hardly generalizable to the universe of cases.

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
430 Inequalities, Grievances, and Civil War

Binary Dependent Variable. American Journal Gellner, Ernestof(1964) Thought Science

Political and Change. London: Widenfeld &
42(4): 1260-1288. Nicolson.

Boix, Carles (2008) Economic Roots of Civil Wars and Revolutions in Gellner, Ernest (1983) Nations and Nationalism. Ithaca, NY: Cornell
the Contemporary World. World Politics 60(3): 390-437. University Press.
Booth, John A. (1991) Socioeconomic and Political Roots of National Gleditsch, Kristian Skrede, and Michael D. Ward (1999) A Revised
Revolts in Central America. Latin American Research Review 26(1): 33 List of Independent States since 1816. International Interactions 25
-73. (4): 393-413.
Boswell, Terry, and William J. Dixon (1993) Marx's Theory of Gleditsch, Nils Petter, Peter Wallensteen, Mikael Eriksson,
Rebellion: A Cross-Nation Analysis of Class Exploitation, Economic Margareta Sollenberg, and Havard Strand (2002) Armed
Development, and Violent Revolt. American Sociological Review 58(5): Conflict 1946-2001: A New Dataset. Journal of Peace Research 39(5):
681-702. 615-637.

Brubaker, Rogers (1996) Nationalism Reframed: Nationhood and Goldstone,

the Jack A. (2001) Toward a Fourth Generation of
National Question in the New Europe. Cambridge: Cambridge Revolutionary Theory. Annual Review of Political Science 4: 139-187.
University Press. Goldstone, Jack A., Robert H. Bates, David L. Epstein, Ted
Buhaug, Halvard (2006) Relative Capability and Rebel Objective in Robert Gurr, Michael B. Lustik, Monty G. Marshall, Jay
Civil War. Journal of Peace Research 43(6): 691—708. Ulfelder, and Mark Woodward (2010) A Global Model for
Buhaug, Halvard (2010) Dude, Where's My Conflict? Conflict Forecasting Political Instability. American Journal of Political Science
Management and Peace Science 27(2): 107-128. 54(1): 190-208.
Buhaug, Halvard, and Jan Ketil R0d (2006) Local Determinants of Gurr, Ted Robert (1970) Why Men Rebel. Princeton, NJ: Princeton
African Civil Wars, 1970-2001. Political Geography 25(3): 315-335. University Press.
Buhaug, Halvard, Lars-Erik Cederman, and Jan Ketil R0d (2008) Gurr, Ted Robert (1993) Minorities at Risk: A Global View of Ethnopolitical
Disaggregating Ethno-Nationalist Civil Wars: A Dyadic Test of Conflicts. Washington, DC: United States Institute of Peace Press.
Exclusion Theory. International Organization 62(3): 531-551. Hale, Henry E., and Rein Taagepera (2002) Russia: Consolidation or
Buhaug, Halvard, Kristian Skrede Gleditsch, Helge Holtermann, Collapse? Europe-Asia Studies 54(7): 1101-1125.
Gudrun 0stby, and Andreas For0 Tollefsen (2011) It's the Hegre, Havard, and Nicholas Sambanis (2006) Sensitivity Analysis of
Local Economy, Stupid! Geographic Wealth Dispersion and Empirical Results on Civil War Onset. Journal of Conflict Resolution
Conflict Outbreak Location. Journal of Conflict Resolution 55(5): 814— 50(4): 508-535.
840. Hegre, Havard, Gudrun 0stby, and Clionadh Raleigh. (2009)
Butler, Christopher, and Scott Gates (2009) Asymmetry, Parity, and Poverty and Civil War Events. A Disaggregated Study of Liberia.
(Civil) War: Can International Theories of Power Help Us Journal of Conflict Resolution 53(4): 598-623.
Understand Civil War? International Interactions 35(3): 330-340. Hegre, Havard, Joakim Karlsen, Havard Mokleiv Nygard, Henrik
Cederman, Lars-Erik, and Luc Girardin (2007) Beyond Urdal, and Havard Strand. (2013) Predicting Armed Conflict,
Fractionalization: Mapping Ethnicity onto Nationalist Insurgencies. 2011-2050. International Studies Quarterly, forthcoming.
American Political Science Review 101(1): 173-185. Heston, Alan, Robert Summers, and Bettina Aten. (2009) Penn
Cederman, Lars-Erik, and Kristian Skrede Gleditsch (2009) World Table Version 6.3. University of Pennsylvania. Available at
http://pwt.econ.upenn.edu/. Accessed June 10, 2011.
Introduction to Special Issue on 'Disaggregating Civil War'. Journal
of Conflict Resolution 53(4): 487-495. Horowitz, Donald L. (1985) Ethnic Groups in Conflict. Berkeley, CA:
Cederman, Lars-Erik, Andreas Wimmer, and Brian Min (2010)University
Why of California Press.
Hosmer,
Do Ethnic Groups Rebel? New Data and Analysis. World Politics 62 David W., and Stanley Lemeshow (2000) Applied Logistic
(1): 87-119. Regression,2nd edn. Hoboken, NJ: Wiley.
Cederman, Lars-Erik, Nils B. Weidmann, and Kristian Skrede Huntington, Samuel P. (1968) Political Order in Changing Society. New
Gleditsch (2011) Horizontal Inequalities and Ethnonationalist Haven, CT: Yale University Press.
Civil War: A Global Comparison. American Political Science Review 105 Jarstad, Anna K., and Desiree Nilsson (2008) From Words to Deeds
(3): 478-495. The Implementation of Power-Sharing Pacts in Peace Accords.
Collier, Paul, and Anke Hoeffler (2004) Greed and Grievance in Conflict Management and Peace Science 25(3): 206-233.
Civil War. Oxford Economic Papers 56(4): 563-595. Kalyvas, Stathis N. (2006) The Logic of Violence in Civil War. New York
Combes, Pierre-Philippe, Miren Lafourcade, Jacques-Francois Cambridge University Press.
Thisse, and Jean-Claude Toutain (2011) The Rise and Fall of Kanbur, Ravi, Prem Kumar Rajaram, and Ashutoshi Varshney
Spatial Inequalities in France: A Long-run Perspective. Explorations (2010) Ethnic Diversity and Ethnic Strife: An Interdisciplinary
in Economic History 48(2): 243-271. Perspective. World Development 39(2): 147-158.
Cunningham, David E., Kristian Skrede Gleditsch, and Idean Laitin, David D. (2007) Nations, States and Violence. Oxford: Oxford
Salehyan (2009) It Takes Two: A Dyadic Analysis of Civil University
War Press.
Lang,
Duration and Outcome. Journal of Conflict Resolution 53(4): Nicholas R. (1975) The Dialectics of Decentralization: Economic
570-597.
Davies, James Chowning (1962) Toward a Theory of Revolution. Reform and Regional Inequality in Yugoslavia. World Politics 27(3):
American Sociological Review 6(1): 5-19. 309-335.

Easterly, William, and Ross Levine (1997) Africa's Growth Tragedy:Lichbach, Mark I. (1989) An Evaluation of 'Does Economic Inequality
Policies and Ethnic Divisions. Quarterly foumal of Economics 112(4): Breed Political Conflict' Studies. World Politics 41(4): 431-470.
1203-1250. Montalvo, Jose G., and Marta Reynal-Querol (2005) Ethnic
Emirbayer, Mustafa, and Chad Alan Goldberg (2005) Pragmatism, Polarization, Potential Conflict, and Civil Wars. American Economic
Bourdieu, and Collective Emotions in Contentious Politics. Theory Review 95(3): 796-816.
and Society 34(5/6): 469-518. Moore, BarringtonJr (1966) Social Origins of Dictatorship and Democracy.
Fearon, James D. (2003) Ethnic and Cultural Diversity by Country. Boston: Beacon Press.

Journal of Economic Growth 8(2): 195-222. Muller, Edward (1972) A Test of a Partial Theory of Potential
Fearon, James D. (2010) Governance and Civil War Onset. Background paper Political Violence. American Political Science Review 66(3): 928-959.
to the World Development Report 2011. Washington, DC: World Bank. Muller, Edward N., and Mitchell A. Seligson (1987) Inequality
Fearon, James D., and David D. Laitin (2003) Ethnicity, Insurgency, Insurgency. American Political Science Review 81(2): 425-451.
and Civil War. American Political Science Review 97(1): 75-90. Murshed, S. Mansoob, and Scott Gates (2005) Spatial-Horizontal
Forsberg, Erika (2008) Polarization and Ethnic Conflict in a Widened Inequality and the Maoist Insurgency in Nepal. Review of
Strategic Setting. Journal of Peace Research 45(2): 283-300. Development Economics 9(1): 121-134.
Gamson, William (1992) Talking Politics. New York: Cambridge Nordhaus, William D. (2006) Geography and Macroeconomics: New
University Press. Data and New Findings. Proceedings of the National Academy of Sciences
Gates, Scott, Havard Hegre, Mark Jones, and Havard Strand (2006) of the United States of America 103(10): 3510-3517.
Institutional Inconsistency and Political Instability: Polity Duration, Oberschall, Anthony (1978) Theories of Social Conflict. Annual
1800-2000. American Journal ofPolitical Science 50 (4): 893-908. Review of Sociology 4: 291-315.

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms
 Halyard Buhaug, Lars-Erik Cederman and Kristian Skrede Gleditsch 431

Tajfel, Henri, and

0stby, Gudrun (2008) Polarization, Horizontal Inequalities and John C. Turner (1979) An Integrative Theory of
Violent
Civil Conflict. Journal of Peace Research 45(2): 143-162.
Intergroup Conflict. In The Social Psychology of Intergroup Relations,
0stby, Gudrun, Ragnhild Nordas, and Jan Ketil R0d (2009) edited by William G. Austin and Stephen Worchel. Monterey, CA:
Regional Inequalities and Civil Conflict in Sub-Saharan Africa. Brooks Cole.

International Studies Quarterly 53(2): 301-324. Tarrow, Sidney (2007) Inside Insurgencies: Politics and Violence in an
Rabushka, Alvin, and Kenneth A. Shepsle (1972) Politics in Plural Age of Civil War. Perspectives on Politics 5(3): 587-600.
Themner, Lotta, and Peter Wallensteen (2011) Armed Conflict,
Societies: A Theory of Democratic Instability. Columbus, OH: Charles.
E. Merrill. 1946-2010. Journal of Peace Research 48(4): 525-536.
Regan, Patrick M., and Daniel Norton (2005) Greed, Grievance, and Tilly, Charles (1978) From Mobilization to Revolution. New York:
Mobilization in Civil Wars. Journal of Conflict Resolution 49(3): 319 McGraw-Hill.
336. Tilly, Charles (1999) Durable Inequality. Berkeley and Los Angeles:
Russett, Bruce M. (1964) Inequality and Instability: The Relation of University of California Press.
Land Tenure to Politics. World Politics 16(3): 442-454. Tilly, Charles (2007) Poverty and the Politics of Exclusion. In Moving out
Sambanis, Nicholas (2001) Do Ethnic and Nonethnic Civil Wars Have of Poverty: Cross-Disciplinary Perspectives on Mobility, edited by Deepa
the Same Causes? A Theoretical and Empirical Inquiry. Journal of Narayan and Patti Petesch. Washington DC: The World Bank.
Conflict Resolution 45(3): 259-282. Vanhanen, Tatu (1999) Domestic Ethnic Conflict and Ethnic Nepotism:
Sambanis, Nicholas (2004) Using Case Studies to Expand Economic A Comparative Analysis. Journal of Peace Research 36(1): 55-73.
Models of Civil War. Perspectives on Politics 2(2): 259-279. Ward, Michael D., Brian Greenhill, and Kristin M. Bakke (2010)
Scott, James C. (1976) The Moral Economy of the Peasant: Rebellion and The Perils of Policy by P-value: Predicting Civil Conflicts. Journal of
Subsistence in Southeast Asia. New Haven, CT: Yale University Press. Peace Research 47(5): 363-375.
Weidmann, Nils B., Doreen Kuse, and Kristian Skrede Gleditsch
Skocpol, Theda (1979) States and Social Revolutions: A Comparative Analysis
of France, Russia, and China. Cambridge: Cambridge University Press. (2010) The Geography of the International System: The CShapes
Stewart, Frances (ed.) (2008) Horizontal Inequalities and Conflict: Dataset. International Interactions 36(1): 86-106.
Williams, Robin M. Jr (2003) The Wars Within: Peoples and States in
Understanding Group Violence in Multiethnic Societies. Basingstoke:
Palgrave Macmillan. Conflict. Ithaca, NY: Cornell University Press.
Stewart, Frances, and Arnim Langer (2008) Horizontal Inequalities:Wimmer, Andreas, Lars-Erik Cederman, and Brian Min (2009) Ethnic
Explaining Persistence and Change. In Horizontal Inequalities and Politics and Armed Conflict: A Configurational Analysis of a New
Conflict: Understanding Group Violence in Multiethnic Societies, edited by Global Data Set. American Sociological Review 74(2): 316-337.
Frances Stewart. Houndmills: Palgrave Macmillan. WUCHERPFENNIG, JULIAN, NlLS B. WEIDMANN, LUC glrardin, LARS-ERIK
Stewart, Frances, Graham Brown, and Luca Mancini. (2005) Why Cederman, and Andreas Wimmer (2011) Politically Relevant
Horizontal Inequalities Matter: Some Implications for Ethnic Groups across Space and Time: Introducing the GeoEPR
Measurement. CRISE Working Paper 19. Dataset. Conflict Management and Peace Science 28(5): 423-437.

This content downloaded from

128.122.149.92 on Fri, 28 Jun 2024 15:38:44 +00:00
All use subject to https://about.jstor.org/terms