Reynal-Querol (2002) Ethnicity, Political Systems, and Civil Wars

Ethnicity, Political Systems, and Civil Wars
Author(s): Marta Reynal-Querol

Source: The Journal of Conflict Resolution, Vol. 46, No. 1, Understanding Civil War (Feb.,
2002), pp. 29-54
Published by: Sage Publications, Inc.
Stable URL: http://www.jstor.org/stable/3176238 .
Accessed: 03/09/2013 06:56
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
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 support@jstor.org.
Sage Publications, Inc. is collaborating with JSTOR to digitize, preserve and extend access to The Journal of
Conflict Resolution.
http://www.jstor.org
This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

All use subject to JSTOR Terms and Conditions
Ethnicity,PoliticalSystems,and Civil Wars
MARTAREYNAL-QUEROL
Institutd'Analisis Economic (IAE), CSIC
Theeffect of ethnicdivisionon civil warandthe roleof politicalsystems in preventingthese conflicts are

analyzed, using the importanceof religious polarizationand animist diversityto explain the incidence of
ethniccivil war.Findingsshowthatreligiousdifferencesarea social cleavagemoreimportantthanlinguistic
differencesin the developmentof civil war,andbeing a consociationaldemocracysignificantlyreducesthe
incidence of ethnic civil war.
Social and ethnic conflicts are recurrentphenomenaaffecting many countries.Eth-

nicity is at the centerof politics in dividedsocieties. The most importanttensionsin the
worldcould be found,amongothers,in Lebanon,Israel,Algeria,Nigeria,Sudan,Ethi-
opia, India, and Yugoslaviaduringthe 1990s.
In the study of the social andpolitical causes of civil war,a crucialissue is the dis-
tinctionbetween ethnic andrevolutionarycivil wars. In this study,we concentrateon
ethnic civil warsandarguethatthe maincauses of them aresocial andpoliticalcharac-
teristics. First, we claim thatreligiously divided societies are more prone to intense
conflict than countries where people have conflicting claims to resources based on
interestgroups or languagedivisions. This is because religious identity is fixed and
nonnegotiable.Disputesamongidentitygroupsbasedon theirreligiousnaturearepar-
ticularlydifficult to negotiate,raisingthe odds of violence.
Second, we use indices of polarizationbased on rent-seekingmodels of conflict,
instead of traditionalfragmentationindices, to capturethe level of religious conflict.
We arguethatpolarizationindices aremoresuitableto proxylatentreligiousconflicts.
We show empiricallythat religious polarizationand animist diversity are the most
importantfactors to explain the incidence of ethnic civil war. Therefore,the results
suggest thatreligiousdivisionsaremoreimportantthanlanguagedivisionsandnatural
resourcesto explain social ethnic conflicts.
However,not all ethnicallydivided societies evolve into civil war.There are also
experiences of good relationshipsamong individuals of different cultures within a
country.Forinstance,afterthe violent riots of 1969, therehas been a reasonablerela-
tionshipbetween ChineseandMuslimsin Malaysia.Althoughthereis a latenttension
AUTHOR'S NOTE: I would like to thankJosep MariaColomer,Joan M. Esteban,and Jose Garcia-

Montalvo and two anonymousreferees for many valuablecomments. Financialsupportis acknowledged
from the SpanishMinisteriode Educaciony Cultura(GrantSEC1034-2000), the EuropeanCommission
ResearchDG programHPCFCT2000-00211,andthe EuropeanUnion projecton Polarizationand Conflict
HPHA-CT2000-00052.
OFCONFLICT
JOURNAL Vol.46 No. 1,February
RESOLUTION, 200229-54
? 2002 Sage Publications
29

30 JOURNAL RESOLUTION
OF CONFLICT
between these two communities, the proportionalsystem introducedin 1970 has

helped to avoid more violent riots. This suggests that violent conflicts can be attenu-
ated by alternativeinstitutionalincentives.
The question of the best political environmentto reduce the likelihood of armed
conflicts in potentiallyconflictive societies is not a new issue. However,the solution
proposed on many occasions is to enhance political rights and civil liberties, even
thoughthe empiricalevidence thatsupportsthis is weak. We show thatthe combina-
tion of political system and democracylevel has a high explanatorypower over the
probabilityof ethniccivil war.Moreover,we findempiricallythatthe establishmentof
consociationaldemocracies-proportionalrepresentationsystems thatproducecoali-
tion politics-can preventethniccivil warsgeneratedbecauseof religiousdifferences.
This article is divided into six sections. The first is this introduction.The second
section presentsa brief review of the literatureon the causes of civil wars. The third
section concentrateson the social and political causes of ethnic conflicts. The fourth
section presentsan empiricalinvestigationof the causes of ethnic civil war.The final
section concludes the study.
THE CAUSES OF CIVIL WARS
Studying social conflict has been considered an issue of political science for
decades. Yet we are recently observing an increasingnumberof contributionsfrom
economics. Partof this literaturehas studiedthe effect of economic and social factors
on the probabilityof civil war. Collier and Hoeffler (1998) investigatethe generic
causes of civil wars with special emphasis on economic factors. Using the index of
ethnolinguisticfragmentation(ELF),they find thatmore fragmentedsocieties arenot
moreproneto civil warthanthe rest,butthe dangerof civil warincreaseswhen society
achievemidlevel valuesof the indexELF.However,afterthose initialfindings,Collier
and Hoeffler (2000) arguethatconflicts arefar more likely to be causedby economic
opportunitiesthanby grievance.However,the studyof the causes of civil wardeserves
additionalstudies,especially on the ethnic and political issues thatmay influence the
incidence of civil war.Ouranalysis,based on these previousstudies,is focused on the
ethnic andpoliticalcauses of civil war,using a theoreticallybased index of latentcon-
flict and giving special importanceto the religious dimensions of ethnicity,using a
new dataset thattriesto overcomethe commoncriticismof the WorldChristianEncy-
clopedia (Barret1982) data.Moreover,most of the literaturehas considereddemoc-
racy the only political variablethat may affect the probabilityof civil war.Here we
arguethatthe level of inclusivenessof the system, togetherwith the level of democ-
racy, matters. These results clarify the role of democracy itself. Elbadawi (1999),
based on the analysisof the resultsof some comparativestaticexperiments,finds that
ethnically polarized societies have a higher risk of suffering a civil war. Ellingsen
(2000) finds that the differentaspects of multiethnicity(the size of the largestgroup,
the numberof groups,the size of the minoritygroup,and ethnic affinities)are impor-
tant in explaining domestic conflicts. Her results give an importantrole to political
regimes and socioeconomic variablesto reducethe level of conflict.

Reynal-Querol/ETHNICITY,POLITICALSYSTEMS,AND WARS 31
However,not all civil warsareof the same nature.Forthis reason,it makessense to

separatethe study of their determinantsas functions of the type of war. Recently,
Sambanis (2001) analyzed the causes of ethnic civil war, separatingthe ethnic war
from the revolutionary/ideologicwar. He finds that ethnic civil wars are predomi-
nantlydue to politicalratherthaneconomic grievances.Ouranalysisalso corroborates
the importanceof sociopolitical variablesin explaining the incidence of ethnic civil
war.Reynal-Querolanalyzes the political causes of revolutionarycivil war and finds
thata presidentialsystemwith a low level of democracyis the most importantpolitical
cause of an ideological civil war.Finally,the low level of inclusivenessof the political
system, representedby a presidentialsystem, explains why LatinAmericancountries
have a higherincidenceof revolutionarycivil wars.In this study,we concentrateon the
determinantsof ethnic conflicts.'
SOCIOPOLITICAL FACTORS AND ETHNIC CONFLICTS
THEROLEOF RELIGION,ETHNICITY,
ANDIDEOLOGY
Manyconflicts relatedto religiousdifferencesariseeveryday.Even thoughecono-

mists have not paidspecialattentionto the issue of religionto explaincivil wars,many
researchersin political science have emphasizedthe importanceof religious differ-
ences as a cause of ethnic conflicts.
The theory of Samuel Huntington(1996) about the culturalfragmentationof the
world initiateda large debate. Samuel Huntingtonproposeda model to interpretthe
new realityof the world,basedon the fact thatthe explanatoryfactorsarenot ideologi-
cal but cultural.Following Huntington,we have to grantreligions a fundamentalrole
in worldpolitics. In the moder world,religionis a centraland,in manysituations,the
primaryforce thatmotivatesandmoves humans.In such situations,whatcounts is not
political ideology or economic interests.Faith and family, blood and beliefs are the
aspectswith which people identifythemselves,the characteristicsfor which they fight
and die. People belongingto differentreligions have differentversions of many rela-
tions among individuals and authorities.Following Huntington,one of the most
importantcauses of futureconflict amongcivilizationsis thattheircharacteristicsand
differences are less mutableand, therefore,more difficult to reach agreementsand
solve thanpolitical and economic differences.More thanethnicity,religion discrimi-
nates and differentiateshumansin a sharpandexclusive way, even morethanbelong-
ing to a countrywould do. A personcan be half Frenchandhalf SaudiArabianand, at
the same time, be a citizen of bothcountries.However,it is difficultto be half Catholic
and half Muslim.
Following Horowitz(1985), in pluralsocieties in Asia, Africa, and the Caribbean,
partiestend to be organizedalong ethnic lines. Most of them in Africa belong to ani-
mist religions. In WesternEuropeand NorthAmerica,religion, social class, and lan-
1. The complete version of this article includes an examinationof the causes of revolutionarycivil
wars.

Lijphart (1984) sostiene que lo que mas diferencia entre los regimenes democraticos son los
aspectos socioeconomicos y los religiosos. Reynal Querol dice que las diferencias religiosas
son mas relevantes que las diferencis linguisticas como divisiones (cleavage) sociales que
pueden convertirse en conflcito debido a dos razones: 1. la religion es un factor de identidad
con un fuerte razgo excluyente (dificlmente se tenga mas de una religion); 2. la religion
32 JOURNAL
OFCONFLICT
RESOLUTION organiza una determina cosmovision, que difiere de otras religiones (2002, p. 32).
¿podria algo de esto estrapolarse al comportamiento electoral?
guage arethe basic dimensionsof a politicalparty.Lijphart(1984) found,in a sample

of 22 democraticregimes, thatthe two dimensionsthat most frequentlydifferentiate
systems are the socioeconomic and the religious.
These authors,among others,claim thatreligious differencesare more important
than language differencesas a social cleavage thatcan develop into a conflict. There
aretwo basic reasonswhy religiousdifferencescan generatemore violence thanother
social cleavages. First,thereis no doubtof the exclusivity of religion. One can speak
two or more languages,but you can have only one religion. Religion can be used as a
sign of identity,strongerthanlanguagein the sense thatyou exclude those from other
religions, whereas speakingtwo languages diffuses the division line among groups.
Second, religious differences,which are the basis of differencesamong civilizations,
imply differentways of understandingthe world,social relationships,andso on. Even
if differentgroupsspeakdifferentlanguages,they could sharethe same way of under-
standingthe worldandrelationshipsif they belong to the same civilization.However,
this is more difficult for people of differentreligions.
HOWTO CAPTURESOCIALCONFLICT
To show the importanceof ethnicdiversityin ethnic civil wars,at least two impor-
tantaspectshave to be addressed.First,it is necessaryto clarifythe conceptof ethnic-
ity to select the variablesthat captureethnic diversity and analyze which of these
dimensions of ethnicity are more importantto explain social conflict. Following
Horowitz(1985), the inclusiveconceptionof ethnicitycoversdifferencesidentifiedby
skin color, language,religion, or some otherattributeof common origin. Because of
datarestrictions,in this study we deal with religion and languagedifferences.More-
over,as shown in the previoussection,religionis the most importantethnicdimension
in explaining social conflict.
Second, we have to summarizethe informationfrom each of these dimensions in
one syntheticindex. The most importantissue is the appropriateprocedureto use to
summarizein an index the concept of social conflict. Is it social fragmentationor
social polarizationthatmakesethnictensionsstronger?Thereis no easy answerto this
question.First, we need to study the mechanismthroughwhich these groupsinteract
and analyzein which situationstensionsarisemoreeasily. Second, the threevariables
that define ethnicity (language,religion, and color) can work in differentways. For
instance,the tensions caused by languagedifferencesand the loss of communication
that they generate can emerge in a situation very different from those generatedby
religion.
The measuresusedregularlyin the empiricalliteratureto quantifyethniccharacter-
istics are fragmentationindices, even thoughthereis no theoreticalsupportfor these
kindsof indices. However,rent-seekingmodels suggest thatpolarizationmeasuresare
more appropriatethanfragmentationindices to capturesocial conflict. Any index of
polarizationpoints out thatthe situationthatleads to the point of maximumtension is
when there are two social groups with the same size. These kinds of measuresdiffer
from an index of fragmentationbecause the index of polarizationcapturesto what

extent it is the distributionof the groups from a bimodal distribution.However, the

fragmentationindex increasesmonotonicallywith diversity.
The commonly used measureof linguistic differencesis the so-called index of lin-
guistic fragmentationof TaylorandHudson(1972). In fact,this indexis consideredthe
only measureof ethnicdiversityby manyauthors.This indicatorcapturestheprobabil-
ity that two randomly selected individuals in a country will belong to different
ethnolinguisticgroups.However,we follow the literatureon rent seeking2that shows
thatsocial conflict is higherif the underlyingdistributionof the individualcharacteris-
tics is bimodal. We apply two simple indices of polarization:the one proposed by
Reynal-Querol(2001a) and an applicationof Estebanand Ray's (1994) well-known
index of polarization.
TheindexofpolarizationofReynal-Querol3(2001a). This indextakesthe following

form:
N (
IRC1= 1- (0.5- Ii)21Ci / .25, buscar cuando cambio, ya que
la division despues es de 0.5 y
i=1 se hace en ls parantesis.
where ni is the proportionof each religion and N is the numberof religions.

This index providesa rankingorderof the differentdistributionsof the population.
It is an index of polarizationwith the usualpropertiesof these indices' properties(see
the appendix).As a matterof fact, the propertiesof the index proposedabove aresimi-
lar to those in the measureof conflict in Estebanand Ray (1999).
Observethat IRC1 can be writtenas
N N
IRC1=X[N -4(/2-Ki_)2 i]= f(i).
i=l i=1
The behaviorof the index criticallydependson the propertiesof theffunction. By

differentiation,one can computef' andf":
f'= [-3t2 + 2i, -14]4 f" =6[ - i].
Theffunction reachesa minimumfor n = 1/6 anda maximumfor i = 1/2. It is con-

vex for X < 1/3 and concave for n > 1/3 (see Figure 1).
Understandingthe shapeof the function,it is crucialto understandthe propertiesof
the proposedpolarizationindex. The intuitionbehind this is clear and simple. If we
transferpopulationfromone groupto another,the effect on the conflict level is differ-
ent depending on the size of the groups. Imagine a populationcomposed of three
groups distributedin the following way (0.5, 0.25, 0.25). If we transferpopulation
from one small groupto the other,the conflict increases.We arein the concave partof
the function. However,if the distributionis (0.45, 0.45, 0.1), and we transferpopula-
tion from one big groupto the other,the conflict decreases.This is because we are in
2. A basic readingis Estebanand Ray (1999).

3. To check the polarizationpropertiesof the index, see Reynal-Querol(2000a).

34 JOURNALOF CONFLICTRESOLUTION
1/6 1/3 1/2
Figure 1
the convex region. Whatis the intuitionbehindthis result?In the first case, even that
transferimplies thatthe distributionis more unequalin the new situation:one of the
small groups is largerin respect to the big group, which means thatwe are closer to
polarization.In the secondcase, the transferimplies thatone of the big groupsbecame
smaller,and thereforethe new situationis less polarized.Notice thatthe resultsimply
thatthis index does not satisfy the propertiesof the Lorenzcurveaboutconcavity.In a
Lorenzcurve,this effect of moving people between small or big groupsis the same. It
is importantto notice anotherdifferencecomparedwith the Lorenzcurvedominance:
our measureis global andthe Lorenzcurveis not. Althoughthe Lorenzcriteriaestab-
lish the impacton inequalityof a local transferindependentlyof the shapeof the restof
the distribution,in our case, the effect on polarizationof the transferof population
from one groupto anothercannotbe establishedwithoutknowingthe entiredistribu-
tion. This is a propertythatalso has the measureof polarizationproposedby Esteban
and Ray (1994).
Estebanand Ray (1994) indexof polarization.To show thatthe resultsobtainedin

the regression analysis are robust to other polarizationmeasures, we compare the
index IRC1to an application4of religionsto an existing measureproposedby Esteban
and Ray (1994).5
P*(n)=KE E IC
l+a~ jd=IRC2,
i=l j=1
where d = 1 if i j and d = 0 if i =j.

4. The original polarization measure proposed by Esteban and Ray (1994) isP *(I,Y)=
1 +a -
KE n=l j Yiy Yjlfor some constantk > Oand a E (0, a*], where a* - 1.6, yi is income per
capita, and ni is the relativesize of the group.
5. The religiousdistributionof a society satisfiesthe featurethatsays any distributionshouldbe satis-
fied to apply the index.

POLITICAL
/ETHNICITY,
Reynal-Querol ANDWARS 35
SYSTEMS,
Using this application, the index can be written in the following way:
1
IRC2 = n + ", which is a strict convex function for a > 1. For each possible a,
thereis a differentpolarizationfunction.In practice,we will choose a = 1.5.6The fact
thata is not determinedfromthe theoreticalmodel makesthe use of this secondpolar-
ization index just a matterof comparisonwith respect to the Reynal-Querol(2001a)
index.7The majordifferencebetweenthese two indices is in the form of the function;
whereasIRC1 has a convex andconcavepartas explainedabove,IRC2 is strictlycon-
vex for a > 1.
THEROLEOF POLITICALINSTITUTIONS
This study also analyzes the political characteristicsthat can preventor promote
civil wars. The literaturehas consideredonly repressionas a political characteristic
thatcan affect notonly economic developmentbutalso civil wars.CollierandHoeffler
(2000), Sambanis(2001), Hegreet al. (2001), andEllingsen (2000) find thatmidlevel
democraciesare more proneto civil war than high-level democraciesand high-level
autocracies.Reynal-Querol(2001b) shows theoreticallyand empiricallyhow coun-
tries with alternativepolitical systems have differentprobabilitiesof experiencinga
civil war.She develops a simple theoreticalmodel thatcapturesthe basic relationship
between the political system and rebellion.Accordingto the model, the proportional
system turnsout to have a lower probabilityof rebellionthanthe majoritariansystem.
The intuitionbehindthis resultis thatin the proportionalsystem, the opportunitycost
of rebellion is higher than in a majoritariansystem. The main idea is that the more
inclusive the political system, the higherthe opportunitycost of rebellion,and there-
fore the lower the probabilityof rebellion.Fromthe empiricalanalysis,she shows how
the structureof a politicalsystem is an importantmechanismthatcan affectthe proba-
bility of civil war in a democraticsystem. The result clarifies the role of democracy
itself. She observesthatsome countrieswithhigh levels of democracysufferperiodsof
violence; therefore,havinghigh levels of civil libertiesand freedomdoes not seem to
protectthem againstviolence. She arguesthatthe representationsystem of the voters
in governmentis moreimportantthanthe level of democracyper se. Empirically,she
finds thatthe more inclusive the system, the smallerthe probabilityof civil war.
Moreover,the level of repressionalso influences the opportunitycost of rebellion.
The morepoliticalrightsandcivil libertiesthe countryhas, the higherthe opportunity
cost of rebellion,andthereforethe lowerthe probabilityof groupsto rebel.However,a
little freedom is needed to let groups organize.This is the intuitionthatjustifies the
findings thatmidlevel democraciesare more proneto suffercivil wars.
6. This value comes to be in the rangeof feasible values for a.

7. However,it would be interestingto analyzeempiricallywhetherthe argumentsaboutpolarization
can be appliedto language groups.Because of datarestrictions,this will be done in futureresearch.

36 OFCONFLICT
JOURNAL RESOLUTION
ETHNICDIVERSITYANDPOLITICALSYSTEMS:
PLURALITYVERSUSCONSOCIATIONALISM
Ethnicconflict in pluralsocieties andhow to controlit often havebeen examinedin

the political science literature.However,there is no agreementabout which system
promotesor reducesthis conflict in a potentiallyconflictive society. Not all ethnically
fragmentedsocieties evolve into violence. Therefore,the questionis whetherthe dem-
ocratic system can affect social behaviorin heterogeneoussocieties.
There are two main theories in the literature.The first, defended by Horowitz
(1985), arguesthatin pluralsocieties, the majoritariansystemis betterthanthe propor-
tional system because it motivatesthe creationof coalitions among minoritiesor the
smallestgroup.On the otherhand,Lijphart(1984) arguesthatin suchpluralsocieties,
the proportionalsystemis betterthanthe majoritariansystembecauseit allows the for-
mation of segmented parties, and it does not artificiallyforce the establishmentof
largerbut less representativeparties.8
Empiricalwork on the importanceof political systems in explainingcivil wars is
scarce, and, as pointedout before, thereis a puzzle in the political science literature
about which system is betterto implementin ethnically divided societies to reduce
social conflict. For this reason, in this article,we also consider the representationof
voters,capturedby the political system, as a factorthatcan reducethe effect of social
cleavages in promotingcivil wars.We shouldmentionthe endogeneityproblemof the
political systems thatcan exist when analyzingthis problem.It may be thatcountries
wherethe proportionalsystem is able to persistfor any lengthof time will tendto have
a low level of polarization.However,some countrieswith a high level of polarization
implementproportionalsystems afterperiodsof violence and achieve peaceful peri-
ods thatdo not breakdown, as in the case of Malaysia or South Africa. However,we
will controlfor this problemin the empiricalanalysis,introducingan interactionterm.
Therefore,if it is the case thatcountrieswith low polarizationtend to implementpro-
portionalsystems, then the interactiontermswill be 0.
EMPIRICAL ANALYSIS
THEDATA
For the purpose of the empirical investigation,and given the determinantsdis-

cussed earlier,we need to obtaindataon civil warsandtheirtype, level of democracy,
political systems, naturalresources,level of development,linguistic fragmentation,
and religiouspolarization.Forthe definitionof civil war,we use datafrom Doyle and
Sambanis (2000). This definition is nearly identical to the definition of Singer and
Small (1994; Small and Singer 1982) and Licklider(1993, 1995). Data on the type of
8. Moreover,he arguesthat"ina politicalsystemwith clearlyseparateandpotentiallyhostile popula-

tion segments, virtuallyall decisions areperceivedas entailinghigh stakes,and strictmajorityrule places a
strainon the unity and peace of the system"(Lijphart1984, 28).

civil warcome fromthe statefailuredataset.9An ethnic waris definedas an episode of

violent conflict betweengovernmentsandnational,ethnic,religious, or othercommu-
nal minorities (ethnic challengers) in which the challengers seek major changes in
their status.
Thereare differentsourcesof dataon the level of democracy.The FreedomHouse
datasource,usuallyreferredto as Gastil'sindex of democracy,has beenthe most com-
monly used amongeconomists.The disadvantageof this sourceis thatit does not pro-
vide data before the 1970s. A more recent work on levels of democracydata is pro-
vided by the Polity IIIproject.10Even thoughthe criteriafor the constructionof these
datasets aredifferent,they look very similar,andthe correlationamongthem is about
0.9.
To capturethe characteristicsof the political system, we use as the basic source of
informationdatain Colomer(2000). He takesdatafor 123 attemptsat democratization
and majordemocraticinstitutionalchanges in 84 countrieswith more than 1 million
inhabitantsduringthe 125 year-periodfrom 1894 to 1999. He distinguishesthe fol-
lowing categories of democraticinstitutionalformulas:parliamentary-majoritarian,
presidentialandsemipresidential,andparliamentary-proportional representation.The
countriesincludedarethe ones consideredto be free by the FreedomHouse database.
Using this study, we generatea time series of cross sections for 138 countriesfrom
1960 to 1995, organizedin 5-year periods.1 We capturethe democraticrule of the
countriesat the beginningof each periodusing these data.Moreover,for the nonfree
countries,we used datafrom the FreedomHouse and Polity III datasources. There-
fore, we obtain five possible categories that define five dummy variables:not free,
partially free, parliamentary-majoritarian, presidential and semipresidential, and
parliamentary-proportional.
To proxy the loot of rebellion,we use, as do Collier and Hoeffler (2000), the share
of naturalresourcesexports in gross domestic product(GDP). The data on primary
exportscome from the WorldDevelopmentIndicators(WDI). To proxy the opportu-
nity cost, we take data on income per capita from the Penn WorldTable (PWT56)
(SummersandHeston 1991). The educationdatacome fromBarroandLee (1996) and
representsthe averageyearsof schoolingin the totalpopulation.The dataon linguistic
fragmentationcome fromthe index of TaylorandHudson(1972), andthe dataon reli-
gious fragmentationcome fromBarro(1997), who uses the same index as the linguis-
tic fragmentationbut with religious data.12
To capturereligious polarization,we use data from Reynal-Querol(2001a). The
dataareconstructedusingessentiallythe informationcontainedin L'Etatdes religions
dans le monde, which takes partof the data from the WorldChristianEncyclopedia
(Barret1982) butuses also nationalsourcesand TheStatesmanYear'sBookof 1987.'3
In some countries,there may be conflict inside a religious group. We consider three
9. Data and definitionsof ethnic and revolutionarycivil war are availableat http://www.bsos.umd.
edu/cidcm/dtfail/.
10. See http://www.colorado.edu/IBS/GAD/spacetime/data/Polity.htm.
11. These data are summarizedin Reynal-Querol(2001a).
12. The inclusionof othervariablessuch as growthrateor indices of income inequalitydo not alterthe
main results we find.
13. For a complete descriptionof the data,see Reynal-Querol(2001a).

38 JOURNAL
OFCONFLICT
RESOLUTION
cases: the animists, the Christians,and the Muslim groups.'4The index of animist
diversity(AD) capturesthe numberof followers of animistcults in each country.'5To
avoid the level effect generatedby this variable,the regressionsthatincludethis index
consider also as an explanatoryvariablethe total populationof the country.
Therearesome differencesbetweenthe dataon religionsused by Barro(1997) and
Reynal-Querol(2000a). Barro'sdataset, whichcomes fromthe WorldChristianEncy-
clopedia (Barret1982), has the advantageof being a time series, providinginforma-
tion for 1970, 1975, and 1980. However,this source has several shortcomings.First,
and probablythe most important,the data do not consider the possibility of double
practice, very common in sub-SaharanAfrica and Latin America countries. Com-
paringthisto the othersourcesof information,we realizethe dataarebiasedtowardthe
Christianreligion. A clear exampleis the case of Zaire,wherethe distributionof reli-
gions is consideredto be similarto Spainor Italy.The distributionof religious groups
between 1970 and 1980 does not change in many countries.There are only approxi-
mately 17 countriesthat recordchanges in proportions.But those changes occur in
countrieswherethereis doublepractice,andtheyusuallyimply an increasein the per-
centage of Christiansand a reduction in the size of animist followers. Moreover,
Barro'sdatado not representwith sufficientdetail all the religions. We have observed
in the WorldChristianEncyclopedia(Barret1982) thatthereis too muchinertiain the
growthrate of some religions, mainly Catholicism.
These shortcomingsmakes the use of Barro's(1997) data somehow problematic.
However,the literatureused Barro'sdata set to compute measuresof religious frag-
mentation. Therefore, we also consider Barro's data when comparing the results.
Because of these shortcomings,we use datain Reynal-Querol(200 la) to constructthe
index of religious polarization.
We wanted to use alternativeethnic variables, as in Vanhaven(1999), mainly
because they capturethe threedimensionsof ethnicity.However,therearetwo impor-
tantshortcomingsin the use of these data.First,theydo not differentiateamongChris-
tians and animist cults, which are very typical in African countries. Second, even
thoughhe reportsthe percentageof the biggest ethnicgroup,he does not providedata
on the size of the othersmall groups,which makesthe constructionof a polarization
measureimpossible.
Another source of data we would like to use is Ellingsen (2000). However, like
Vanhaven(1999), she reportsthe percentageof the biggest ethnic groupbutdoes not
providedataon the size of the othersmallergroups,which makesthe constructionof a
polarizationmeasureimpossible.
14. We do not considerotherreligions because they do not have the featuresthat make these three
groupsinternallyproblematic.Forthe Muslimsandthe Christians,we applythe same measureof polariza-
tion, IRC1.We use the proportionsof differentkind of Christiansfor the polarizationmeasureof Christians
anddataon the votes forMuslimpoliticalpartiesto calculatea polarizationmeasureforthe Muslimgroup.
15. Theanimistgroupsaredifferent.Theyincludea variablenumberof traditionalreligionstypicalof
primitivesocieties. Religion in these societies pervadesall social domains.Therefore,the argumentof the
interactionbetweenreligionsexplainedby therent-seekingmodelsdoes not work.The existenceof so many
differentkinds of organizationsbased on differentbeliefs makes the communicationacross these groups
more difficult. The degree of communicationloss depends on the numberof animist cults. See Reynal-
Querol (2000a) for a complete descriptionof the treatmentof animist cults.

THEECONOMETRIC
SPECIFICATION
Given the natureof the data,the econometricspecificationshouldaccommodatea

discretevariablewith the panel datastructure.Forthis purpose,a reasonablechoice is
the logistic model with individualeffects. To analyzethe effect of ethnicityandpoliti-
cal systems on ethniccivil war,we adopta generalspecificationderivedfromthe loot-
ing andjustice-seekingmodel, includingalternativeexplanatoryvariablesto show the
generalityof our findings.
The dependentvariableis the incidenceof ethniccivil war,ratherthanonset ethnic
civil wars, following other studies such as Elbadawi and Sambanis (forthcoming).
Some authorshave arguedthatit would be betterto analyze the onset of ethnic civil
warto avoid the analysisbeing influencedby time dependence.Therefore,we also do
the same analysis, changingthe dependentvariableof the onset of ethnic civil war to
corroboratethe resultswe findusing the incidenceof ethniccivil war.We do notreport
the results because there are no importantchanges in our findings and conclusions.
Moreover,to check thatourresultsare also robustto otherestimationprocedures,we
do the same analysisbutexploitthe panelcharacteristicsof the data.We use a random-
effects estimationand find thatthe resultsare qualitativelythe same. Finally,we also
comparethe resultsassuminga logit andprobitestimationfor the probabilityfunction.
The resultspresentedin this studyare robustto the use of all the differentestimation
proceduresexplainedhere.
Forall the empiricalexercises,we considera sampleof 138 countriesanddatafrom
1960 to 1995, organizedin periodsof 5 years.All the independentvariablesare taken
at the beginningof the period.The dependentvariableis a dummythattakesa value of
1 if the countryexperiencesan ethnic civil war duringthe period and 0 otherwise.
Theresultspresentedherearethe ones whenusing a logit pool estimationspecifica-
tion. Notice that even though Collier and Hoeffler (2000) never include the level of
GDP percapitaandthe level of educationtogether,we decide to includebothvariables
at the same time becausethis formulationcorrespondsto the usual specificationof the
new growth literature.When both variablesare included together,they have a poor
explanatorypower.If we includethemseparately,as CollierandHoefflerdo, they turn
out to be significant.A surprisingresultis the poorexplanatorypowerof the proxy for
naturalresources, which is opposite to the findings in Collier and Hoeffler (1998,
2000). They find thatnaturalresourcesarean importantvariablein explainingthe inci-
dence of civil war.This differencein the resultscould be causedby the differentsam-
ple we used andalso the differentdataset on civil wars.Notice thatwe restrictthe anal-
ysis to ethnic civil wars.Moreover,it is interestingto mentionthateven thoughwe do
not present the results, we find naturalresourcesto be a very importantvariablein
explaining the incidence of ideological/revolutionarycivil war and other kinds of
political violence such as coups or revolutions.Some of the reasons for these results
could be thatin ethniccivil wars,thejustice motivationprevailsoverthe looting moti-
vation, butin ideological civil wars,the looting for resourcescan be strongenough to
starta revolution.Whenthe society is notclearlydivided,thenthe motivationfor fight-
ing could come fromlooting for resources,and the existence of naturalresourcescan
be an importantcause. However,in ethnicallydividedsocieties, theexclusion of social

40 JOURNAL
OFCONFLICT
RESOLUTION
groupsis enoughto evolve into a civil warwithoutthe need of naturalresourcesin the

country.These results suggest that if we want to study the causes of civil wars, we
should probablyanalyzethese two types of civil warseparatelyto distinguishthe dif-
ferentcauses. To showtherobustnessof ourresults,we controlfor othervariables.The
economic variablesaddedarethe investmentshareof GDP andthe consumptionshare
of GDP, which are not directlyrelatedto the looting andjustice-seeking model. The
idea behindthe inclusionof these variablesis that,probablyif the countryis using the
resourcesfor investmentandconsumption,the opportunitycost of the resourcesdedi-
catedto supportviolence is higher.In otherwords,if the countrycan investandpeople
can consume and thereforeincrease their utility, the opportunitycost of fighting is
higher.
The polarizationmeasuresareconstantacrosstime. We includein all regressionsa
dummy variablefor each religion thatparticipatesin the constructionof the polariza-
tion index to avoid thatthe significanceof the index comes from the types of religions
ratherthanfromtheirpolarization.In this way, we tryto ensurethatthe index captures
only religious polarizationindependentlyof which religions exist in the country.
To control for the region, we introducedummy variablesfor sub-Saharan,Asian,
and Latin American countries.Many of the ethnic civil wars take place in Africa;
therefore,by includingthese variables,we wantto see if thereis still some effect thatis
not capturedby the fact of being an Africancountry.
Moreover,we noticedthatthe inclusion of primaryexportsreducesthe numberof
observationsbecause of missing data. Therefore,to see if our findings are robustto
more observations,we also did the analyses withoutincluding this variable,and the
results are even strongerthanthe ones presentedhere.
REGRESSIONRESULTS
Religious Polarization and Ethnic Civil War

In Table 1, we analyzethe effect of religious polarizationand animistdiversityon
the incidence of ethnic civil war.We show the results of the logit-pooled estimation
using the specificationdescribedin the previoussection. The most importantresultis
the importanceof religious polarizationand animistdiversityin explainingthe inci-
dence of ethnic civil war.
In column2, we finda positiveandsignificanteffect of animistdiversityon the inci-
dence of ethnic civil war.This resultshows the importanceof the existence of animist
cults, which are very typical in sub-Saharancountries.Moreover,this also suggests
thatthe most common sourcesof dataused in the literature,which do not considerthe
animistcults, are usuallyincludedtogetherwith the Christiansandmiss an important
dimension of religious diversity.The results in columns 3 and 4 show a positive and
significanteffect of religious polarizationin explainingthe incidence of ethnic civil
war.This resultis robustto the use of differentpolarizationmeasures.It contrastswith
recentresultsin the literatureon the causes of civil warthatclaim thateconomic fac-
tors aremore importantthanethnic characteristics.In regressions5 and 6, we include

TABLE 1
Logit Pool Estimationfor the Incidenceof Ethnic Civil Warfrom 1960-1995:
Religious Polarizationand Animist Diversity
Model 1 2 3 4 5 6
Constant -13.32 -22.00 -23.7 -25.44 -26.77 -29.78

(-2.3) (-2.84) (-3.11) (-3.26) (-3.38) (-3.61)
Lpop 0.77 0.43 0.64 0.72 0.37 0.39
(3.76) (1.68) (2.78) (3-16) (1.53) (1.65)
Lgdp 0.87 1.48 1.19 1.32 1.52 1.67
(1.67) (2.11) (1.87) (2.07) (2.26) (2.51)
Educ -0.21 -0.29 -0.34 -0.36 -0.27 -0.28
(-1.26) (-1.32) (-1.56) (-1.73) (-1.24) (-1.31)
Ex 0.10 0.13 0.11 0.13 0.11 0.10
(1.04) (1.07) (0.98) (1.16) (0.91) (0.83)
Ex2 -0.00 -0.00 -0.00 -0.00 -0.00 -0.00
(-1.1) (-1.03) (-0.96) (-1.02) (-0.93) (-0.87)
I -0.14 -0.14 -0.08 -0.10 -0.11 -0.11
(-3.23) (-2.84) (-1.69) (-2.14) (-2.1) (-2.29)
C -0.01 0.01 -0.00 -0.00 0.02 0.02
(-0.34) (0.52) (-0.00) (-0.02) (0.55) (0.80)
IRC1 5.7 5.15
(3.27) (3.08)
IRC2 8.02 8.52
(2.93) (2.95)
Trib 7.38 6.26 6.75
(3.11) (2.78) (3.07)
Democ 0.78 0.68 0.59 0.58 0.69 0.71
(3.51) (2.54) (2.36) (2.37) (2.52) (2.59)
Democ2 -0.09 -0.09 -0.07 -0.07 -0.08 -0.08
(-3.54) (-2.72) (-2.45) (-2.53) (-2.55) (-2.65)
Safrica -0.38 -1.4 -1.34 -1.22 -2.48 -2.49
(-0.54) (-1.26) (-1.23) (-1.11) (-1.93) (-1.91)
Asiae 1.46 3.85 4.5 4.46 4.76 4.82
(2.04) (2.71) (3.17) (3.14) (3.16) (3.16)
Laam -0.58 -0.38 -1.02 -0.62 -0.77 -0.57
(-0.95) (-0.39) (-0.97) (-0.64) (-0.68) (-0.51)
R2 0.3279 0.4075 0.4066 0.3888 0.4518 0.4442
Number 403 369 374 374 369 369
NOTE:Numbersin parenthesesare t statistics.In all the regressions,I include religious dummyvariables.

Educ = averageyears of schooling in the total population;Lpop = log of the populationat the beginningof
the period;Lgdp= log of the realgrossdomesticproduct(GDP)percapitaof the initialperiod(1985 interna-
tional prices);Ex = shareof primaryexportsin GDP;Ex2 = squareof ex; I = investmentshareof GDP; C =
consumptionshareof GDP;Democ andDemoc2 = democracylevel fromPolity IIIdatasource;IRC1= reli-
gious polarization(Montalvoand Reynal-Querol2000); IRC2 = religious polarization(Estebanand Ray
1994); Trib= animistdiversity;Safrica= dummyvariablefor sub-Saharancountries;Asiae = dummyvari-
able for Asian countries;Laam= dummy variablefor Latin Americancountries.
religious polarizationand animistdiversitytogether.We find that even if we include

them together,they still have a positive and significanteffect on the probabilityof an
ethnic civil war.This resultis opposite to the findings of Collier and Hoeffler (2000),

who reportno effect of polarization.The differencein the resultsis due to the fact that
they analyzethe causes of all civil wars;moreover,the datathey use on religionscome
from Barro (1997), which, as we mentionedbefore, have importantshortcomings.
Moreover,Sambanis(2001), using Vanhaven(1999) data,finds thatethnic heteroge-
neity is linearlyand positively correlatedwith the onset of ethnic civil war.However,
becausethe variablehe uses capturesthe inclusivedefinitionof ethnicity,it is difficult
to comparethe resultshe finds with the effect of religious polarizationfound here.
The resultsin Table1 show the importanceof the religiouscompositionof the soci-
ety as an importantsocial cleavage that affects the incidence of ethnic civil war.'6
These resultsareimportantfor threereasons.First,we considerpolarizationmeasures
insteadof fragmentationmeasuresto capturelatentconflicts, following the results of
the rent-seekingliterature.Second,we use the religiousdimensionof ethnicityinstead
of the linguistic one, as an importantcause of conflict, following some political scien-
tist theories.Finally,we consideranimistdiversity,very typical in sub-Saharancoun-
tries, as a special case of religious conflict.
Religious Polarization versus Linguistic Fragmentation

The literaturethatstudiesthe ethnic causes of social conflict has limited ethnicity
only to linguisticfragmentationand,in some cases, to religiousfragmentation.Collier
andHoeffler(2000) foundno effect of polarization,butgreatersocial fractionalization
actually reduces the risk of conflict. Collier, Hoeffler, and Soderbom (1999) and
Elbadawiand Sambanis(forthcoming)find empiricalevidence that ethnic diversity,
measuredby linguisticfragmentation,increasesthe probabilityof civil warat low lev-
els butthenreducesit at higherlevels. In this section,we analyzethe effect of linguistic
fragmentationon ethnic civil war and comparethis effect when we include religious
and linguistic fragmentationtogether.Table2 shows the resultof this analysis.In col-
umns 1 and 2, we find a positive and significanteffect of linguistic fragmentationon
ethniccivil war.Theresultsfoundherecorroboratethe implicittheoriesbehindthe use
of linguistic fragmentationindices, based on the idea that communicationproblems
increase with diversity,and this can cause conflict. However,these findings are very
sensitive to the inclusion of other ethnic variables.Columns 3 and 4 show how the
effect of languagefragmentationdisappearswhenreligiouscharacteristicsof a society
areincluded,whereasthe effect of religiouspolarizationandanimistdiversityremains
positive and significantno matterwhich polarizationindex we use.'7
These resultssuggestthatreligiousdifferencesin a countryaremoreimportantthan
linguistic differences as a social cleavage that can develop into civil war.Moreover,
they supportthe claims of Huntington(1996) aboutthe importanceof religiousdiffer-
ences in the explanationof domestic conflicts.
16. Wealso analyzetheeffect of ChristianpolarizationandMuslimpolarization.We finda significant

and positive effect on the incidence of ethnic civil war.
17. We also analyzethe interactionof both variables.However,we do not reportthe resultsbecause
we find thatreligious tensions,capturedthroughreligious polarizationand animistdiversity,affect ethnic/
religiouscivil warmorethanlinguisticfragmentationdoes, andthe presenceof bothsocial characteristicsin
a countrydoes not imply a higherprobabilityof conflict.

TABLE2
Religious and LinguisticFragmentation
Model 1 2 3 4
Constant -10.1 (-2.40) -19.24 (-2.83) -36.24 (-3.68) -40.50 (-3.75)

Lpop 0.61 (3.68) 0.67 (2.97) 0.35 (1.30) 0.36 (1.37)
Lgdp 0.08 (0.17) 0.79 (1.14) 2.22 (2.16) 2.50 (2.37)
Educ -0.01 (-0.05) -0.1 (-0.44) -0.35 (-1.20) -0.37 (-1.32)
Ex 0.07 (0.72) 0.19 (1.58) 0.22 (1.53) 0.19 (1.37)
Ex2 -0.00 (-0.88) -0.01 (-1.41) -0.01 (-1.40) -0.01 (-1.25)
I -0.61 (-1.14) -0.08 (-1.15) -0.09 (-1.34)
C 0.04 (1.22) 0.04 (0.88) 0.05 (1.12)
Elf60 2.75 (2.84) 4.25 (3.14) 2.31 (1.53) 2.41 (1.61)
IRC1 5.73 (2.75)
IRC2 9.75 (2.66)
Trib 6.10 (2.46) 6.41 (2.57)
Democ 0.77 (3.41) 0.63 (2.67) 0.48 (1.61) 0.49 (1.65)
Democ2 -0.1 (-3.65) -0.08 (-2.67) -0.06 (-1.68) -0.06 (-1.73)
Safrica -0.91 (-0.95) -2.24 (-1.65) -2.12 (-1.60)
Asiae 1.13 (1.34) 5.82 (3.12) 6.07 (3.16)
Laam 0.64 (0.77) 0.12 (0.09) 0.42 (0.32)
R2 0.3369 0.3880 0.4832 0.4777
Number 374 374 345 345

the period;Lgdp= log ofthe realgross domesticproduct(GDP)percapitaof the initialperiod(1985 interna-
consumptionshareof GDP;Elf60 = linguisticfragmentation;Democ andDemoc2 = democracylevel from
Polity III data source; IRC1 = religious polarization(Montalvo and Reynal-Querol2000); IRC2 = reli-
gious polarization (Estebanand Ray 1994); Trib= animist diversity; Safrica = dummy variablefor sub-
Saharancountries;Asiae = dummyvariablefor Asian countries;Laam= dummyvariablefor LatinAmeri-
can countries.
Religious Polarization versus Religious Fragmentation

To show not only the importanceof thereligiouscharacteristicsof a society butalso
the importanceof the mechanismthroughwhich social cleavage works, we compare
religious polarizationto religious fragmentation.18 Table 3 shows the results. In col-
umns 1 and 2, we observethatthe religious fragmentationmeasurehas no significant
effect on ethniccivil war.We test the robustnessof this effect with the inclusionof reli-
gious polarization(columns3 and4) andfind threeimportantresults:religious polar-
ization and animistdiversityhave a positive andsignificanteffect on civil warno mat-
ter which polarizationmeasurewe use. Also, religious fragmentationdoes not have a
significanteffect on ethniccivil war.However,the effect of religious fragmentationis
significantat the 10%level. This is importantbecauseit gives us some intuitionabout
18. When using religiousfragmentationdata,we reportthe resultsusing Barro's(1997) databecause
the literaturehas used these data when computing an index of religious fragmentation.We find similar
results when using Reynal-Querol(2001a) data and the fragmentationindex.

TABLE3
Religious Polarizationversus Religious Fragmentation
Model 1 2 3 4
Constant -14.41 (-2.18) -35.85 (-3.32) -27.58 (-2.58) -34.43 (-3.15)

Lpop 0.93 (3.58) 0.28 (0.92) 0.31 (0.77) 0.32 (0.84)
Lgdp 0.92 (1.46) 2.68 (2.90) 1.97 (2.19) 2.28 (2.51)
Educ -0.39 (-1.79) -0.61 (-2.19) -0.11 (-0.36) -0.13 (-0.44)
Ex 0.09 (0.80) 0.20 (1.32) 0.19 (1.26) 0.13 (0.89)
Ex2 -0.00 (-0.70) -0.01 (-1.19) -0.01 (-1.24) -0.00 (-0.96)
I -0.14 (-2.75) -0.13 (-2.26) -0.11 (-1.78) -0.12 (-2.02)
C -0.02 (-0.82) 0.05 (1.45) -0.00 (-0.13) 0.01 (0.25)
Rff -1.45 (-0.76) 3.08 (0.99) -9.85 (-1.89) -9.82 (-1.98)
IRC1 10.75 (3.13)
IRC2 19.13 (3.30)
Trib 8.78 (2.92) 7.06 (2.34) 7.95 (3.05)
Democ 0.64 (2.48) 0.69 (2.47) 0.59 (1.98) 0.62 (2.09)
Democ2 -0.08 (-2.52) -0.08 (-2.59) -0.07 (-1.94) -0.07 (-2.07)
Safrica 0.08 (0.07) -1.59 (-1.23) -1.71 (-1.13) -1.64 (-1.11)
Asiae 2.51 (2.48) 6.6 (3.33) 5.33 (3.01) 5.80 (3.22)
Laam -1.07 (-1.24) 0.87 (0.65) -3.43 (-1.63) -2.74 (-1.51)
R2 0.3556 0.4343 0.5029 0.4986
Number 323 310 310 310

the period;Lgdp= log of the realgrossdomesticproduct(GDP)percapitaof the initialperiod(1985 interna-
consumptionshareof GDP;Democ andDemoc2 = democracylevel fromPolity IIIdatasource;IRC1= reli-
gious polarization(Montalvoand Reynal-Querol2000); IRC2 = religious polarization(Estebanand Ray
1994); Trib= animistdiversity;Safrica= dummyvariablefor sub-Saharancountries;Asiae = dummyvari-
able for Asian countries;Laam= dummyvariablefor Latin Americancountries.
the effect of religious diversityon ethnic civil war.It seems thatthe fragmentationof
the society on religiousgroupsis preventingcountriesfromviolence ratherthaninduc-
ing them to conflict. When analyzing the causes of civil wars, Collier and Hoeffler
(2000) also find evidence that religious fragmentationmakes countriessafer.More-
over, Elbadawiand Sambanis(forthcoming)find a quadraticinteractiontermof reli-
gious andethnicdiversitythatis negativelyassociatedwiththe incidenceof civil war.
Political Causes of Ethnic Civil War

In this section, we analyzethe effect of politicalrightsandcivil libertieson the inci-
dence of ethniccivil waraftercontrollingfor the social effects analyzedin the previous
sections. The results are reportedin Table4. We use data from the Polity III data set
(columns 1 to 4) andfromthe FreedomHouse dataset (columns5 to 6). We find what

otherauthorshavepredicted.Countrieswith a midleveldemocracyhavea higherprob-

ability of suffering a civil war.19These findings are consistent with Collier and
Hoeffler(2000), Sambanis(2001), Hegreet al. (2001), andEllingsen(2000). The intu-
ition behindthese resultsis thatsome level of freedomis neededto let people organize
to starta civil war.
We have shown that democracyseems not to be a sufficient condition to prevent
countriesfrom gettinginvolvedin a civil war.We now introducethe institutionalvari-
ables created in Reynal-Querol(2001b). We define autocracy(AUTO), partialfree-
dom (PF), majoritariansystem (MAJO),presidentialsystem (PS), and proportional
representation(PR).In the firststep, we analyzethe effect of the differentinstitutional
systems by using dummyvariables.However,because there is no countrythat has a
proportionalsystem and has experiencedan ethnic civil war duringthe next 5 years,
the predictionpowerof this dummyis perfect.This fact makesthe logit panel a badly
defined specification. Therefore, to avoid this problem, we construct, as Reynal-
Querol(2001b) did, variablesthatorderthe differentsystems accordingto the level of
inclusiveness of theirvoting rules.Forthe politicalinstitution,we tryto summarizein
one variable the informationcontained in the five dummies referredto earlier.We
orderthe five dummies with respect to the inclusiveness of their systems. The most
inclusive rule is unanimity.We know that nonfree systems are less inclusive than
nonauthoritarian countries,and pluralitysystems are less inclusive thanproportional
representationsystems. A numberof countrieshave presidentialsystems. The theory
does not incorporatethis directly.However,thereis a sense thatsocieties with a presi-
dentialsystem anda proportionalsystem in the assembly aremore inclusivethanpure
majoritariansystems. By definition, the election of a presidentis by majorityrule;
therefore,whatmakesa differencewith respectto the presidentialsystem is the voting
rule followed in the assembly.It would be ideal to have datathatdistinguishbetween
the kind of presidentialsystems in terms of their different level of inclusiveness,
dependingon the voting rule followed in the assembly.However,we are not awareof
the existence of sucha dataset. Therefore,if we orderthe systems by the level of inclu-
siveness, presidentialsystems are less inclusive than proportionalrepresentationand
equal or more inclusive thanmajoritarianrule systems, dependingon the voting rule
thatis followed in the assemblyelections. Therefore,we createa variablecalled INCV,
which has value of 0 if the systemis not free, 1 if it has a majoritariansystem, 2 if it has
a presidentialsystem, and 3 if it has a proportionalsystem. Alternatively,mainly
because following this orderis difficult to ensurethat presidentialsystems are more
inclusive thanmajoritarian,we havecreatedanothervariable,called INCV1, thathas a
value of 0 if the countryis not free, 1 if it has a majoritarianorpresidentialsystem, and
2 if it has a proportionalsystem.
We reportthe resultsin Table5. In columns 1 and 2, we analyze the effect of the
political system, controllingfor the democracylevel and using the two variablesthat
capturethe level of inclusiveness,INCV and INCV1. We find thatthe level of inclu-
siveness of the politicalsystemhas a negativeandsignificanteffect on the incidenceof
19. However,whenincludingthe squareof the autocracy,column4, the effect of this variableremains

not significant.

TABLE4
Logit Pool Estimationfor the Incidenceof EthnicCivil Warfrom 1960-1995:
Political Rights and Civil Liberties
Model 1 2 3 4 5 6
Constant -25.79 -24.85 -26.77 -26.68 -25.87

(-3.33) (-3.27) (-3.28) (-3.49) (-3.13)
Lpop 0.38 0.32 0.37 0.31 0.51
(1.57) (1.30) (1.53) (1.29) (1.94)
Lgdp 1.26 1.23 1.52 1.37 1.26
(1.95) (1.98) (2.26) (2.18) (1.85)
Educ -0.33 -0.33 -0.27 -0.31 -0.29
(-1.55) (-1.58) (-1.24) (-1.47) (-1.26)
Ex 0.09 0.09 0.11 0.15 0.11
(0.85) (0.86) (0.91) (1.25) (0.89)
Ex2 -0.00 -0.00 -0.00 -0.00 -0.00
(-0.87) (-0.88) (-0.93) (-1.19) (-0.92)
I -0.09 -0.08 -0.11 -0.08 -0.09
(-1.76) (-1.67) (-2.10) (-1.64) (-1.70)
C 0.01 0.01 0.01 0.01 0.01
(0.42) (0.36) (0.55) (0.29) (0.22)
IRC1 5.28 5.53 5.15 5.42 4.34
(3.18) (3.33) (3.08) (3.31) (2.71)
Trib 6.38 6.50 6.26 6.55 5.80
(2.73) (2.77) (2.78) (2.84) (2.50)
Democ 0.02 0.69
(0.19) (2.52)
Democ2 -0.08
(-2.55)
Autoc -0.07 0.36
(-0.87) (1.21)
Autoc2 -0.05
(-1.51)
Freedom -1.02
(-1.34)
PF 0.24
(0.33)
Safrica -1.96 -2.03 -2.48 -2.16 -1.89
(-1.54) (-1.62) (-1.93) (-1.81) (-1.44)
Asiae 5.48 5.61 4.76 5.94 5.45
(3.78) (3.84) (3.16) (3.91) (3.60)
Laam -0.03 -0.10 -0.77 -0.14 -0.00
(-0.03) (-0.09) (-0.68) (-0.13) (-0.00)
0.4248 0.4276 0.4518 0.4365 0.4297
Number
Number 369 369 369 369 346

Results do not changeif we controlfor othersocial variables.Educ = averageyearsof schooling in the total
population;Lpop= log of the populationatthebeginningof the period;Lgdp= log of therealgross domestic
product(GDP) per capitaof the initial period(1985 internationalprices);Ex = shareof primaryexportsin
GDP; Ex2 = square of ex; I = investmentshare of GDP; C = consumptionshare of GDP; Democ and
Democ2 = democracy level from Polity III data source; IRC1 = religious polarization(Montalvo and
Reynal-Querol2000); trib = animist diversity;Autoc and Autoc2 = autocracylevel from Polity III data
source;PF = dummyvariablefor partiallyfree countries(FreedomHouse datasource);Freedom= dummy
variablefor free countries(FreedomHouse datasource);Safrica= dummyvariablefor sub-Saharancoun-
tries;Asiae = dummyvariableforAsian countries;Laam= dummyvariablefor LatinAmericancountries.

SYSTEMS,AND WARS 47
Reynal-Querol/ETHNICITY,POLITICAL
TABLE5
LogitPoolEstimation
fortheIncidenceof EthnicCivilWarfrom1960-1995:
PoliticalSystem
Model 1 2 3 4 S 6 7 8
Constant -30.07 -29.39 -27.54 -26.33 -29.10 -28.37 -25.89 -25.24

(-3.58) (-3.50) (-3.41) (-3.26) (-3.50) (-3.42) (-3.30) (-3.18)
Lpop 0.57 0.55 0.46 0.42 0.55 0.52 0.44 0.41
(2.20) (2.11) (1.75) (1.59) (2.11) (2.00) (1.68) (1.54)
Lgdp 1.81 1.79 1.61 1.57 1.74 1.72 1.61 1.65
(2.48) (2.46) (2.41) (2.37) (2.39) (2.37) (2.40) (2.42)
Educ -0.29 -0.30 -0.31 -0.32 -0.22 -0.26 -0.29 -0.32
(-1.25) (-1.33) (-1.44) (-1.50) (0.92) (-1.07) (-1.31) (-1.41)
Ex 0.23 0.21 0.20 0.19 0.23 0.21 0.18 0.19
(1.56) (1.51) (1.52) (1.45) (1.51) (1.47) (1.39) (1.36)
Ex2 -0.01 -0.01 -0.01 -0.01 -0.01 -0.01 -0.01 -0.01
(-1.39) (-1.35) (-1.36) (-1.32) (-1.35) (-1.32) (-1.26) (-1.25)
I -0.18 -0.17 -0.13 -0.12 -0.21 -0.19 -0.15 -0.14
(-3.03) (-2.91) (-2.37) (-2.23) (-3.30) (-3.10) (-2.59) (-2.48)
C 0.00 0.01 -0.00 0.00 -0.00 -0.00 -0.00 -0.00
(0.15) (0.21) (-0.02) (0.02) (-0.1I1) (-0.03) (-0.17) (-0.18)
IRCI 5.33 5.23 5.39 5.29 5.58 5.42 5.57 5.67
(3.25) (3.18) (3.35) (3.30) (3.27) (3.21) (3.38) (3.37)
Trib 6.17 6.04 6.35 6.23 6.33 6.22 6.70 6.78
(2.82) (2.84) (2.84) (2.87) (2.75) (2.77) (2.89) (2.98)
Democ 1.01 0.99 0.87 0.86
(3.34) (3.29) (2.76) (2.76)
Dernoc2 -0.09 -0.09 -0.05 -0.05
(-2.89) (-2.85) (-1.39) (-1.41)
Autoc -0.07 -0.10 -0.48 -0.64
(-0.21) (-0.29) (-1.17) (-1.51)
Autoc2 -0.02 -0.02 0.01 0.02
(-0.58) (-0.52) (0.31) (0.57)
Incv -1.46 -2.51 -1.28 0.02 -2.07
(-2.88) (-2.83) (-2.55) (0.03) (-3.16)
Incvl -2.30 -0.42 -4.28
(-2.60) (-0.37) (-3.44)
Demincv -0.35
(-2.49)
Denmincvl -0.54
(-2.35)
Autincv 0.29
(2.53)
Autincvl 0.61
(3.16)
Safrica -3.78 -3.63 -2.77 -2.72 -4.52 -4.37 -3.10 -3.16
(-2.62) (-2.57) (-2.16) (-2.16) (-2.75) (-2.71) (-2.28) (-2.33)
Asiae 5.93 5.73 6.76 6.55 5.62 5.40 6.53 6.55
(3.58) (3.52) (4.20) (4.10) (3.35) (3.27) (4.14) (4.06)
(continued)

48 JOURNAL
OFCONFLICT
RESOLUTION
TABLE5 Continued
Model 1 2 3 4 5 6 7 8
Laam -0.65 -0.73 -0.08 -0.19 -0.67 -0.78 -0.17 -0.26

(-0.54) (-0.62) (-0.07) (-0.17) (-0.50) (-0.60) (-0.14) (-0.22)
R2 0.4877 0.4875 0.4644 0.4666 0.5140 0.5094 0.4811 0.4933
Number 369 369 369 369 369 369 369 369

the period;Lgdp= log of thereal gross domesticproduct(GDP)percapitaof the initialperiod(1985 interna-
consumptionshareof GDP;Democ and Democ2 = democracylevel from Polity IIIdatasource;Autoc and
Autoc2 = autocracylevel fromPolity IIIdatasource;IRC1= religious polarization(Montalvoand Reynal-
Querol 2000); Trib = animist diversity;Incv and Incvl = level of inclusiveness of the political system;
Demincv = democ * incv; Demincvl = democ ? incvl; Autincv = autoc * incv; Autincvl = autoc * incvl;
Safrica= dummyvariablefor sub-Saharancountries;Asiae = dummyvariablefor Asian countries;Laam=
dummy variablefor LatinAmericancountries.
ethnic civil war. Controllingfor the level of autocracyinstead of democracy (co!-

umns 3 and 4), we find the same results on the effect of INCV and INCV1.20These
findings show the importanceof the level of inclusiveness of the political system in
preventingethnic civil war.
We include in the regressionthe interactionbetween the level of inclusivenessand
the democracylevel (columns 5 and 6 of Table5) and the level of inclusiveness with
the autocracylevel (columns 7 and 8). Columns5 and 6 show thatthe level of inclu-
siveness of the political system in democraticcountrieshas a negativeand significant
effect on the incidence of ethnic civil war. Moreover,this result indicates that the
higherthe democraticlevel of a country,the largerthe effect of the level of inclusive-
ness on ethnic civil war.These resultsarecorroboratedwhen using the level of autoc-
racy insteadof democracy(columns 7 and 8).
These findings suggest thatwe need to controlnot only for the level of democracy
but also for the type of political system. Not all political institutionsworkin the same
way,andthe level of representationof thepopulationis a key elementif we wantto pre-
vent countriesfrom ethnic civil war.This inclusiveness can be achievedby applying
consociationaldemocraciesratherthanmajoritariansystems.
Policy Implications
In this section, we analyze whetherin religiously polarized and animist diverse
societies, the existence of a political system with a high level of representationof the
populationreducesthe effect of this latentconflict on the incidenceof ethniccivil war.
Table 6 shows the results of this analysis.
20. If we do the same analysisbutwithoutincludingthe squareof the democracyandautocracyvari-

ables, we findthatdemocracyhas no significanteffect on the incidenceof ethnic/religiouscivil war,butthe
autocracylevel has a negativeandsignificanteffect. The level of inclusivenesscontinuesto have a negative
and significanteffect.

In columns 1 and2, we includethe IRC1measuretogetherwith the level of democ-

racy,the level of inclusiveness,INCV,andtheirinteraction.We find thatin religiously
polarized countries,the effect of polarizationon the incidence of ethnic civil war is
reducedby the effect of having a political system with a high level of inclusiveness.
These resultsarerobustto the use of differentvariablesthatcapturethe level of inclu-
siveness (column 2) andto the inclusionof the squareof the democracyvariable(col-
umns 3 and4). Moreover,they arealso robustto the use of differentpolarizationmea-
sures (columns 5 and 6). In regression7 of Table6, we analyze whetherthe political
system can help to reducethe effect of animistdiversityon ethniccivil war.The results
show thatthe level of inclusivenessdoes not help in reducingthe risk of a warcaused
because of animist diversity.In column 8 of Table 6, we analyze all these effects
together.We find the same results as when we analyze these effects separately.
Because of this result,we test one of the argumentsof Horowitz(1985). He argues
that in ethnicallydividedsocieties, the pluralityrule system is a mechanismthatpro-
motes the creationof multiethnicparties,which helps to reduce ethnic tensions. We
analyze whethersystems with pluralityrule help to reducethe effect of animistdiver-
sity on the incidence of ethniccivil war.In column9 of Table6, we findthatthe effect
of animistdiversityon ethniccivil waris reducedby the establishmentof majoritarian
systems. We analyze in column 10 of Table 6 all the effects together.From these
results, we conclude thatthe inclusion of consociationaldemocraciesis an important
measurethatsignificantlyreducesthe incidence of ethnic civil wargeneratedby reli-
gious polarization,andthe effect of animistdiversityon ethniccivil waris reducedby
the establishmentof majoritariansystems.
On one hand, the empirical evidence seems to corroboratethe arguments of
Lijphart(1977, 1984) on which political system is better to reduce ethnic conflict
because consociationaldemocraciesreducethe effect of religious polarizationon the
incidence of ethniccivil war.On the otherhand,the evidence also supportspartof the
argumentby Horowitz(1985) becausethe conflict generatedby animistdiversitycan-
not be reducedby applyingconsociationaldemocraciesbut by applyingmajoritarian
systems.
Therefore, in societies with a high degree of religious polarizationand animist
diversity,it will be necessary to design new political systems that consider, on one
hand,a majoritariansystemto institutionalizetherelationshipamongthe differentani-
mist groupsand, on the otherhand,a proportionalsystem to managethe relationship
between all the animistgroupsand the othergroupsin the society.
We test this implicationin the case of a countrywith a significantamountof animist
groups and anotherreligious group.Therefore,the majoritariansystem works in line
with Horowitz's (1985) theories,helping to promotethe creationof multiethnicpar-
ties. In this context, the animist groups are very different from the other religious
groups,andtheircommonelementscan providean incentivefor the creationof multi-
ethnicparties.The resultsthereforeapply when a significantnumberof ethnicgroups
competeforpowerwith otherreligions.The differencesbetweenthe ethnicgroupsand
otherlargereligionsimplythe needfor cohesion amongthe animistorethnicgroupsto
defend theirown identitiesagainstlargerreligions such as Christianityor Islam.How-

TABLE 6
Logit Pool Estimationfor the Incidence of Ethnic Civil Warfrom 196
Political Systems with a High Level of Inclusion and Representa
Model 1 2 3 4 5 6 7
Constant -13.29 -12.65 -11.52 -11.51 -12.32 -11.77 -7.24

(-3.12) (-2.98) (-2.68) (-2.69) (-2.86) (-2.74) (-1.89)
Lpop 1.04 1.06 1.00 1.02 1.01 1.02 0.67
(6.37) (6.56) (6.06) (6.24) (6.38) (6.54) (4.40)
Lgdp -0.03 -0.09 -0.11 -0.14 -0.13 -0.18 -0.38
(-0.08) (-0.23) (-0.28) (-0.34) (-0.33) (-0.45) (-0.96)
Educ -0.39 -0.39 -0.39 -0.39 -0.36 -0.37 -0.25
(-2.76) (-2.74) (-2.63) (-2.60) (-2.58) (-2.59) (-1.81)
I -0.07 -0.07 -0.07 -0.07 -0.06 -0.07 -0.06
(-2.18) (-2.28) (-2.40) (-2.43) (-2.18) (-2.22) (-2.15)
C -0.04 -0.04 -0.04 -0.04 -0.04 -0.04 -0.03
(-1.97) (-2.04) (-2.21) (-2.16) (-2.10) (-2.12) (-1.72)
IRC1 3.80 3.68 3.48 3.52
(3.89) (3.69) (3.54) (3.50)
IRC2 4.99 4.80
(3.08) (2.88)
Trib 6.12
(3.19)
Democ 0.17 0.22 0.70 0.69 0.17 0.22 0.15
(2.76) (3.33) (3.59) (3.55) (2.81) (3.36) (2.42)
Democ2 -0.6 -0.05
(-2.84) (-2.55)
Incv -0.38 -0.44 -0.54 -1.01
(-1.06) (-1.09) (-1.52) (-2.98)
Incvl -1.06 -1.03 -1.35
(-1.75) (-1.57) (-2.21)

Cflndem -0.29 -0.29
(-2.23) (-2.00)
Cflnldem -0.47 -0.45
(-2.33) (-2.12)
Cf2Indem -0.48
(-1.99)
Cf2Inldem -0.70
(-2.01)
Ltlndem 0.23
(0.50)
Ltlnldem
MAJOc
Ltmaj
R2 0.3771 0.3877 0.3964 0.4031 0.3616 0.3728 0.3704

Number 606 606 606 606 606 606 596
NOTE:Numbersin parenthesesare t statistics. In all the regressions,I includereligious dummyvariablesandregional dumm

the totalpopulation;Lpop= log of the populationat the beginningof the period;Lgdp = log of the real gross domestic produ
ternationalprices);I = investmentshareof GDP; C = consumptionshareof GDP; Democ and Democ2 = democracylevel fr
ization(MontalvoandReynal-Querol2000); IRC2= religious polarization(Estebanand Ray 1994); Trib= animistdiversity
litical system; Cflndem = confl * demlncv; Cflnldem = confl * demlncvl; Cf2Indem = conf2 * demlncv; Cf2Inldem =
Ltlnldem = Itrib* demlncvl.

52 OFCONFLICT
JOURNAL RESOLUTION
ever,these resultscannotbe generalizedwhen societies aredivided by animist/ethnic

groups because in this case, the differencesamong them appearto be stronger.
CONCLUSIONS
This study has analyzed the social and political causes of ethnic civil war.
Threeimportantfindingsarerobustfor the use of differentestimationproceduresand
specifications.
The firstis the importanceof religiouspolarizationandanimistdiversityin explain-
ing the incidenceof ethniccivil warno matterwhatothermeasuresareincluded.These
resultscorroboratethe resultsof the literatureon rent-seekingmodels thatstudy social
conflict andthe importanceof the distributionof the groupsin a countryas a key ele-
ment for understandingthe causes of social conflicts. Second, we find thatreligious
polarizationis moreimportantas a social cleavagethatcan develop into civil warthan
are linguistic differences. This result corroborateswhat Huntington(1996) claims
aboutthe importanceof religiousdifferencesto explaindomestic conflicts. Third,we
find thata consociationaldemocracyis a politicalsystem thatsignificantlyreducesthe
incidence of ethniccivil war.Mainly,it reducesthe probabilityof ethniccivil wargen-
eratedby religious polarization.
The study of the causes of civil war deserves additionalstudies, especially of the
ethnicandpoliticalissues thatmay influencetheincidenceof civil war.Ouranalysis-
based on the study by Collier and Hoeffler (2000), who argue that conflicts are far
more likely to be causedby economic opportunitiesthanby grievance-is focused on
the ethnicandpoliticalcauses of civil war.We use a theoreticallybasedindex of latent
conflict and give special importanceto the religious dimensions of ethnicity,using a
new dataset thattriesto overcomethe commoncriticismto the WorldChristianEncy-
clopedia (Barret1982) data.Moreover,herewe arguethatwhatmattersarethe level of
inclusiveness of the system and the level of democracyinsteadof the level of democ-
racy per se.
These results are importantfor four reasons.First, we considerpolarizationmea-
sures insteadof fragmentationmeasuresto capturelatentconflict. Second, we use the
religious dimensionof ethnicityinsteadof the linguistic one as an importantcause of
conflict. Third,we consideranimistdiversity,verytypicalin sub-Saharancountries,as
a special case of religious conflict. Finally, most of the literaturehas considered
democracyas the only politicalvariablethatmay affectthe probabilityof civil war,and
even the empiricalevidence aboutthis is very week. We arguethatwe need to control
not only for democracybutalso for the political system. The resultsclarifythe role of
democracyitself.

APPENDIX
Property 1: It attainsits maximumat the bipolarsymmetricdistribution.

Property2: Suppose we startwith the populationequally distributedover two groups. Let us
transfer x from each original group to two newly created groups. Then the index is
nonmonotonicwith respectto x, and it reachesits minimumat four equal-sizedgroups.
Property3: Considerthe populationdividedinto N groupsof size 1/N. Polarizationdecreases
with N.
Indeed, the index is as follows:
N
IRCl= 1- (0.5-i)2 =/0. -4N( /- )2(N) = 1-4(-2 N)2
i=1
Clearly,the index is strictlydecreasingin N.

Property4: If we startwith a uniformdistributionover N groups, any merging of k adjacent
groups will increase polarization.
This is a direct corollaryof property3.
Property5: There aretwo groupswith size t1 andn2.Takeany one groupandsplit it into m > 2
groupsin such a way thatnI = i > ii Vi=2 ...n + , where cis the new vectorof population
n+l-
sizes, and clearly
. i=2 i =712. Then polarizationunder ii is smaller than undern. (See
Reynal-Querol2001a for the proofs.)
REFERENCES
Barret,D., ed. 1982.WorldChristian

encyclopedia.Oxford,UK:OxfordUniversityPress.
Barro,RobertJ. 1997.Determinants of economicgrowth:A cross-country empiricalstudy.Cambridge,
MA:MITPress.
Barro,RobertJ.,andJong-Wha Lee.1996.International
measures of schoolingyearsandschoolingquality.
AmericanEconomicReview,PapersandProceedings 86 (2):218-23.
Collier,P.,andA.Hoeffler.1998.Oneconomiccausesof civilwars.OxfordEconomic Papers50:563-73.
.2000. Greedandgrievances.WorldBankDECRGWorking Paper,Washington,DC.
Collier,Paul,AnkeHoeffler,andMansSoderbom.1999.Onthedurationof civilwar.Washington, DC:
WorldBank,Development EconomicResearchGroup.
Colomer,J. M. 2000. Political institutions,democracyand social choice. Oxford,UK: OxfordUniversity
Press.
Doyle,MichaelW.,andNicholasSambanis. 2000.International A theoretical
peacebuilding: andquantita-
tiveanalysis.American PoliticalScienceReview94 (4):779-801.
Elbadawi,Ibrahim A. 1999.Civilwarsandpoverty:Theroleof externalinterventions,politicalrightsand
economicgrowth.Paperpresented attheWorldBank'sConference on"CivilConflicts,CrimeandVio-
lence,"February,Washington, DC.
Elbadawi, Ibrahim,andNicholasSambanis. Forthcoming.Howmuchwarwillwe see?Estimating theinci-
denceof civilwarin 161countries,1960-1999.Journalof ConflictResolution.
Ellingsen,Tanja.2000.Colorfulcommunity orethnicwitches'brew?Journalof ConflictResolution 44 (2):
228-49.
Esteban,J., andD. Ray.1994.Onthemeasurement of polarization.
Econometrica 62 (4):819-51.
.1999. Conflictanddistribution.
Journalof EconomicTheory87:379-415.

Hegre,H., T.Ellingsen,N. P.Gleditsch,andS. Gates.2001. Towardsa democraticcivil peace?Opportunity,

grievance,and civil war, 1816-1992. AmericanPolitical Science Review95 (1): 33-48.
Horowitz, D. 1985. Ethnicgroups in conflict. Berkeley:Universityof CaliforniaPress.
Huntington,S. 1996. The clash of civilizations and the remakingof world order. New York:Simon &
Schuster.
Licklider,Roy. 1993. How civil warsend: Questionsand methods.In Stoppingthe killing:How civil wars
end, edited by Roy Licklider.New York:New YorkUniversityPress.
.1995. The consequencesof negotiatedsettlementsin civil wars, 1945-1993. AmericanPolitical
Science Review 89 (3): 681-90.
Lijphart,A. 1977. Democracyinplural societies:A comparativeexploration.New Haven,CT:YaleUniver-
sity Press.
. 1984. Democracies,patternsof majoritarianand consensus governmentin twenty-onecountries.
New Haven, CT: Yale UniversityPress.
Montalvoand Reynal-Querol.2000. A theoryof religious conflict and its effects on growth.IVIEWorking
Paper2000-04.
Reynal-Querol,M. 2001a. Religious andethnic conflict, political systems and growth.Ph.D. diss, London
School of Economics.
. 2001b. Why some democraciesfail: Stability,political systems and civil war.Mimeographed.
Sambanis, Nicholas. 2001. Do ethnic and nonethniccivil wars have the same causes? A theoreticaland
empiricalinquiry(partI). Journalof ConflictResolution45 (3): 259-82.
Singer, J. D., and M. Small. 1994. Correlatesof war project:Internationaland civil war data, 1816-1992.
Computerfile. Inter-UniversityConsortiumfor Political and Social Research,Ann Arbor,MI.
Small, M., and J. D. Singer. 1982. Resort to arms: Internationaland civil war, 1816-1980. Beverly Hills,
CA: Sage.
Summers,R., andA. Heston. 1991. The PennWorldTable(Mark5): An expandedset of internationalcom-
parisons, 1960-1992. QuarterlyJournalof Economics 106:327-68.
Taylor,C., and M. C. Hudson. 1972. The world handbookof political and social indicators.2d ed. New
Haven, CT: Yale UniversityPress.
Vanhaven,Tatu. 1999. Domestic ethnic conflict and ethnic nepotism:A comparativeanalysis. Journal of
Peace Research 36 (1): 55-73.


Reynal-Querol (2002) Ethnicity, Political Systems, and Civil Wars

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Reynal-Querol (2002) Ethnicity, Political Systems, and Civil Wars

Uploaded by

Copyright:

Available Formats

Ethnicity, Political Systems, and Civil Wars

Author(s): Marta Reynal-Querol

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

Theeffect of ethnicdivisionon civil warandthe roleof politicalsystems in preventingthese conflicts are

Social and ethnic conflicts are recurrentphenomenaaffecting many countries.Eth-

AUTHOR'S NOTE: I would like to thankJosep MariaColomer,Joan M. Esteban,and Jose Garcia-

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

between these two communities, the proportionalsystem introducedin 1970 has

THE CAUSES OF CIVIL WARS

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

However,not all civil warsareof the same nature.Forthis reason,it makessense to

SOCIOPOLITICAL FACTORS AND ETHNIC CONFLICTS

Manyconflicts relatedto religiousdifferencesariseeveryday.Even thoughecono-

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

guage arethe basic dimensionsof a politicalparty.Lijphart(1984) found,in a sample

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

extent it is the distributionof the groups from a bimodal distribution.However, the

TheindexofpolarizationofReynal-Querol3(2001a). This indextakesthe following

where ni is the proportionof each religion and N is the numberof religions.

The behaviorof the index criticallydependson the propertiesof theffunction. By

Theffunction reachesa minimumfor n = 1/6 anda maximumfor i = 1/2. It is con-

2. A basic readingis Estebanand Ray (1999).

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

1/6 1/3 1/2

Estebanand Ray (1994) indexof polarization.To show thatthe resultsobtainedin

where d = 1 if i j and d = 0 if i =j.

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

6. This value comes to be in the rangeof feasible values for a.

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

Ethnicconflict in pluralsocieties andhow to controlit often havebeen examinedin

For the purpose of the empirical investigation,and given the determinantsdis-

8. Moreover,he arguesthat"ina politicalsystemwith clearlyseparateandpotentiallyhostile popula-

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

civil warcome fromthe statefailuredataset.9An ethnic waris definedas an episode of

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

Given the natureof the data,the econometricspecificationshouldaccommodatea

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

groupsis enoughto evolve into a civil warwithoutthe need of naturalresourcesin the

Religious Polarization and Ethnic Civil War

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

Constant -13.32 -22.00 -23.7 -25.44 -26.77 -29.78

NOTE:Numbersin parenthesesare t statistics.In all the regressions,I include religious dummyvariables.

religious polarizationand animistdiversitytogether.We find that even if we include

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

Religious Polarization versus Linguistic Fragmentation

16. Wealso analyzetheeffect of ChristianpolarizationandMuslimpolarization.We finda significant

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

Constant -10.1 (-2.40) -19.24 (-2.83) -36.24 (-3.68) -40.50 (-3.75)

NOTE:Numbersin parenthesesare t statistics.In all the regressions,I include religious dummyvariables.

Religious Polarization versus Religious Fragmentation

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

Constant -14.41 (-2.18) -35.85 (-3.32) -27.58 (-2.58) -34.43 (-3.15)

NOTE:Numbersin parenthesesare t statistics.In all the regressions,I include religious dummyvariables.

Political Causes of Ethnic Civil War

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

otherauthorshavepredicted.Countrieswith a midleveldemocracyhavea higherprob-

19. However,whenincludingthe squareof the autocracy,column4, the effect of this variableremains

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

Constant -25.79 -24.85 -26.77 -26.68 -25.87

NOTE:Numbersin parenthesesare t statistics.In all the regressions,I include religious dummyvariables.

This content downloaded from 128.151.244.46 on Tue, 3 Sep 2013 06:56:44 AM

Constant -30.07 -29.39 -27.54 -26.33 -29.10 -28.37 -25.89 -25.24