ethnic politics and armed conflict: a configurational...

Karl Marx predicted that revolutionary classstruggles would transform the world dur-

ing the twentieth century. Instead, it turned outto be the age of ethnonationalist conflicts. Warsfought in the name of national liberation or eth-nic autonomy comprise only one fifth of the

wars between the Congress of Vienna (1814)and the Treaty of Versailles (1919). FromVersailles to 2001, however, the share of eth-nonationalist wars rose to 45 percent, and sincethe Cold War ended it has reached 75 percent.1

Ethnic demands and grievances play a promi-

Ethnic Politics and Armed Conflict: AConfigurational Analysis of a New GlobalData Set

Andreas Wimmer Lars-Erik CedermanUCLA ETH Zurich

Brian MinUCLA

Quantitative scholarship on civil wars has long debated whether ethnic diversity breeds

armed conflict. We go beyond this debate and show that highly diverse societies are not

more conflict prone. Rather, states characterized by certain ethnopolitical configurations

of power are more likely to experience violent conflict. First, armed rebellions are more

likely to challenge states that exclude large portions of the population on the basis of

ethnic background. Second, when a large number of competing elites share power in a

segmented state, the risk of violent infighting increases. Third, incohesive states with a

short history of direct rule are more likely to experience secessionist conflicts. We test

these hypotheses for all independent states since 1945 using the new Ethnic Power

Relations (EPR) data set. Cross-national analysis demonstrates that ethnic politics is as

powerful and robust in predicting civil wars as is a country’s level of economic

development. Using multinomial logit regression, we show that rebellion, infighting, and

secession result from high degrees of exclusion, segmentation, and incohesion,

respectively. More diverse states, on the other hand, are not more likely to suffer from

violent conflict.

AMERICAN SOCIOLOGICAL REVIEW, 2009, VOL. 74 (April:316–337)

Direct correspondence to Andreas Wimmer([email protected]). The authors wish to thankthe many individuals who helped assemble the dataset on which this article relies. While we cannot listthe dozens of country and regional experts whogenerously shared their knowledge, we should liketo at least mention Dennis Avilés, Yuval Feinstein,Dmitry Gorenburg, Wesley Hiers, Lutz Krebs,Patrick Kuhn, Anoop Sarbahi, James Scarritt,Manuel Vogt, Judith Vorrath, Jürg Weder, andChristoph Zürcher. Luc Girardin implemented thesoftware for the online expert survey. The data proj-ect relied on f inancial support from UCLA’sInternational Institute and the Swiss National

Science Foundation through the project“Democratizing Divided Societies in BadNeighborhoods.” For encouraging comments andcriticisms, we are grateful to Michael Ross as wellas audiences at the department of sociology of theUniversity of Arizona, the Conference onDisaggregating the Study of Civil War andTransnational Violence held at the University ofEssex, the Program of Order, Conflict, and Violenceat Yale, and the Mannheim Center for EuropeanSocial Research.

1 These figures are based on the data set assembledby Wimmer and Min (2006) and concern wars withmore than 1,000 battle deaths.

Delivered by Ingenta to :UCLA Library

Tue, 07 Apr 2009 18:50:37

nent role in most conflicts reported in the dailynews—from Iraq to Darfur, Kenya to Tibet,Israel and Palestine to Burma. What can thesocial sciences offer to an understanding ofthese conflicts? When do lines of conflict fol-low ethnic divides and what are the causal mech-anisms linking ethnicity to conflict?

There is no satisfactory answer to these ques-tions in the burgeoning quantitative literature oncivil wars that has emerged over the past decade.The most influential school of thought dis-misses ethnicity as an explanatory factor alto-gether, arguing that ethnic grievances are toowidespread to explain the rare event of civilwar. In this view, rebels fight wherever gov-ernments are militarily weak or lootableresources can feed an insurgent organization(the greed-and-opportunity perspective). Otherscholars maintain that ethnicity does matter,and that more ethnically diverse states are morelikely to see conflict (the diversity-breeds-con-flict tradition). Yet a third group examines theconditions under which discriminated ethnicminorities will rebel (the minority-mobiliza-tion school). We argue that all three traditionstend to misconceive the relationship betweenethnicity and conflict.

To get this relationship right, we first need torecognize that the modern state is not an ethni-cally neutral actor or a mere arena for politicalcompetition, but a central object of and partic-ipant in ethnopolitical power struggles. Why isthis the case? Our answer takes an institution-alist point of departure. Contrary to empires,nation-states are governed in the name of “theirpeoples,” which provides incentives to alignpolitical loyalties along ethnic divides. To gainlegitimacy, political elites in control of execu-tive-level state power will favor co-ethnics whendeciding with whom to ally and to whom todistribute public goods. Politics will then cen-ter on the question of which ethnic group con-trols which share of executive government, andthe struggle over state power will pit ethnical-ly defined actors against each other. In thisview, ethnic politics is not exclusively a strug-gle to rectify the grievances of minority groups,as the minority-mobilization school assumes,but it is more generally and fundamentally aboutthe distribution of state power along ethnic lines.The diversity-breeds-conflict school relies ondemographic indices of heterogeneity that over-look how ethnicity relates to the state. Rather

than high degrees of diversity, it is ethnic exclu-sion from state power and competition over thespoils of government that breed ethnic conflict.

We propose a configurational model thatidentifies three constellations in which thisstruggle over the state is most likely to escalateinto armed conflict. First, armed rebellions aremore likely when the state excludes large sec-tions of the population from central state poweron the basis of their ethnic background. Second,the likelihood of infighting increases when alarge number of ethnic elites shares govern-ment power and engages in competitive rival-ry. Third, both rebellion and infighting will bemore likely and take on secessionist forms whensegments of the population have a short andtroubled history of direct rule by the center. Weexamine these hypotheses with quantitativeanalysis of all states since World War II usinga new data set on Ethnic Power Relations (EPR).This data set records all politically relevant eth-nic groups, minorities and majorities, and theirdegree of access to executive-level statepower—from total control of the government toovert political discrimination and exclusion.The EPR data set overcomes the limitations ofexisting data sets, especially the widely usedMinorities at Risk data set, which focuses exclu-sively on disadvantaged minorities and is thusunable to capture the dynamics of ethnic poli-tics at the power center. The EPR data set is alsoan improvement over conventional demographicindices of diversity that are only tangentiallyrelated to the ethnopolitical struggle over thestate.

Ethnic politics, our findings reveal, helps toexplain the dynamics of war and peace, contraryto what the greed-and-opportunity school main-tains. Second, our results demonstrate that morediverse states are not more war-prone, in con-trast to the expectations of the diversity-breeds-conflict school. Third, disaggregated analysisusing multinomial logit regressions shows thatdifferent kinds of ethnic conflicts result from dif-ferent causal processes: rebellions are morelikely the higher the share of the excluded pop-ulation; the chance of infighting increases as thenumber of power sharing elites augments; andsecessions are more frequent in incohesive statesthat lack a long history of direct rule by thecenter. We thus follow in the footsteps of otherscholars in the quantitative literature who arguethat different types of wars have different caus-

ETHNIC POLITICS AND ARMED CONFLICT—–317


Tue, 07 Apr 2009 18:50:37

es (Buhaug 2006; Sambanis 2001), and we sup-port the recent trend of closely investigatingthe various mechanisms that lead to armed con-flicts (Kalyvas 2007).

ETHNICITY AND CONFLICT:GETTING THE RELATIONSHIPRIGHT

Two major shortcomings characterize the quan-titative literature on ethnicity and violence. First,the mechanisms linking ethnicity to conflictare specified in theoretically problematic andempirically unsatisfactory ways. Second, quan-titative approaches tend to overaggregate thedependent variable and treat ethnic conflicts asthough they have uniform causes. We first dis-cuss the problem of specifying relevant mech-anisms, focusing on three prominent schools ofquantitative research on the outbreak of civilwars: greed and opportunity, ethnic diversitybreeds conflict, and minority mobilization.

The most influential articles argue that eth-nicity plays no role in predicting the onset ofcivil wars. According to authors in this tradition,the increase in ethnic conflicts during the twen-tieth century does not capture any meaningfultrend, but is due to the unfortunate tendency ofboth scholarly observers and rebels themselvesto attribute conflict to primordial ethnic identi-ties—a collective delusion of sorts (Laitin2007:20–27). More important than ethnic iden-tity or political exclusion along ethnic lines arethe material and organizational incentives tostage a rebellion against government. Accordingto Fearon and Laitin’s (2003) well-known insur-gency model, wars erupt when governmentsare weak and rebels have ample opportunitiesto hide from troops while recruiting unemployedyoung men for whatever cause: national liber-ation, revolutionary progress, the spread of truereligion, or rich bounty. Similarly, Collier andHoeffler (2004) maintain that civil wars occurwhere rebellions are most feasible, rather thanwhere actors are motivated by ethnic inequali-ty or social marginalization. More specifically,they argue that lootable economic resourcesmake organizing and sustaining a rebel organ-ization easier (see also Collier, Hoeffler, andRohner 2006).

A second group of scholars insists that eth-nicity does matter. They suggest various reasonswhy ethnically diverse states experience more

conflict. Some argue that high degrees of eth-nic diversity contradict the assumption of cul-tural homogeneity on which modernnation-states are based, thus triggering waves ofseparatist wars and ethnic cleansings (Gellner1991; Nairn 1993). Vanhanen (1999), the mostardent proponent of the diversity-breeds-conflictargument, relies on van den Berghe’s sociobi-ological theory of ethnic nepotism, according towhich humans tend to favor kin and quasi-kin,such as co-ethnics, over others. As a result,more ethnically heterogeneous states will havemore conflict. Finally, Sambanis (2001) drawson organizational economy models to arguethat more ethnically divided societies face high-er risks of ethnic war because shared ethnicitydecreases the collective action costs associatedwith organizing a rebel force. Since the likeli-hood of ethnic rebellion does not depend ongroup size, he expects “the relationship betweenethnic war and ethnic divisions [to be] linear andpositive” (Sambanis 2001:266; see also Easterlyand Levine 1997).

These two positions—the greed-and-oppor-tunity school and the diversity-breeds-conflicttradition—rely on the same type of demographicdiversity indicators to test their core assumptionregarding ethnicity and conflict. Many use alinguistic fractionalization index, calculated asthe likelihood that two randomly drawn indi-viduals would speak a different language. Thisis a poor indicator for capturing the politicaldynamics associated with ethnic conflict. First,not all ethnic groups matter for politics (Chandraand Wilkinson 2008; Posner 2004). Second,ethnic conflicts are not the outcome of every-day encounters between individuals; they are theresult of interactions between the state and eth-nopolitical movements that challenge stateauthority (Cederman and Girardin 2007).

Given these conceptual and measurementproblems, it is not surprising that empiricalstudies produce conflicting results when usingfractionalization indices. Some find that ethnicfractionalization does not explain high-intensi-ty conflicts (defined as more than 1,000 battledeaths per year) (Collier and Hoeffler 2004;Fearon and Laitin 2003). Others show that eth-nic fractionalization is very important if thedependent variable includes low-intensity wars(Hegre and Sambanis 2006) or if one focuses onethnic wars (Sambanis 2001) or secessionistconflicts only (Buhaug 2006). Some find a par-

318—–AMERICAN SOCIOLOGICAL REVIEW


Tue, 07 Apr 2009 18:50:37

abolic relationship between ethnic fractional-ization and the prevalence of civil war (Elbadawiand Sambanis 2000). Still others maintain thatpolarization between two equally sized ethnicgroups, rather than fractionalization, bestexplains conflict (Montalvo and Reynal-Querol2005).2

We move beyond these demographic indica-tors of ethnic diversity in the analyses that fol-low by introducing a new data set that recordspolitically relevant groups and their access toexecutive state power. This allows for a directtest of how ethnic politics affects war and peace,rather than relying on demographic proxies farremoved from how ethnicity works in politicalpractice. Once we account for the politicaldynamics of ethnic exclusion and competition,diversity in and of itself has no effect on the like-lihood of civil conflict.

The third major approach is the minority-mobilization school. These scholars analyze therelationship between ethnicity and conflict at thegroup level, rather than the state level. Comingfrom a political mobilization perspective, Gurr(1993a) and others explore the conditions underwhich ethnic minorities protest or rebel. Theyfind various factors that account for the politi-cal behavior of ethnic groups, including, as willbe familiar to students of social movements,the strength of communal grievances and thepolitical opportunity structure provided by dif-ferent political regimes. Gurr and colleagueshave also assembled a large, worldwide dataset on these “Minorities at Risk” (MAR). TheMAR data set has produced a quantum leap inthe study of ethnic politics and has provided aninvaluable service to researchers in politicalscience (Elkins and Sides 2007; Saideman andAyres 2000; Toft 2003; Walter 2006) and soci-ology (Chai 2005; Olzak 2006).

The minority-mobilization perspective comesmuch closer than the other schools to the empir-ically observable mechanisms linking ethnici-ty to conflict. We thus incorporate some of theirinsights into the model of ethnic politics devel-oped below. Their perspective, however, is lim-ited by its focus on minority groups only. Thishas two consequences. First, the state appears

as ethnically neutral, making it impossible tograsp the dynamics of ethnic politics in thepower center. Second, the MAR coding schemedoes not fit countries with ruling minorities orcomplex coalitions of ethnically defined elites(e.g., Nigeria, India, and Chad).3 In such coun-tries, ethnic conflict will be pursued in the nameof excluded majorities (rather than minorities)or ethnic groups that share power (and are thusnot at risk). Roughly half the observations in ourdata set conform to such ethnopolitical con-stellations and thus escape the logic of the MARapproach. By reducing its focus to the politicalmobilization of discriminated minorities, theminority-mobilization model overspecifies theconditions under which ethnicity leads to con-flict.

All major schools in the quantitative litera-ture fail to specify convincing mechanisms link-ing ethnicity and conflict. They either rely on aversion of the ethnic diversity argument that isunrelated to the logic of ethnic politics, or theydefine ethnic conflicts too narrowly as a mat-ter of minority mobilization. A second problemin the existing literature is that it conceives eth-nic conflict as a unitary phenomenon caused byuniform factors.4 Qualitative comparative workshows the importance of taking different eth-nopolitical constellations into account and ofacknowledging the causal heterogeneity of theprocesses that lead to ethnic conflict. The fol-lowing four vignettes of well-known ethnic con-flicts illustrate this point.

In Ireland, when segments of the educatedCatholic middle class, inspired by the U.S. civilrights movement, mobilized against their long-standing exclusion from power, the state appa-


2 Ellingsen (2000) finds support for both a linearrelationship to fractionalization and a U-shaped rela-tionship to polarization.

3 The MAR data set tries to address these limita-tions by including five “advantaged” minorities whobenefit from political discrimination and control astate apparatus. MAR also includes a series of “com-munal contenders” (i.e., groups that share powerwith others while at the same time mobilizing inprotest or rebellion); these are mostly in Africa (Gurr1993b). Ethnically defined elites that do not mobi-lize their constituencies in protest are omitted.

4 The MAR data set comes closest to a more dis-aggregated perspective by coding different types ofethnic groups. Gurr’s (1993b) analysis, however,mostly focuses on the difference between peacefulprotest and violent rebellion, irrespective of thesegroup differences.


Tue, 07 Apr 2009 18:50:37

ratus, controlled entirely by Protestants eliteswho ruled Northern Ireland as an outpost ofthe British state, reacted with repression andintimidation. The ensuing escalation reinvigo-rated the Irish nationalist underground army,which fought to unite Northern Ireland withthe rest of the country. This in turn led to theemergence of Protestant militias and terroristgroups opposed to the nationalist project(Bardon 2001).

In Bosnia shortly before independence, theleadership of the Serbian territories withdrewfrom the provincial government they had sharedwith Croatian and Bosniak politicians.Mobilization for war proceeded quickly on bothsides. Serbian militias, supported by the armyof neighboring Yugoslavia, soon attackedCroatian and Bosniak villages that they intend-ed to incorporate into the territory of a futureSerbian state (Burg and Shoup 1999).

In January 1994, the now iconic comman-dante Marcos led a group of masked men andwomen to the main square of San Cristobal dela Casas and announced that the indigenouspeoples of Chiapas and Mexico would no longeraccept their fate as second-class citizens. Hedemanded profound constitutional, economic,and political change. Decades of political mobi-lization preceded his rebellion, including left-wing organizations fighting for land reformand members of the lower clergy inspired by lib-eration theology. The central government react-ed to this provocation by sending the army tooccupy indigenous villages that supposedly har-bored members of the Zapatista army. After aseries of armed encounters, the Zapatistas even-tually withdrew into the Lacandon jungle(Collier and Lowery Quaratiello 1994; Wimmer1995).

Most recently, in Iraq after the fall of SaddamHussein, former Baathist officers and high levelfunctionaries joined Sunni clerics, tribal lead-ers from the Sunni triangle, and foreign jihadistsin a fragile alliance to fight the new power hold-ers from the Shiite south of the country. Theystruggled against what they perceived as an ille-gitimate government controlled by Shiite apos-tates and Kurdish separatists. Opposing anyfederalization and power sharing on the nation-al level, they dreamt of restoring the ethnocrat-ic regime they once controlled. Meanwhile,factions within the Shiite block jockeyed forpower, exploiting the unpopularity of the new

government and its dependence on U.S. militarypower. The Sadr Army harnessed the support ofmarginalized urban youth to oppose power shar-ing with Sunni and Kurdish political parties,advocating instead a strong, central state underShiite command (Bengio 2004; Cole 2003;Wimmer 2003).

The factors affecting these four conflicts andthe mechanisms at play are quite different.While Irish Catholics and indigenousChiapanecos represent excluded groups thatmobilized against the state, representatives ofBosnian Serbs and Shiite Arabs were partnersin coalitional governments. Serbian Bosniakelites and Iraqi ethnoreligious factions faced adisorganized and ethnically fragmented state,while Catholics in Northern Ireland and theZapatistas in Mexico opposed an entrenchedstate apparatus. The IRA and the Bosnian Serbnationalists developed separatist agendas aimedat joining established neighboring states, whilethe Zapatistas and Iraqi groups focused onchanging ethnic power relations within existingstates. It seems doubtful that any single indica-tor can accurately grasp these different eth-nopolitical dynamics. The power configurationsare different, as are the mechanisms and logicrelating ethnicity to conflict. In the followingdiscussion, we introduce a configurationalapproach that links different ethnopolitical con-stellations with distinct causal pathways lead-ing to specific types of ethnic conflict.

AN INSTITUTIONALIST,CONFIGURATIONAL THEORY OFETHNIC POLITICS AND CONFLICT

Our theory of ethnic politics and conflict isbased on two pillars. First, we rely on institu-tionalist theories that show how establishedstructures of political legitimacy provide incen-tives for actors to pursue certain types of polit-ical strategies. Second, our model follows aconfigurational logic. Depending on the con-figuration of political power, similar politicalinstitutions can produce different consequences,while similar consequences can result from dif-ferent constellations of power. The institution-alist part of the argument specif ies theconditions under which political loyalties willalign along ethnic cleavages; the configura-tional part explains when we expect such eth-nic politics to lead to armed violence.



Tue, 07 Apr 2009 18:50:37

INSTITUTIONAL INCENTIVES FOR ETHNIC

POLITICS

We derive the institutionalist part of the argu-ment from Wimmer’s (2002) theory of nation-state formation and ethnic politics. It states thatethnicity matters for politics, not because of auniversal, naturally-given tendency to favor(ethnic) kin over non-kin (as sociobiologistsargue), nor because of a primordial attachmentof individuals to their identities, nor because itprovides lower costs for political organization(as the political economy tradition maintains).Rather, ethnicity matters because the nation-state itself relies on ethnonational principles ofpolitical legitimacy: the state is ruled in thename of an ethnically defined people and rulersshould therefore care for “their own people.” Asa result, ethnicity and nationhood have muchgreater political significance in nation-statesthan they do in other types of polities such asempires or city-states.

Given this institutional environment, politi-cal office holders have incentives to gain legit-imacy by favoring co-ethnics or co-nationalsover others when distributing public goods andgovernment jobs; judiciary bodies have incen-tives to apply the principle of equality before thelaw more for co-ethnics or co-nationals thanfor others; the police have incentives to provideprotection for co-ethnics or co-nationals, but lessfor others; and so forth. The expectation of eth-nic preference and discrimination works theother way too. Voters prefer parties led by co-ethnics or co-nationals, delinquents hope forco-ethnic or co-national judges, and citizensprefer to be policed by co-ethnics or co-nation-als.

Not all modern nation-states are characterizedby such ethnic and national favoritism, howev-er. As we discuss elsewhere, this favoritism ismore likely in poor states that lack the resourcesfor universal inclusion, as well as in states withweak civil society institutions where other,nonethnic channels for aggregating politicalinterests and rewarding political loyalty arescarce (Wimmer 2002). In such states, politicalleaders and followers orient their strategiestoward avoiding dominance by ethnic or nation-al others—they strive for the self-determina-tion and self-rule that are at the core ofnationalist ideology. This motive is at the sametime material, political, and symbolic: “ade-quate” or “just” representation in a central gov-

ernment offers material advantages, such asaccess to government jobs and services; legaladvantages such as the benefits of full citizen-ship rights, a fair trial, and protection from arbi-trary violence; and symbolic advantages such asthe prestige of belonging to a “state-owning”ethnic or national group. The aggregate conse-quence of these strategic orientations is a strug-gle over control of the state between ethnicallydefined actors—or ethnic politics for short(Esman 1994; Rothschild 1981).

Such ethnic politics may lead to a process ofpolitical mobilization, counter-mobilization,and escalation. Political leaders appeal to theideal of self-rule and fair representationenshrined in the nation-state model to mobilizetheir followers against the threat of ethnic dom-inance by others. These demands may stir thefear of ethnic dominance among other politicalelites and their ethnic constituencies and resultin a process of counter-mobilization. The con-flicting demands may finally spiral into armedconfrontation. Our theory does not explicitlyaddress the logic of this escalation process (seeOlzak 2006; Tarrow and Tilly 2006) but seeksto specify the ethnopolitical configurations thatmake it more likely.

ETHNOPOLITICAL CONFIGURATIONS OF

POWER AND TYPES OF ETHNIC CONFLICT

To accomplish this task, we first introduce someconceptual tools to describe different configu-rations of actors and the power relations betweenthem (see Figure 1). Borrowing from Tilly’s(1978) polity model, we distinguish betweenvarious social groups that control or have accessto the central government (the inner circle ingrey), those who are excluded from govern-ment but are still citizens of the country (the nextcircle in white), and finally, the social worldbeyond the territorial boundaries of the state.Each ethnopolitical constellation of power isthus defined by three types of boundaries: (1)the territorial boundaries of a state that definewhich ethnic communities are considered alegitimate part of a state’s citizenry, (2) theboundary of inclusion separating those whoshare government power from those who are notrepresented at the highest levels of government,and (3) the division of power and the number ofethnic cleavages among the included sections ofthe population.



Tue, 07 Apr 2009 18:50:37

Each boundary can become the focus of eth-nopolitical conflict: who is included or exclud-ed from state power, how power is shared amongethnic elites and their constituencies, and whichethnic communities should be governed by astate. We can thus distinguish between threetypes of ethnic conflict, depending on which ofthese boundaries is at stake and which actors arechallenging each other over its location. Whenexcluded segments of a population fight to shiftthe boundaries of inclusion, we call these con-flicts rebellions. When ethnic elites in power arepitted against each other in a struggle over thespoils of government, we speak of infighting.Secession aims at changing the territorial bound-aries of a polity and can be pursued by bothexcluded and included groups.

WAR-PRONE CONFIGURATIONS:HYPOTHESES

Following the logic of our configurational argu-ment, we propose separate hypotheses for rebel-

lions, infighting, and secession. First, a highdegree of ethnic exclusion will increase the like-lihood of rebellion (Hypothesis 1) because itdecreases a state’s political legitimacy. Thismakes it easier for political leaders to mobilizea following among their ethnic constituenciesand challenge the government.5 We expect thatthe most war-prone configurations are ethnoc-racies, that is, the rule of an elite with a smallethnic constituency (e.g., the Tutsi in Burundi,white settlers in Rhodesia, and Sunni rule underSaddam Hussein).

Second, we assume that infighting is morelikely to occur when many partners share gov-ernment power, that is, in states characterizedby a segmented center. The greater the numberof political partners, the more likely allianceswill shift, increasing the fear of losing out in the


Figure 1. Ethnopolitical Constellations of Power and Conflict

5 For additional specifications of the mechanismsleading to successful ethnic mobilization, see Hechterand Levi (1979), Gurr (1993b), and Wimmer (1997).


Tue, 07 Apr 2009 18:50:37

ongoing struggle over the distribution of gov-ernment spoils.6 In such configurations, an elitefaction is more likely to mobilize its ethnic fol-lowers and challenge its power sharing part-ners by demanding a bigger share of thegovernment cake. In states with only one eth-nically defined elite in power, such ethnicinfighting is logically impossible. Thus, thegreater the number of power sharing elites, thegreater the likelihood of violent infighting(Hypothesis 2). We expect countries character-ized by a high degree of center segmentation,such as Lebanon and India, to be particularlyconflict-prone.

Third, we hypothesize that states with a longhistory of indirect rule are more likely to seesecessionist conflicts (Hypothesis 3). In suchstates, large segments of the population are notaccustomed to being governed directly by thepolitical center. These groups can be more eas-ily mobilized for a secessionist project with theargument that only independence will avoid thedanger or reality of alien rule (Hechter 2003).An example is Bosnia, which spent the nine-teenth and most of the twentieth century underOttoman, Habsburg, and later Yugoslavian rule.Fourth, we postulate that secession is more like-ly in large states (Hypothesis 4). Large states areless likely to have penetrated the outer reachesof their territory in the past, and thus the pop-ulation is less accustomed to direct rule.Imperial past and population size are both meas-urements of state cohesion, that is, the degreeto which the population takes a state’s territo-rial borders for granted and identifies with astate independent of who controls its govern-ment. An earlier literature in political anthro-pology refers to this aspect of an ethnopoliticalconfiguration as “institutional pluralism.”7

Secessionist groups claiming to represent powersharing partners or excluded populations aremore likely to challenge states that lack coher-ence. Low state cohesion thus reinforces thedynamics of exclusion and segmentation andleads challengers to secessionist paths.

Additional factors may halt the spiral ofmobilization, counter-mobilization, contesta-tion, and escalation and instead lead to a pathof accommodation and de-escalation. First, richstates’ governments can better accommodateprotest movements through redistribution poli-cies and by co-opting the movements’ leadershipinto the power elite, such as in the aftermath ofthe civil rights movement in the United States.The same holds true for dissatisfied membersof a power sharing arrangement: new govern-ment institutions can be created and staffedwith their followers, and new infrastructureprojects can be directed toward their ethnic con-stituency. Both rebellions and infighting, there-fore, should be less likely the greater a state’slevel of development (Hypothesis 5). Our modelincorporates one of the most robust findings inthe civil war literature (Hegre and Sambanis2006)—that civil wars happen in poor coun-tries—and gives it a new interpretation in linewith theories of contestation and violence (seeTarrow and Tilly 2006:145).

Second, the likelihood that a particular actorwill instigate conflict depends on the entirepower configuration, not just on that actor’sposition within that configuration. More specif-ically, we expect that power sharing partnersare less likely to fight each other when there isa high risk of rebellion by the excluded popu-lation. We assume that the likelihood of infight-ing decreases as the degree of exclusionincreases (Hypothesis 6) and as states becomelarger (and thus more incoherent) (Hypothesis7). Our configurational theory posits that exclu-sion and cohesion will have opposite effects ondifferent types of ethnic conflict. Ethnocracies


6 Horowitz (1985) offers many insights into themechanisms through which such elite competitionescalates into violent conflict, including mutual out-bidding of ethnic parties, the holding of a close elec-tion that resembles an ethnic census (see alsoWilkinson 2004), and the logic of military coupsand counter-coups.

7 Existing typologies are also based on exclusion,elite segmentation, and state cohesion as main aspectsof ethnopolitical configurations of power. Hechter andLevi (1979), Horowitz (1985), Lustick (1979), andWimmer (2002) distinguish highly exclusionarystates and those with high levels of elite segmenta-

tion. Anthropologists working in “complex societies”have analyzed different degrees of institutional plu-ralism (Despres 1968; Simpson 1995; Smith 1969),referring to the cohesion dimension. Cohen (1978)combines cohesion and exclusion, whileSchermerhorn (1970) combines segmentation andexclusion. Young (1976) and Rothschild (1981) offerthe most comprehensive typologies building on allthree aspects.


Tue, 07 Apr 2009 18:50:37

will have more rebellions (Hypothesis 1) but lessinfighting among the included population(Hypothesis 6); incoherent states will have moresecessions (Hypothesis 4) but less infighting(Hypothesis 7). Only a disaggregated researchdesign distinguishing between different types ofethnic conflicts can test these hypotheses.

RELATION TO EXISTING THEORETICAL

TRADITIONS AND EMPIRICAL FINDINGS

Our configurational theory incorporates andreconciles two sets of theoretical propositionsthat are usually seen as mutually exclusive.First, much debate centers on whether exclusionand segregation (the “internal colonialism”model of Hechter [1975]) or competition andincreased contact (Horowitz 1985; Olzak andNagel 1986) are more conflict-prone. Our the-ory maintains that both hierarchical exclusionand vertical competition are relevant mecha-nisms that link ethnic politics to violence, butthey affect different types of actors, as definedby actors’ positions in the ethnopolitical powerconfiguration. Our theory also specifies whatcompetition and exclusion are about: they arenot primarily about individual goods such ashousing or jobs (as maintained by competitiontheory), nor more generally the fruits of mod-ernization (as argued in Horowitz 1985). Rather,competition and exclusion concern control overthe state and the public goods and services at itsdisposal.

Our approach also avoids the popular dis-tinction between “greed” and “grievance” the-ories of civil war (introduced by Collier andHoeffler 2004). While the alliteration is cer-tainly seductive, and the dichotomy resonateswell with Western traditions of opposing thematerial to the ideal, it makes little empiricalsense. As argued above, ethnic politics simul-taneously concerns material interests, such asaccess to government controlled jobs, services,and contracts; idealist motives, such as therecognition of one’s ethnic heritage by the state;and genuine political goals, such as access tostate power. Because political domination byethnic others also affects one’s economic, legal,and symbolic standing, it is pointless to try todisentangle these intertwined and mutually rein-forcing motives (see Tarrow and Tilly 2006). Thecrucial question is not whether rebels are cool-ly calculating materialists or hot-blooded ide-

alists fighting for a cause, but rather what causaldynamics lead actors with complexly inter-twined motives down the path toward conflict.

Our institutionalist theory of ethnic config-urations and conflict builds on previous empir-ical research while extending it in newdirections. To date, no scholar has proposed ortested hypotheses regarding center segmenta-tion—that is, how the number of power sharingelites influences infighting. In line with ourhypothesis that low state cohesion is related tosecession, quantitative research based on theMAR data set (Gurr 1993b; Walter 2006) showsthat previous political autonomy predicts thelikelihood of secession at the group level.Similarly, on the basis of a new data set, Roeder(2007) demonstrates that previous provincialautonomy greatly increases the likelihood ofnationalist mobilization.8 Buhaug (2006) showsthat population size affects secessionist con-flicts only, but he offers a different explanationfor this finding.

Quantitative tests of the exclusion hypothe-sis (Hypothesis1) produce more conflictingresults. Gurr (1993b:179) uses his Minorities atRisk data to demonstrate that political disad-vantage increases the likelihood of armed rebel-lion, while political discrimination decreasesit. Using data covering all countries from 1945to 2001, Fearon and Laitin (2003:85) find thata lack of minority language rights and a con-stitutional preference for certain religiousgroups do not increase the likelihood of high-intensity civil war. Wimmer and Min (2006)also use a global data set and aggregate coun-try-level MAR data; they demonstrate that coun-tries with more politically discriminated groupsare more likely to have civil wars. Olzak(2006:124) also aggregates MAR data to thecountry level for a subset of 55 countries from1965 to 1989. She arrives at the conflictingconclusion that both ethnic discrimination andthe granting of ethnic group rights are associ-


8 Two other factors that we do not incorporate intoour theory are known to increase the likelihood ofsecession: kin groups across the border (Gurr 1993b;Saideman and Ayres 2000; see also Davis and Moore1997; but see Walter 2006) and geographic concen-tration and peripheral location (Buhaug, Cederman,and Rød 2008; Saideman and Ayres 2000; Toft 2003;Walter 2006).


Tue, 07 Apr 2009 18:50:37

ated with higher intensity of ethnic rebellion.Cederman and Girardin (2007) made a firstattempt to code ethnic groups’ access to statepower in the countries of Eurasia and foundevidence that exclusion breeds conflicts. Somehave contested this finding (Fearon, Kasara,and Laitin 2007),9 but more recently, Buhaugand colleagues (2008) confirmed the initialresults using Eurasian data that include addi-tional geo-coded variables.

Existing tests of the exclusion argument arethus rather inconclusive.10 We argue that this isbecause of measurement problems and datalimitations. Most researchers define exclusionnarrowly, focusing on a small number of minor-ity rights rather than explicitly measuring accessto state power. The corresponding data thus donot capture ethnic power relations in a broader,nonlegalistic way and depend too much on thedominant majority versus discriminated minor-ity scheme of the MAR data set. Data sets thatuse a broader definition of exclusion are limit-ed in geographic scope and purely cross-sec-tional and therefore do not record changes inethnic power relations over time. There is thusample room to improve on the existing researchto test the exclusion argument in a more ade-quate and comprehensive way. This is the aimof the new data set we have assembled.

THE ETHNIC POWER RELATIONS(EPR) DATA SET, 1946 TO 2005

The Ethnic Power Relations (EPR) data setidentifies all politically relevant ethnic cate-gories around the world and measures access toexecutive-level state power for members of theseethnic categories in all years from 1946 to 2005.For the sake of brevity, we introduce only themajor aspects of the data set here and refer

readers to the Online Supplement on the ASRWeb site (http://www2.asanet.org/journals/asr/2009/toc068.html) for more details about cod-ing procedures and rules. The data set containstwo parts. The first is a country-year data set thatcodes all politically relevant ethnic groups andtheir degree of access to central state power.11

The second is a conflict data set, based on thewidely used PRIO/Uppsala Armed ConflictData Set that includes all armed conflicts withmore than 25 battle deaths. We extend the dataset with new codings of whether rebels pur-sued ethnic or nonethnic goals, as well aswhether they aimed at secession. We then linkconflicts to politically relevant ethnic groups ifrebels claimed to fight in the name of a partic-ular ethnic community.

POLITICALLY RELEVANT ETHNIC GROUPS

AND ACCESS TO POWER

Following the constructivist, Weberian tradi-tion, we define ethnicity as a subjectively expe-rienced sense of commonality based on a beliefin common ancestry and shared culture. Thisdefinition includes ethnolinguistic, ethnoso-matic (or “racial”), and ethnoreligious groups,but not tribes and clans that conceive of ances-try in genealogical terms, nor regions that do notdefine commonality on the basis of sharedancestry. Ethnic categories may be hierarchicallynested and comprise several levels of differen-tiation, not all of which are politically relevantat a particular time. (On the notion of ethnici-ty underlying this project, see Wimmer 2008.)

An ethnic category is politically relevant if atleast one significant political actor claims to rep-resent the interests of that group in the nation-al political arena, or if members of an ethniccategory are systematically and intentionallydiscriminated against in the domain of publicpolitics. We do not distinguish between degreesof representativity of political actors who claimto speak for an ethnic group, nor do we code theheterogeneity of political positions voiced byleaders claiming to represent the same com-munity (Brubaker 2004). The coding schemeallows us to identify countries or specific peri-


09 Fearon and colleagues (2007) propose an alter-native measurement strategy that records the ethnicbackground of each country’s head of state. Thisdoes not capture broader, institutionalized structuresof inequality, however, and necessitates ad hocchanges in the data to avoid misleading codings (e.g.,Georgian dominance of the Soviet Union under Stalinor Quebecois hegemony in Canada under Trudeau).

10 Others have tested an exclusion argument forsecessionist minority rebellion only, using the MARdata set, and arrived at contradicting results as well(Saideman and Ayres 2000; Walter 2006).

11 The data set includes all 155 sovereign stateswith a population of at least 1 million and a surfacearea of at least 500,000 square kilometers as of 2005.


Tue, 07 Apr 2009 18:50:37

http://www2.asanet.org/journals/asr/2009/toc068.html

http://www2.asanet.org/journals/asr/2009/toc068.html

ods in which political objectives, alliances, ordisputes were never framed in ethnic terms,thus avoiding using an ethnic lens for countriesnot characterized by ethnic politics, such asTanzania and Korea.

Because politically relevant categories andaccess to political power may change over time,coders divided the 1946 to 2005 period andprovided separate codings for each subperiod.This was also necessary when the list of polit-ically relevant categories changed from oneyear to the next (either because certain cate-gories ceased to be or became relevant for thefirst time, or because higher or lower levels ofethnic differentiation became salient). Next, wecoded the degree of access to power enjoyed bypolitical leaders who claimed to represent var-ious groups.

We focus only on executive-level power, thatis, representation in the presidency, cabinet, andsenior posts in the administration, including thearmy. The weight given to these institutionsdepends on their de facto power in a given coun-try. In all cases, coders focused on absoluteaccess to power irrespective of the question ofunder- or overrepresentation relative to thedemographic size of an ethnic category.

We categorized all politically relevant ethnicgroups according to the degree of access tocentral state power by those who claimed torepresent them. Some held full control of theexecutive branch with no meaningful partici-pation by members of any other group, someshared power with members of other groups, andsome were excluded altogether from decision-making authority. Within each of these threecategories, coders differentiated between furthersubtypes, choosing from monopoly power, dom-inance, senior or junior partner in a power shar-ing arrangement, regional autonomy, powerless,and discriminated (see the ASR OnlineSupplement for details of the coding scheme).For the present analysis, we distinguish onlybetween power-holding groups (whatever theirshare of power) and the excluded population (fora disaggregated analysis on the group level,using the full array of power categories, seeCederman, Wimmer, and Min [2009]).

WAR CODING

The conflict data set created for this project isbased on the widely used Uppsala/PRIO Armed

Conflicts Data Set (ACD) (Gleditsch et al.2002). ACD defines armed conflict as anyarmed and organized confrontation betweengovernment troops and rebel organizations, orbetween army factions, that reaches an annualbattle-death threshold of 25 people. Massacresand genocides are not included because the vic-tims are neither organized nor armed; commu-nal riots and pogroms are excluded because thegovernment is not directly involved.

To date, the ACD has been of limited use forethnic conflict analysis because it does not con-tain information on whether a conflict should beclassified as ethnic. To overcome this limitation,we conducted new research and coded eachconflict for whether rebel organizations pur-sued ethnonationalist aims and recruited alongethnic lines. We also coded whether rebelsaimed at establishing a new independent state.

We distinguish between ethnic and nonethnicconflicts using the aims of the armed organi-zation and their recruitment and alliance struc-tures (this is in line with other ongoing codingprojects, e.g., Sambanis 2009). We identify as“ethnic” the aims of achieving ethnonationalself-determination, a more favorable ethnic bal-ance-of-power in government, ethnoregionalautonomy, the end of ethnic and racial dis-crimination, language and other cultural rights,and so forth. In ethnic wars, armed organizationsalso recruit fighters predominantly among theirleaders’ethnic group and forge alliances on thebasis of ethnic similarity.

We looked at the aims and recruitment pat-terns of each armed organization involved in aconflict separately. In some complex cases (e.g.,Afghanistan, Burma, Chad, Uganda, Angola,and Zaire), we disaggregated a conflict intosubconflicts when the nongovernmental sidemade different ethnic claims and rebel organi-zations acted independent from each other. Ourdata set thus contains a higher number of con-flicts than the original ACD data (see the ASROnline Supplement for details).

We then linked all ethnic conflicts to thepolitically relevant ethnic category in the EPRdata set. To avoid endogeneity problems, wemade sure that the coding of ethnic power rela-tions reflects the power constellation before theoutbreak of conflict in cases where politicalchanges occurred in the same year as a conflict.To test our configurational theory of ethnicconflict, we then divided ethnic conflicts into



Tue, 07 Apr 2009 18:50:37

those fought in the name of ethnic groupsexcluded from central government power (rebel-lions) and those fought in the name of powerholders (infighting). We further subdividedrebellions and infighting depending on whetherthey aimed to establish a separate, independentstate or join another existing state. This produceda fourfold typology with separatist rebellions,nonseparatist rebellions, separatist infightings,and nonseparatist infightings

Our data set includes 215 armed conflictsfought between 1946 and 2005, 110 of whichwere ethnic conflicts. Of the 215 conflicts, 60had secessionist aims, the vast majority of whichwere also ethnic in character. Among the 110ethnic conflicts, 20 were fought by groups inpower and 90 by excluded groups (see Table 1).One half of the conflicts reached the standardthreshold of civil war (more than 1,000 battledeaths in a year).

VARIABLES AND DATA SOURCES

EXCLUSION, CENTER SEGMENTATION,STATE COHESION

To test Hypothesis 1, we compute the share ofthe excluded population in the total populationthat is ethnopolitically relevant. We call this theshare of the excluded population for short. Weassume that increases in the share of the exclud-ed population have a greater effect on the like-lihood of conflict at lower levels of exclusionthan at higher levels, and we therefore use alogged transformation of this variable.12 Wemeasure the degree of center segmentation

(which according to Hypothesis 2 is associatedwith higher conflict probability) by counting thenumber of power sharing groups represented byethnic elites. The number of power sharing part-ners ranges from 1 to 14 (in India). FollowingHypothesis 3, the cohesion of a state decreas-es the longer the pre-independence history ofindirect rule in an empire and the larger thesize of the population. We rely on a measure ofa state’s past imperial history that calculatesthe percentage of years spent under imperialrule between 1816 and independence (Wimmerand Min 2006). We count as imperial rule allyears during which a territory was a colonial orimperial dependency (including of the SovietUnion and other communist empires) or theheartland of a landbased empire (e.g., Turkeyunder the Ottomans or Austria under theHabsburgs, but not the “mother country” of anempire with seaborne colonies, like Portugal).

OTHER VARIABLES

We control for other robustly significant vari-ables in civil war research, especially thoseidentified in Hegre and Sambanis’s (2006) meta-analysis. We include linguistic fractionaliza-tion (as found in Fearon and Laitin’s data set)to show its limited significance once ethnicpolitics variables are included. GDP per capi-ta13 and a state’s population size also play impor-


Table 1. The Conflict Data Set

Ethnic Conflicts

Infighting Rebellions Nonethnic Conflicts Total

Secessionist 09 48 003 060Nonsecessionist 11 42 102 155

Total Infighting/Rebellions 20 90Total 110 105 215

12 We hypothesize that the initial break with the eth-nonational principles of legitimacy of modern nation-states carries more political risk than does the shiftto an even more exclusionary ethnocracy.

13 Our GDP per capita data are in constant 2000US Dollars. Data for 5,737 observations (79 per-cent) come from Penn World Table 6.2. Using growthrates from the World Bank’s World DevelopmentIndicators provided 229 more observations (3 per-cent). Using Fearon and Laitin’s data, we calculatedannual growth rates and extended our values back to1946. Total data coverage is 7,105 observations (99.6percent).


Tue, 07 Apr 2009 18:50:37

tant roles in our theory of ethnic politics (accord-ing to Hypotheses 4 and 5).

Democratic civil peace theory states thatdemocracies are better able than other politicalregimes to solve internal disputes. Autocracies,on the other hand, can suppress rebellions byusing force or threatening mass violence. Civilwars should therefore be less likely in stronglydemocratic and strongly autocratic societies(Ellingsen 2000; Hegre et al. 2001; Mansfieldand Snyder 2005; Müller and Weede 1990).14

We use Polity IV data and the widely adoptedcutoffs of +6 and –6 to identify democracies,autocracies, and anocracies (states that are nei-ther democracies nor autocracies).

Fearon and Laitin’s (2003) insurgency modelmaintains that wars break out when govern-ment forces are weak and when mountainousterrain allows rebels to hide and retreat. Weinclude measures of mountainous terrain andprevious regime change (which should weakenthe government vis-à-vis the rebels) to evaluatetheir main argument. We adopt the mountain-ous terrain data from their data set; we defineregime change as any change in the Polity scoreof 3 points or more over the prior three years.

Ross (2003) developed a theory of how theavailability of natural resources affects differ-ent types of conflict. He expects that whenrebels can obstruct the extraction of naturalresources, as with oil, the likelihood of seces-sionist movements increases (see also Collierand Hoeffler 2004). Buhaug (2006), on the otherhand, argues that oil matters in conflicts over anexisting state because oil resources are usuallycontrolled by the central government. Thisincreases the incentives to capture a state, ratherthan to secede from it. To measure the impactof oil, we generate an oil production per capi-ta variable based on data from Wimmer andMin (2006).

MODELS AND FINDINGS

Our data set includes 7,155 country-year obser-vations covering 155 sovereign states in allyears after independence from 1946 to 2005. Weuse the standard modeling approach in the lit-erature on civil war, regressing a range of inde-

pendent variables on a binary dependent vari-able coded as 1 in the first year of an armed con-flict and 0 otherwise. We create a civil conflictonset variable that includes both ethnic andnonethnic onsets, as well as a more narrow eth-nic conflict onset variable. For the ethnic con-flict onset variable, we disaggregate further todistinguish between the political status of thegroups instigating the conflict (excluded orpower sharers) and the aims of these parties(secession or other aims).

We test our models against two versions ofthese dependent variables, both common in theliterature. The first version includes all obser-vations, including those in which another warwas already ongoing, and adds a dummy con-trol for such ongoing war. The second versiondrops ongoing war years by coding them asmissing, thereby omitting additional wars thatbegin while a first conflict is ongoing. Thiscoding of the dependent variable results inapproximately 15 percent fewer observations. Inthis article, we present results using the first ver-sion (for models based on the second version,see the supplement on the first author’s home-page: http://www.sscnet.ucla.edu/soc/faculty/wimmer/AppendixEthnicPolitics.pdf).The results of the two models are almost iden-tical.

We control for possible time trends by includ-ing the number of peace years since the outbreakof a war, as well as a cubic spline function onpeace years following Beck, Katz, and Tucker(1998). We also add a calendar year variable tocapture possible changes in the geopolitical cli-mate over time. For the sake of space, we do notshow the time control variables in the follow-ing tables (see supplement on first author’shomepage). As a robustness check, we tested ourmodels with regional controls and without timecontrols and found no large differences in ourmain findings (again, see supplement on firstauthor’s homepage). Throughout, we specifyrobust standard errors clustered by country toaccount for the nonindependence of observa-tions from the same state. Because armed con-flict is a rare event, we also ran our modelsusing the “rare events” logit estimator and foundno substantive differences to our main findings(see supplement on first author’s homepage).

Our analysis proceeds in four steps, eachleading to a more fine-grained, disaggregatedanalysis of conflict onset. First, we determine


14 Sambanis (2001) and Reynal-Querol (2002)confirm this hypothesis for ethnic wars only.


Tue, 07 Apr 2009 18:50:37

http://www.sscnet.ucla.edu/soc/faculty/wimmer/AppendixEthnicPolitics.pdf

http://www.sscnet.ucla.edu/soc/faculty/wimmer/AppendixEthnicPolitics.pdf

whether ethnic politics matters at all in pre-dicting the onset of armed civil conflicts.Second, we focus only on ethnic conflicts whilemaintaining our global purview and keepingall country-years in the analysis. Third, we eval-uate whether exclusion and segmentation pre-dict rebellions and infighting, respectively.Finally, we disaggregate further to determinehow state cohesion affects both rebellion andinfighting and drives them toward secessionistgoals.

EXPLAINING ARMED CONFLICT

We first test whether ethnic politics matters forunderstanding conflict and peace (Hypotheses1, 2, and 3). To ensure that our results do notdepend on our coding of civil conflicts, we alsorun our model on high-intensity wars only, aswell as against war codings from the well-knowncivil war data sets assembled by Fearon andLaitin (2003) and Sambanis (2004).

Table 2 shows that ethnic politics is animportant part of the puzzle in explainingcivil wars. The results challenge greed-and-opportunity theories of civil war, according towhich ethnicity is unrelated to conflict. Thetable also demonstrates that once ethnic pol-itics is measured directly, the ethnic diversi-ty index loses significance—contrary to whatthe diversity-breeds-conflict school assumes.Rather than diversity as such, it is politicalexclusion along ethnic lines that breeds eth-nic conflict.

The share of the excluded population, thecentral variable in our configurational modelof ethnic conflict, is significant for all modelspecifications: when using Fearon and Laitin’sor Sambanis’s coding of dependent variables(i.e., excluding low-intensity wars); when drop-ping all ongoing war years from the sample orleaving them in; and with or without addi-tional control variables. Ethnic exclusion is asconsistently related to conflict as is GDP percapita, one of the most robust explanatory fac-tors in the study of civil wars (Hegre andSambanis 2006).

In contrast, the number of power sharingpartners (Hypothesis 2) does not have a robustimpact on civil war onset. This is not surpris-ing, given that only 20 of the 200 conflicts in thisanalysis were initiated by actors representingethnic groups in power. Moreover, high degrees

of exclusion should have a mitigating effect onthe likelihood of infighting (Hypothesis 6), sowe expect to see the effects of center segmen-tation only when disaggregating the dependentvariable. The imperial past variable is positivebut insignif icant (Hypothesis 3). We willdemonstrate further that the lack of state coher-ence substantially increases the likelihood ofethnic secessionist conflicts.15

How do other theories of civil war fare in ourtest? Regime change and mountainous terrainplay a key role in the insurgency model butreceive rather limited support here, althoughthe mountainous variable helps explain one cod-ing of high-intensity civil wars (Model 4) andone version of the ACD conflict coding (Model2).16 Oil production per capita is associatedwith resource competition theories and receivesmixed support (Models 2 and 5). Meanwhile, thefindings for democratic civil-peace theory aremore robust: anocracy increases the risk of con-flict in all models except those run on the high-intensity ACD wars.

EXPLAINING ETHNIC CONFLICT

This is the first time that the ethnic exclusionargument has been statistically confirmed basedon a global data set that measures degrees ofexclusion directly and at the polity level, ratherthan the group level. The robustness of thisfinding is remarkable, given that we regress onall civil conflicts in the data set. Our model ofethnic politics makes no claims to explainnonethnic wars, such as the civil war in Korea


15 Among a large number of robustness checks(available in the supplement on the first author’shomepage), we controlled for endogeneity (the pos-sibility that past conflict determines future conflict)by running models that include a variable for thenumber of past conflicts. This did not affect ourresults (see Table 4d in the supplement on the firstauthor’s homepage).

16 Sambanis (2004) and Collier and Hoeffler(2004) also find no support for the mountainousvariable—but it appears in Hegre and Sambanis’s(2006) list of the “25 most robust variables,” as doespolitical instability. We also experimented with Fearonand Laitin’s “new state” variable (results not shown),but we found it extremely sensitive to alternativecodings (e.g., three instead of two years of inde-pendence).


Tue, 07 Apr 2009 18:50:37

330—–AMERICAN SOCIOLOGICAL REVIEWTa

ble

2.E

thn

ic P

olit

ics

and

Arm

ed C

onfl

ict

(log

isti

c re

gres

sion

)

DV

: On

set

of C

onfl

icts

DV

: On

set

of E

thn

ic C

onfl

ict

Mod

el N

o.:

12

34

56

78

9

AC

D H

igh

Fea

ron

an

d

AC

D H

igh

F

earo

n a

nd

C

onfl

ict

Dat

aset

:A

CD

AC

DIn

ten

sity

Lai

tin

Sam

ban

isA

CD

AC

DIn

ten

sity

Lai

tin

Eth

nic

Poli

tics

Var

iabl

es—

Exc

lude

d po

pula

tion

.188

7**

.129

1*.2

859*

*.2

564*

*.2

792*

*.4

192*

*.3

191*

*.5

347*

*.3

667*

*(.

0513

)(.

0558

)(.

0834

)(.

0779

)(.

0808

)(.

0862

)(.

0875

)(.

1351

)(.

1214

)—

Cen

ter

segm

enta

tion

.086

2**

.058

7.0

562

.077

1.0

177

.155

4**

.112

0**

.127

2*.0

969

(.02

95)

(.03

89)

(.04

55)

(.05

86)

(.04

91)

(.03

12)

(.03

7)(.

053)

(.07

47)

—Im

peri

al p

ast

.207

5.4

579

.728

5.7

899*

.593

2.6

401

.930

1*1.

1793

1.57

61**

(.26

14)

(.28

86)

(.44

41)

(.35

68)

(.33

07)

(.44

77)

(.44

26)

(.63

04)

(.42

44)

Oth

er V

aria

bles

—L

ingu

isti

c fr

acti

onal

izat

ion

.629

8.1

244

–.02

83.0

261

1.28

00**

.256

3.5

99(.

3227

)(.

4597

)(.

4274

)(.

3989

)(.

3997

)(.

4974

)(.

6156

)—

GD

P p

er c

apit

a–.

1239

**–.

1093

**–.

1902

**–.

1267

**–.

1750

**–.

1446

**–.

1256

**–.

1921

**–.

1554

**(.

0271

)(.

0276

)(.

0546

)(.

0374

)(.

0472

)(.

0415

)(.

0448

)(.

0746

)(.

0585

)—

Popu

lati

on s

ize

.155

6**

.139

7**

.086

5.2

354*

*.2

135*

*.2

171*

*.2

102*

*.1

884*

.360

9**

(.05

59)

(.05

32)

(.06

36)

(.06

72)

(.06

16)

(.07

14)

(.06

56)

(.07

57)

(.08

94)

—M

ount

aino

us te

rrai

n.1

241*

.190

1.1

581*

.132

0.1

749

.325

8*.0

701

(.06

01)

(.11

17)

(.07

94)

(.07

65)

(.09

84)

(.14

83)

(.10

9)—

Poli

tica

l ins

tabi

lity

.345

4.4

555

.269

3.2

655

.154

4.2

979

–.04

41(.

1764

)(.

2852

)(.

2754

)(.

2412

)(.

2726

)(.

3958

)(.

3549

)—

Ano

crac

y.4

292*

*.4

014

.721

8**

.647

8**

.446

9*.5

681

.973

8**

(.16

25)

(.25

11)

(.23

69)

(.18

63)

(.22

63)

(.29

29)

(.26

14)

—O

il p

rodu

ctio

n pe

r ca

pita

.017

1**

.005

1.0

056

.017

6*.0

180*

.027

7**

.006

4(.

0063

)(.

0162

)(.

0165

)(.

0078

)(.

0091

)(.

0083

)(.

0284

)—

Ong

oing

war

–.98

32**

–.96

78**

–1.2

732*

*–2

.165

5**

–1.4

045*

*.0

359

–.06

97–.

7636

–2.2

861*

*(.

3620

)(.

3733

)(.

4690

)(.

4277

)(.

4435

)(.

6123

)(.

6166

)(.

6271

)(.

5551

)—

Con

stan

t–1

6.15

44–1

4.28

10–9

.022

5–4

9.47

49**

–46.

1852

**–3

7.12

96*

–32.

6393

*–2

6.78

55–8

0.78

09**

(10.

2646

)(1

0.51

60)

(13.

5435

)(1

4.67

98)

(14.

8129

)(1

4.62

91)

(15.

1798

)(1

7.95

73)

(17.

4279

)N

Obs

erva

tion

s6,

938

6,86

56,

865

6,03

45,

818

6,93

86,

865

6,86

56,

034

N C

onfl

ict O

nset

s20

019

782

9712

110

310

250

66

Not

e:T

ime

cont

rols

not

sho

wn;

rob

ust s

tand

ard

erro

rs in

par

enth

eses

.*

p<

.05;

**

p<

.01.


Tue, 07 Apr 2009 18:50:37

or army coups in Brazil. Because half of the con-flicts in our data set are not fought in the nameof ethnic groups, a more focused investigationneeds to exclude nonethnic conflicts, as we doin Models 6 to 9 in Table 2. We thus followSambanis (2001) who shows that, because dif-ferent factors cause ethnic and nonethnic civilwars, they should be analyzed separately (butsee Fearon and Laitin 2003).

Once we focus on ethnic conflicts only, theother two ethnic politics variables become sta-tistically significant. The number of power shar-ing groups is significant in all models except inregressions on Fearon and Laitin’s coding ofhigh-intensity wars. The imperial past variable,which measures the degree of state cohesion andshould predict secessionist conflicts only, reach-es significance in some models (we revisit thisresult further below).

Exclusion, segmentation, and incohesion arealso substantively important for the dynamics ofwar and peace. Increasing the share of theexcluded population from 6 to 32 percent (anincrease of one standard deviation from themean) results in a 25 percent increase in theprobability of ethnic conflict (calculated on thebasis of Model 7). A one standard deviationincrease in center segmentation leads to a 9percent increased risk of conflict, while a sim-ilar increase in years under imperial rule increas-es the chance of armed conflict by 13 percent.A one standard deviation increase in GDP percapita and population size, the two most robustvariables in the civil war literature, influence theprobability of war by 22 and 13 percent, respec-tively.17

The strength and robustness18 of the exclu-sion, segmentation, and cohesion variables areremarkable because the dependent variablehere does not distinguish between differenttypes of ethnic conflict. Our theory assumes,however, that infighting, rebellion, and seces-sion are caused by different ethnopolitical con-figurations and that the same variable couldtherefore have opposite effects on the likeli-hood of different types of conflict (seeHypotheses 1, 4, 6, and 7). To test this, we

disaggregate the dependent variable furtherand use multinomial logit regressions to pre-dict the onset of different types of ethnic con-flicts.

EXPLAINING REBELLION AND INFIGHTING

We first distinguish between ethnic conflictsfought in the name of excluded groups (rebel-lions) and those begun by power sharing part-ners (inf ighting). We expect that the twoprincipal aspects of ethnic politics affect rebel-lions and infighting differently. As the numberof power sharing elites increases and theiralliances therefore become more unstable, theirlikelihood of fighting wars against each othershould also increase (Hypothesis 2). Centersegmentation should have no effect, however, onrebellions by leaders who claim to representexcluded groups. The size of the excluded pop-ulation should have opposite effects on includ-ed and excluded groups: it should increase thelikelihood of rebellion (Hypothesis 1) and there-fore provide a disincentive for inf ighting(Hypothesis 6).

Table 3 shows that the greater the number ofgroups that share power, the greater the like-lihood that they will fight each other on the bat-tlefield. Infighting is also influenced, and againnegatively, by population size (Hypothesis 7).The larger (and thus more incoherent) a state’spopulation, the less likely elites can afford tofight each other to increase their share ofpower. Contrary to our expectations, infight-ing is not significantly less likely when largesegments of the population are excluded frompower nor in richer countries (inconsistentwith Hypotheses 6 and 5, respectively),although the signs of the coefficients point inthe expected direction.

The size of the excluded population doesinfluence rebellions by excluded groups(Hypothesis 1). Rebellions are less likely inrich countries (Hypothesis 5) where govern-ments can afford to redistribute state resourcesor co-opt the leaders of protest movements.Hypothesis 5 therefore receives mixed support.More populous and linguistically heterogeneousstates are more likely to see rebellions (a find-ing that is mostly driven by secessionist wars,as we will see in the next section). State coher-ence (measured through the imperial past vari-able) does not consistently predict rebellions


17 See the table of first difference, Table S3, in theASR Online Supplement.

18 For a series of robustness checks, see Tables 5in the supplement on the first author’s homepage.


Tue, 07 Apr 2009 18:50:37

or infighting (a result that we also revisitbelow).19

Among the control variables introduced inModel 2, anocracy is no longer significant.Mountainous terrain seems to matter whengroups in power fight each other, but not incountries where rebels try to overthrow the gov-ernment (as the insurgency model would pre-dict). Oil resources do not seem to entice eitherincluded or excluded groups to fight.

EXPLAINING SECESSIONIST AND

NONSECESSIONIST CONFLICTS BY REBELS

AND INFIGHTERS

We now further differentiate between seces-sionist and nonsecessionist wars. Combiningactor types with war aims generates four kindsof ethnic conflict: secessionist wars fought in thename of excluded groups (secessionist rebel-lions for short), nonsecessionist rebellions,secessionist conflict started by power sharinggroups (secessionist infighting for short), andnonsecessionist infighting. We run multinomi-al logit regressions using these four types of eth-nic conflict as possible outcomes.

The results in Table 4 support our expecta-tions. Exclusion and center segmentation havethe same effects on the likelihood of rebellionsand infighting as before, and they also predict


Table 3. Ethnic Conflicts by Actor Type (multinomial logistic regression)

Model 1 Model 2

Infighting by Rebellion by Infighting by Rebellion byType of ethnic conflict: Power Holders Excluded Power Holders Excluded

Ethnic Politics Variables—Excluded population –.0379 .5212** –.3146 .5146**

(.1659) (.0808) (.1802) (.0848)—Center segmentation .3583** .0468 .3285** .0648

(.0568) (.0387) (.0684) (.0433)—Imperial past 2.8363 .4000 3.7934* .4520

(1.5424) (.4405) (1.8819) (.4836)Other Variables—Linguistic fractionalization –.8215 1.5463** 1.1132 1.4589**

(1.1411) (.4868) (1.1328) (.4450)—GDP per capita –.2628 –.0921* –.2148 –.0967*

(.1493) (.0391) (.1248) (.0437)—Population size –.2531* .3832** –.4172** .3818**

(.1184) (.0765) (.1517) (.0826)—Mountainous terrain .6026** .0767

(.2179) (.1189)—Political instability .1255 .1751

(.6731) (.3150)—Anocracy .4277 .4566

(.4815) (.2374)—Oil production per capita .0198 .0196

(.0113) (.0116)—Ongoing war .5618 –.0881 .2301 –.1307

(1.2172) (.6913) (1.1056) (.6827)—Constant –93.2683* –23.6030 –88.1487* –26.0182

(37.7776) (15.9535) (42.1082) (16.3912)N Observations 6,935 6,935 6,865 6,865N Conflict Onsets 20 83 19 83

Note: Time controls not shown; robust standard errors in parentheses.* p < .05; ** p < .01.

19 Dropping the time controls, including the vari-able for the number of past conflicts, or running themodels with additional region controls producealmost no changes to these results (see Tables 6 in thesupplement on the first author’s homepage).


Tue, 07 Apr 2009 18:50:37

the onset of secessionist wars.20 How does statecohesion affect conflict? Conforming toHypothesis 3, having spent more years in impe-rial polities over the past two centuries increas-es the likelihood of secessionist conflictinstigated by both power sharers and the lead-ers of excluded groups. It has no effect, againconfirming our expectations, on nonsecession-ist ethnic conflicts. The size of a state’s popu-lation is also linked with secessions (Hypothesis4). Both a long imperial past and a large popu-lation size suggest the presence of population

segments accustomed to self-rule who are like-ly to resent the shift to direct rule brought aboutby a modern nation-state. As expected, popula-tion size is significant and positive for exclud-ed populations only, and the sign of thecoefficient is negative for power sharing part-ners (Hypothesis 7).21


20 This result depends on using a logged versionof the share of the excluded population. A nonloggedversion, although it does not change any results ofprevious tables, fails to come close to standard sig-nificance levels in Models 2 and 6 in Table 4 (resultsnot shown here).

Table 4. Ethnic Conflict, by Actor and Aim (multinomial logistic regression)

Secession by Secession by Infighting by Rebellion byWar Type: Power Holders Excluded Power Holders Excluded

Ethnic Politics Variables—Excluded population –.2032 .2554* –.4504 .7501**

(.3306) (.1109) (.3156) (.1277)—Center segmentation .4956** .0008 .3176** .0689

(.1164) (.0417) (.0960) (.1001)—Imperial past 14.6269** 1.9524* 1.1870 –.8041

(2.8503) (.8152) (1.6311) (.7777)Other Variables—Linguistic fractionalization 1.4433 1.9997** .9991 .9796

(1.2707) (.6431) (1.6116) (.8709)—GDP per capita –.6017 –.0226 –.1914 –.1833*

(.3302) (.0584) (.1750) (.0814)—Population size –.1882 .4835** –.7321** .2498

(.1925) (.1256) (.1841) (.1329)—Mountainous terrain .6948 .3943 .5656* –.0913

(.3751) (.2211) (.2815) (.1608)—Political instability –35.2497** .3655 1.0312 .0291

(.6728) (.5128) (.7487) (.4485)—Anocracy 1.4050 .2931 .0115 .6333

(.9854) (.3892) (.7129) (.3639)—Oil production per capita –.3692 .0016 .0126 .0296**

(.4031) (.0452) (.0088) (.0085)—Ongoing war 2.6879 –.1664 –.5972 –.0502

(2.9776) (1.0923) (1.7814) (.9068)—Constant –290.3441** –15.6566 12.0956 –45.2199

(41.4419) (22.4369) (68.2112) (23.1803)N Observations 6,865 6,865 6,865 6,865N Conflict Onsets 9 41 10 42

Note: Time controls not shown; robust standard errors in parentheses.* p < .05; ** p < .01.

21 That population size is totally insignificant inregressions on the onset of nonethnic wars (resultsnot shown) supports our interpretation of popula-tion size as a proxy for state coherence. This is con-trary to the interpretation of Fearon and Laitin, whohypothesize that large populations are logisticallyand militarily more difficult for governments to con-trol. Dropping the time controls or running the mod-els with additional region controls produces somesmall changes to these results (for details, see Tables7 in the supplement on the first author’s homepage).


Tue, 07 Apr 2009 18:50:37

Our expectations regarding the effects of lev-els of economic development, however, areagain not fully confirmed. Richer states’ gov-ernments are able to avoid nonsecessionist rebel-lions because they can afford to co-opt theleadership of ethnic protest movements, butthey do not experience less nonsecessionistinfighting. That said, the frequency of violentinfighting is rare (9 for secessionist and 10 fornonsecessionist cases). These results shouldtherefore be interpreted with some caution.

Table 4 again includes linguistic fractional-ization as a control variable. With a disaggre-gated measure of ethnic conflict as thedependent variable, we find that linguistic diver-sity is significant only in predicting secession-ist rebellions (and only in models that includeongoing war years). We therefore suggest thatlinguistic fractionalization captures—in an indi-rect and rough way—an aspect of state coher-ence. It expresses the extent to which the centralstate has linguistically assimilated its populationin past centuries; this provides an indicator ofa state’s capacity to extend its reach over a ter-ritory across a prolonged timeframe. Linguisticfractionalization should thus be linked with theconsequences of low state cohesion, such ashigher risk of secessionist conflict. Table 4shows that once ethnic politics is measured inmore adequate and direct ways, and we havereached the appropriate level of disaggregation,the effects of linguistic fractionalization areindeed very limited.

Among other control variables, anocracy andregime change again have no significant effectson any of the four types of conflict, while moun-tainous terrain is associated with infighting butnot rebellion. Oil resources increase the likeli-hood of nonsecessionist wars fought by exclud-ed groups. This is consistent with Buhaug’shypothesis that oil resources provide incentivesto capture the state but not to secede from it.

Overall, the results of these tables demon-strate that a configurational approach to thestudy of civil wars yields important insightsabout the different mechanisms that generateviolence and war. Measures of ethnic politicshave heterogeneous effects on different types ofethnic conflict, as do other key variables suchas population size and oil. Our configurationalapproach allows us to better understand why eth-nic conflicts and wars might erupt in such dif-

ferent ethnopolitical constellations as seen inBosnia, Northern Ireland, and Mexico.

Bosnian Serbs were part of a segmented powersharing arrangement within which elite compe-tition for control over the newly founded statequickly escalated to incompatible positions anddemands. The weak coherence of the formerYugoslav state, and the high degree of disiden-tification among all but the Bosniak segments ofthe population, further increased the likelihoodof conflict and gave it a secessionist form. InNorthern Ireland, however, the conflict eruptedas a struggle over the political exclusion of thelarge Catholic population. Ireland was long ruledas an internal colony of Great Britain; theNorthern parts of the island disidentified with theBritish state, increasing the likelihood that rebelswould pursue secessionist aims. In 1994 inMexico, commandante Marcos led a group of for-mer peasant activists in a rebellion against theexclusion that the indigenous populations ofChiapas had suffered for centuries. In contrast toNorthern Ireland and Bosnia, the Mexican statehad time over the past two centuries to project itssymbolic and political power over the population,who thus learned to see their membership in thestate as self-evident and legitimate. The rebelliondid not develop into a separatist endeavor, eventhough ample opportunities existed to unite withneighboring Guatemaltecan Mayas and theirrebel organizations.

CONCLUSIONS

This article identifies the conditions underwhich struggles over state power may lead toethnic conflict. The likelihood of armed con-frontation increases as the center of powerbecomes more ethnically segmented and asgreater proportions of a state’s population areexcluded from power because of their ethnicbackground. These conflicts are even morelikely, and more likely to take secessionist form,in incoherent states where the population isnot accustomed to direct rule by the politicalcenter.

These results represent a major challenge tothe greed-and-opportunity school, which dis-counts ethnicity as a relevant factor in explain-ing civil war. To be sure, our argument is not thatethnic identity or grievances, as opposed tointerests and greed, motivate people to foundand join armed organizations. Rather, ethnici-



Tue, 07 Apr 2009 18:50:37

ty may channel the pursuit of power and pres-tige along certain pathways such that the fac-tions that struggle over state control will alignalong ethnic cleavages. Ethnicity is not an aimin itself, but the organizational means throughwhich individuals struggle to gain access tostate power. Our approach specifies the incen-tive structures under which this political logicof ethnic solidarity comes into play, as well asthe conditions under which it leads to armedconflict.

Contrary to the assumptions of the diversity-breeds-conflict school, we show that ethnic con-flicts are not any more likely in diversecountries: ethnodemographic diversity indicesrarely achieve significance and do so only fora circumscribed subset of conflicts.Ethnodemographic indices, and many theoriesof conflict and peace that rely on them, brack-et the crucial fact that the state is neither a neu-tral actor nor a passive arena within which ethnicactors operate. Rather, it is both the prize overwhich contending political actors struggle anda power instrument for those who control it.

These insights have important repercussionsfor the study of ethnic diversity in general.Recently, economists and political scientistshave discovered the unwelcome consequencesof “ethnic diversity” for a range of outcomes,including economic development, public goodsprovision, and levels of social capital and gen-eralized trust. Our study shows that ethnic diver-sity indices lose much of their significance if weinclude variables that measure ethnic exclusionand competition. It is worth asking whether onewould obtain similar results if our measure-ments of ethnic exclusion, center segmentation,and state coherence were used to study eco-nomic development, public goods provision,and social capital. In a new study of economicdevelopment, we show that this is indeed thecase (Min, Cederman, and Wimmer 2009). Thispoints to the possible conclusion that econom-ic development, public goods provision, andconflict are endogenous to the ethnic powerconfigurations analyzed in this article. Theseethnopolitical configurations at the center ofstate power may shape the different trajecto-ries of economic and political development ina much more profound way than hithertoacknowledged.

Our study also goes beyond the minority-mobilization model by showing that ethnic

mobilization and conflict not only involve dis-criminated minorities fighting for their rights.Ethnic conflict often concerns the entire con-figuration of power, most importantly the ques-tion of who has access to state power and whocontrols which share of it. Our results lendthemselves to a broader perspective that is notfocused exclusively on demographic minori-ties at risk, but on the dynamics of ethnic pol-itics at the center of the state. Contrary to theminority-mobilization model, challengers aremost likely to find an armed following amongexcluded majorities, not minorities. In addi-tion, groups in power instigate an importantnumber of conflicts. The policy implicationsare obvious: when minorities rule, or manygroups share power, granting rights to minori-ties will not prevent violence. Rather, nothingless than a fundamental rearrangement of theethnopolitical configurations of power willsecure durable peace.

Andreas Wimmer is professor of sociology at theUniversity of California, Los Angeles. His researchaims to understand the dynamics of nation-state for-mation, ethnicity boundary making, and politicalconflict from a comparative perspective. He has pur-sued various methodological and analytical strate-gies, including anthropological fieldwork, networkanalysis, comparative historical work, and cross-national statistical analysis.

Lars-Erik Cederman has taught at the GraduateInstitute of International Studies in Geneva, Oxford,UCLA, and Harvard. He is now professor of inter-national conflict research at ETH Zurich. His mainresearch interests include computational modeling,quantitative and GIS-based conflict research, nation-alism, integration and disintegration processes, andhistorical sociology.

Brian Min is a PhD candidate in the Department ofPolitical Science at the University of California, LosAngeles. His research examines the provision of pub-lic goods, particularly in ethnically diverse societies.He holds a BA from Cornell University and an MPPfrom the Kennedy School of Government at HarvardUniversity.

REFERENCES

Bardon, Jonathan. 2001. A History of Ulster. London,UK: Blackstaff.

Beck, Nathaniel, Jonathan N. Katz, and RichardTucker. 1998. “Taking Time Seriously: Time-Series-Cross-Section Analysis with a BinaryDependent Variable.” Journal of Political Science42:1260–88.



Tue, 07 Apr 2009 18:50:37

Bengio, Ofra. 2004. “The New Iraq: Challenges forState Building.” Pp. 45–64 in Regime Change inIraq: The Transatlantic and Regional Dimensions,edited by C. P. E. A. Henelt. Florence, Italy:European University Institute.

Brubaker, Rogers. 2004. “Ethnicity without Groups.”Pp. 34–52 in Facing Ethnic Conflicts: Toward aNew Realism, edited by A. Wimmer, D. Horowitz,R. Goldstone, U. Joras, and C. Schetter. Lanham,MD: Rowman and Littlefield.

Buhaug, Halvard. 2006. “Relative Capability andRebel Objective in Civil War.” Journal of PeaceResearch 43:691–708.

Buhaug, Halvard, Lars-Erik Cederman, and Jan KetilRød. 2008. “Disaggregating Ethnic Conflict: ADyadic Model of Exclusion Theory.” InternationalOrganization 62:531–51.

Burg, Steven L. and Paul S. Shoup. 1999. The Warin Bosnia-Herzegowina: Ethnic Conflict andInternational Intervention. New York: Sharpe.

Cederman, Lars-Erik and Luc Girardin. 2007.“Beyond Fractionalization: Mapping Ethnicityonto Nationalist Insurgencies.” American PoliticalScience Review 101:173–85.

Cederman, Lars-Erik, Andreas Wimmer, and BrianMin. 2009. “Understanding Ethno-NationalistConflict: From Factors to Actors.” Zürich, SwissFederal Institute of Technology. Unpublished man-uscript.

Chai, Sun-Ki. 2005. “Predicting Ethnic Boundaries.”European Sociological Review 21:375–91.

Chandra, Kanchan and Steven Wilkinson. 2008.“Measuring the Effect of ‘Ethnicity.’” ComparativePolitical Studies 41:515–63.

Cohen, Ronald. 1978. “Ethnicity: Problem and Focusin Anthropology.” Annual Review of Anthropology7:397–403.

Cole, Juan. 2003. “The United States and Shi’iteReligious Factions in Post-Ba’thist Iraq.” TheMiddle East Journal 57:543–66.

Collier, George and Elisabeth Lowery Quaratiello.1994. Basta! Land and the Zapatista Rebellion inChiapas. Oakland, CA: Institute for Food andDevelopment Policy.

Collier, Paul and Anke Hoeffler. 2004. “Greed andGrievance in Civil War.” Oxford Economic Papers56:563–95.

Collier, Paul, Anke Hoeffler, and Dominic Rohner.2006. “Beyond Greed and Grievance: Feasibilityand Civil War.” Center for the Study of AfricanEconomics, Working Paper 10.

Davis, David R. and Will H. Moore. 1997. “EthnicityMatters: Transnational Ethnic Alliances andForeign Policy Behaviour.” International StudiesQuarterly 41:171–84.

Despres, Leo A. 1968. “Anthropological Theory,Cultural Pluralism, and the Study of ComplexSocieties.” Current Anthropology 9:3–26.

Easterly, William and Ross Levine. 1997. “Africa’s

Growth Tragedy: Policies and Ethnic Divisions.”The Quarterly Journal of Economics112(4):1203–50.

Elbadawi, Nicholas and Nicholas Sambanis. 2000.“Why Are There So Many Civil Wars in Africa?Understanding and Preventing Violent Conflict.”Journal of African Economics 9:244–69.

Elkins, Zachary and John Sides. 2007. “CanInstitutions Build Unity in Multi-Ethnic States?”American Political Science Review 101:693–708.

Ellingsen, Tanja. 2000. “Colorful Community orEthnic Witches’ Brew? Multiethnicity andDomestic Conflict During and After the Cold War.”Journal of Conflict Resolution 44:228–49.

Esman, Milton. 1994. Ethnic Politics. Ithaca, NY:Cornell University Press.

Fearon, James D., Kimuli Kasara, and David D.Laitin. 2007. “Ethnic Minority Rule and Civil WarOnset.” American Political Science Review101:187–93.

Fearon, James D. and David D. Laitin. 2003.“Ethnicity, Insurgency, and Civil War.” AmericanPolitical Science Review 97:1–16.

Gellner, Ernest. 1991. “Nationalism and Politics inEastern Europe.” New Left Review 189:127–43.

Gleditsch, Nils Petter, Peter Wallensteen, MikaelErikson, Margareta Sollenberg, and Håvard Strand.2002. “Armed Conflict 1946–2001: A NewDataset.” Journal of Peace Research 39:615–37.

Gurr, Ted R. 1993a. Minorities at Risk: A GlobalView of Ethnopolitical Conflict. Washington, DC:United States Institute of Peace Press.

———. 1993b. “Why Minorities Rebel: A GlobalAnalysis of Communal Mobilization and Conflictsince 1945.” International Political Science Review14:161–201.

Hechter, Michael. 1975. Internal Colonialism. TheCeltic Fringe in British National Development,1536–1966. Berkeley, CA: University of CaliforniaPress.

———. 2003. “Containing Nationalist Violence.”Pp. 283–300 in Facing Ethnic Conflicts: Towarda New Realism, edited by A. Wimmer, D. Horowitz,R. Goldstone, U. Joras, and C. Schetter. Boulder,CO: Rowman & Littlefield.

Hechter, Michael and Margaret Levi. 1979. “TheComparative Analysis of EthnoregionalMovements.” Ethnic and Racial Studies 2:260–74.

Hegre, Havard, Tanja Ellingsen, Scott Gates, andNils Petter Gleditsch. 2001. “Toward a DemocraticCivil Peace? Democracy, Political Change, andCivil War, 1816–1992.” The American PoliticalScience Review 95:33–48.

Hegre, Havard and Nicholas Sambanis. 2006.“Sensitivity Analysis of Empirical Results on CivilWar Onset.” Journal of Conflict Resolution50:508–35.

Horowitz, Donald. 1985. Ethnic Groups in Conflict.Berkeley, CA: University of California Press.



Tue, 07 Apr 2009 18:50:37

Kalyvas, Stathis N. 2007. “Civil Wars.” Pp. 416–34in Handbook of Political Science, edited by C.Boix and S. Stokes. New York: Oxford UniversityPress.

Laitin, David. 2007. Nations, States, and Violence.Oxford, UK: Oxford University Press.

Lustick, Ian. 1979. “Stability in Deeply DividedSocieties: Consociationalism versus Control.”World Politics 31:325–44.

Mansf ield, Edward D. and Jack Snyder. 2005.Electing to Fight: Why Emerging DemocraciesGo to War. Cambridge, MA: MIT Press.

Min, Brian, Lars-Erik Cederman, and AndreasWimmer. 2009. “Ethnic Exclusion, EconomicGrowth, and Civil War: A New Dataset on EthnicPower Relations in the World, 1946–2005.”University of California, Los Angeles.Unpublished manuscript.

Montalvo, José G. and Marta Reynal-Querol. 2005.“Ethnic Polarization, Potential Conflict and CivilWars.” American Economic Review 95:796–816.

Müller, Edward N. and Erich Weede. 1990. “Cross-National Variation in Political Violence: A RationalAction Approach.” Journal of Conflict Resolution34:624–51.

Nairn, Tim. 1993. “All Bosnians Now?” DissentFall:403–10.

Olzak, Susan. 2006. The Global Dynamics of Raceand Ethnic Mobilization. Stanford, CA: StanfordUniversity Press.

Olzak, Susan and Joane Nagel. 1986. “CompetitiveEthnic Relations.” New York: Academic Press.

Posner, Daniel. 2004. “Measuring EthnicFractionalization in Africa.” American Journal ofPolitical Science 48:849–63.

Reynal-Querol, Marta. 2002. “Ethnicity, PoliticalSystems, and Civil Wars.” Journal of ConflictResolution 46:29–54.

Roeder, Philip G. 2007. Where Nation-States ComeFrom: Institutional Change in the Age ofNationalism. Princeton, NJ: Princeton UniversityPress.

Ross, Michael. 2003. “Oil, Drugs, and Diamonds:How Do Natural Resources Vary in Their Impacton Civil War?” Pp. 47–70 in Beyond Greed andGrievance: The Political Economy of ArmedConflict, edited by K. Ballentine and J. Sherman.Boulder, CO: Lynne Rienner.

Rothschild, Joseph. 1981. Ethnopolitics: AConceptual Framework. New York: Columbia.

Saideman, Stephen M. and R. William Ayres. 2000.“Determining the Causes of Irredentism: LogitAnalyses of Minorities At Risk Data from the1980s and 1990s.” The Journal of Politics62:1126–44.

Sambanis, Nicholas. 2001. “Do Ethnic and NonethnicCivil Wars Have the Same Causes?” Journal ofConflict Resolution 45:259–82.

———. 2004. “What is Civil War? Conceptual and

Empirical Complexities of an OperationalDef inition.” Journal of Conflict Resolution48:814–58.

———. 2009. “What is an Ethnic War?Organization and Interests in Insurgencies.” YaleUniversity, Department of Political Science, NewHaven, CT. Unpublished manuscript.

Schermerhorn, Richard A. 1970. Comparative EthnicRelations: A Framework for Theory and Research.New York: Random House.

Simpson, Jacqueline C. 1995. “Pluralism: TheEvolution of a Nebulous Concept.” AmericanBehavioural Scientist 38:459–77.

Smith, Michael G. 1969. “Institutional and PoliticalConditions of Pluralism.” Pp. 27–66 in Pluralismin Africa, edited by L. Kuper and M. G. Smith.Berkeley, CA: University of California Press.

Tarrow, Sidney and Charles Tilly. 2006. ContentiousPolitics. Cambridge, UK: Cambridge UniversityPress.

Tilly, Charles. 1978. From Mobilization ToRevolution. Reading, UK: Addison-Wesley.

Toft, Monica. 2003. The Geography of EthnicViolence: Identity, Interests, and the Indivisibilityof Territory. Princeton, NJ: Princeton UniversityPress.

Vanhanen, Tatu. 1999. “Domestic Ethnic Conflictand Ethnic Nepotism: A Comparative Analysis.”Journal of Peace Research 36:55–73.

Walter, Barbara. 2006. “Information, Uncertainty,and the Decision to Secede.” InternationalOrganization 60:105–35.

Wilkinson, Steven. 2004. Votes and Violence:Electoral Competition and Ethnic Riots in India.Cambridge, UK: Cambridge University Press.

Wimmer, Andreas. 1995. “Die erneute Rebellion derGehenkten, Chiapas 1994.” 1999. Zeitschrift fürSozialgeschichte des 20. und 21. Jahrhunderts3:59–68.

———. 1997. “Who Owns the State?Understanding Ethnic Conflict in Post-ColonialSocieties.” Nations and Nationalism 3:631–65.

———. 2002. Nationalist Exclusion and EthnicConflicts: Shadows of Modernity. Cambridge, UK:Cambridge University Press.

———. 2003. “Democracy and Ethno-ReligiousConflict in Iraq.” Survival: The InternationalInstitute for Strategic Studies Quarterly 45:111–34.

———. 2008. “The Making and Unmaking OfEthnic Boundaries: A Multi-Level Process Theory.”American Journal of Sociology 113:970–1022.

Wimmer, Andreas and Brian Min. 2006. “FromEmpire to Nation-State: Explaining Wars in theModern World, 1816–2001.” AmericanSociological Review 71:867–97.

Young, Crawford. 1976. The Politics of CulturalPluralism. Madison, WI: University of WisconsinPress.



Tue, 07 Apr 2009 18:50:37

ethnic politics and armed conflict: a configurational...

Documents