Is There a Clash of Civilizations? Evidence from Patterns ... · PDF fileSamuel Huntington’s The Clash of Civiliz- ... Is There a Clash of Civilizations? Evidence from Patterns of

711

Introduction

Samuel Huntington’s The Clash of Civiliz-ations and the Remaking of World Order hassent intellectual shock waves through theinternational affairs academic and policycommunities. According to the editors ofForeign Affairs, the article that Huntington(1993a) wrote in 1993 generated more discussion in three years than any otherarticle they had published since the 1940s.And deservedly so: of all the theoretical

frameworks that have gained visibility in thefield in the last decade, Huntington’s isamong the most thought-provoking andcontroversial.

In Huntington’s vision, the dynamics ofinterstate interaction in the post-Cold Warperiod will be primarily molded by civiliza-tional factors. While the state continues to bethe principal actor in the internationaldomain, its identity and interests will beshaped by its civilizational heritage. And itsbehavior will be affected accordingly: conflictbetween civilizations and cooperation withincivilizations is Huntington’s prediction forthe future of world politics. ‘In this newworld order’, Huntington (1996: 28) main-tains, ‘the most pervasive, important, anddangerous conflicts will not be between socialclasses, rich and poor, or other economically

© 2002 Journal of Peace Research,vol. 39, no. 6, 2002, pp. 711–734Sage Publications (London, Thousand Oaks,CA and New Delhi)[0022-3433(200211)39:6; 711–734; 028843]

Is There a Clash of Civilizations? Evidence fromPatterns of International Conflict Involvement,1946–97*

GIACOMO CHIOZZA

Department of Political Science, Duke University

This article offers an empirical test of Huntington’s thesis in The Clash of Civilizations. Huntingtonargues that states belonging to different civilizations will have a higher propensity to be involved ininternational conflict. This effect should be more prominent in the post-Cold War period. The civiliz-ation factor should also interact with membership in different Cold War blocs, border contiguity,regime type, and levels of modernization, magnifying or depressing the basic effects of these variables.To test these hypotheses, a logit specification with King & Zeng’s solution for rareness of events is usedon the Kosimo data. The Kosimo data allow for an extension of the empirical analysis from both a tem-poral and a substantive point of view. This study shows that state interactions across the civilizationaldivide are not more conflict prone. The first eight years of the post-Cold War era also fail to give supportto Huntington’s thesis. Moreover, while the civilization factor modifies the effects of border contiguityand regime type, this is not sufficient to generate conditions under which differences in civilizationalheritage are associated with greater risks of conflict.

* I would like to thank Chris Gelpi, Hein Goemans, JoeGrieco, Errol Henderson, and Robert Keohane for helpfulcomments to previous versions of this project. The statisti-cal analysis was performed using Stata 7 and King & Zeng’s(2001a,b) Relogit program. A replication dataset is avail-able at the JPR website: http://www.prio.no/jpr/datasets.asp. An appendix containing additional analyses isalso available at the JPR website. Mistakes, omissions, andother infelicities are my own responsibility.

04chiozza (ds) 2/10/02 8:41 am Page 711

http:\\www.sagepublications.com

defined groups, but between peoples belong-ing to different cultural entities’.

Drawing on a vast array of historical andphilosophical sources, Huntington (1996:43) defines a civilization as ‘the highestcultural grouping of people and the broadestlevel of cultural identity people have short ofthat which distinguishes humans from otherspecies. It is defined both by common objec-tive elements, such as language, history,religion, customs, institutions, and by thesubjective self-identification of people.’ Thisperspective differentiates the Clash-of-Civilizations (CofC) thesis from othercurrent arguments about the emergence ofnationalism and ethnic movements, forHuntington’s argument is not just about theemergence of particularistic identities, be theynational, ethnic, or religious. It is a novelargument about shifting loyalties: away fromthe nation-state and towards larger identitygroups that transcend national boundaries.

In this vein, identification with largercultural communities should replacenarrowly defined allegiances to the nation-state. Patterns of enmity between countrieswhich are members of the same civilization –Georgia and Russia, or China and Vietnamare Huntington’s (1996: 155–156) examples– are bound to disappear, no matter whattheir current discrepancies and misunder-standings might be.1 At the same time,patterns of cooperation between countries ofdifferent civilization are likely to becomeshallower, no matter how consolidated theymight seem to be. The logic that drives thisprocess is to be found in a psychological needfor identification: people yearn for group-ings, and blood, faith, and beliefs are theirmost basic and persistent sources.

Several indicators are invoked to substan-tiate these claims. In a fascinating tapestry,

Huntington weaves together historicalevents, state policies, interstate dynamics,mass attitudes, and elite opinions to showthat the civilizational factor is what makessense of them all at a single stroke. ‘If notcivilizations, what?’, Huntington (1993b)provocatively asks, stirring up worldwidedebate.

But, while controversies abound, system-atic empirical analyses are scarcer, and gener-ally unsupportive of Huntington’s (1996)argument. According to Russett, Oneal &Cox (2000), and Henderson & Tucker(2001), intercivilization dyads are less, andnot more, conflict prone.2 But, before sound-ing the death knell of the CofC thesis, weshould acknowledge the limitations of theseempirical tests. First, their coverage of thepost-Cold War period is limited: both studiesstop their analyses in 1992. This might beproblematic insofar as Huntington (1996:13) himself acknowledges that his civiliza-tional approach is not a work of socialscience, but only an interpretation of thelikely evolution of global politics after theCold War. As he concedes (1996: 14), thecivilizational approach is likely to be ill suitedto account for conflict patterns in earliertemporal domains. Given the strict temporalconstraint of those studies, the potentialimpact of the civilization factor might havebeen swamped in a vast array of noisy data ormight not yet have been sufficiently pro-nounced to be detected by large-N statisticalanalyses.

Second, the Cold War poses a demandingempirical challenge for the CofC thesis, forany alleged intercivilizational conflict mightjust have been an epiphenomenon of theconfrontation between the Free World andthe Communist blocs. This is obviously true for the rivalry between the United Statesand the Soviet Union, a Western and an

j ournal o f PE AC E RE S E A RC H volume 39 / number 6 / november 2002712

1 This expectation notwithstanding, Huntington (1996:312–318) depicts a scenario in which a conflict betweenChina and Vietnam sets off a larger civilizational confla-gration.

2 Gurr (1994) and Fox (2001) assess the CofC thesis withrespect to ethnic conflict within states. Both studies findlittle support for Huntington’s (1996) argument.


Orthodox country in civilizational terms.And this is potentially true for many inter-national crises that took place across the IronCurtain. Observing that the likelihood of adispute in intercivilizational dyads waxes andwanes in line with the intensity of the ColdWar, as Russett, Oneal & Cox (2000: 600)do, may not be a sufficient basis for refuta-tion of Huntington’s (1996: 39) argumentabout the restraining effects of the Cold Waron intercivilizational conflict.

Third, previous studies have not investi-gated the possibility that the civilizationfactor might influence the probability of con-flict by altering the operation of other causalvariables. The civilizational status of a pair ofcountries might be a ‘moderator’ variablethat magnifies or depresses the effects of keypredictors of conflict. Should that be thecase, there might be conditions under whichcountries of different civilization mightexperience higher risks of military confron-tation than same-civilization countries.

In this article, a new test is offered ofHuntington’s CofC argument, one that seeksto address the potential drawbacks attendanton earlier investigations of the CofC thesis.This test introduces four specific inno-vations. First, it employs a new conflictdataset, the Kosimo dataset (Pfetsch &Rohloff, 2000), that covers a larger numberof years in the post-Cold War period than theMID and COW data used by Russett, Oneal& Cox (2000), and Henderson & Tucker(2001). This allows for an extension of thedomain of investigation by five years duringthe post-Cold War period, that is from 1993to 1997.3 Second, it differentiates thepatterns of state interaction in the Cold Waracross the civilizational and ideologicaldivides, and investigates how they interact.Third, this study evaluates whether theimpact on the probability of conflict of three

key variables, namely border contiguity,regime type, and modernization, variesacross same-civilization and different-civiliz-ation dyads. This research design offers anassessment of whether the civilizationalstatus of a dyad has an indirect, but not lessimportant, effect on conflict. Finally, a newestimator – King & Zeng’s (2001a,b) rareevents logit model – is used, which shouldallow for more precise estimates of therelevant parameters.

The discussion below proceeds in threesteps. First, there is a specification of a set ofhypotheses on the relationship between civilizational membership and internationalconflict involvement. Second, researchdesign and measurement issues are discussed.The third section puts forward an analysis ofthe results and an assessment of their impli-cations for the CofC hypotheses.

This analysis shows that, as Huntington’scritics have suggested, intercivilization dyadsare in general less likely to get involved ininternational disputes. This is true even inthe post-Cold War years, the period in whichHuntington believes his thesis is most likelyto show its empirical power. Moreover, whileintercivilizational dynamics interact with,and modify, the impact of geographic conti-guity and regime type on the probability ofdispute involvement, these effects are notstrong enough to generate a conflict syn-drome for countries of different civilizations.

Hypotheses on the Clash ofCivilizations

The main thrust of Huntington’s argumentis that countries belonging to differentcivilizations are more likely to find them-selves embroiled in international confronta-tions that might lead to an outburst ofviolence. What characterizes intercivilizationdyads is the lingering presence of conflictualinteractions. As Huntington (1996: 291)writes, ‘Fault line wars are intermittent; fault

Giacomo Chiozza IS TH E R E A CL A S H O F CI V I L I Z AT I O N S? 713

3 The Kosimo dataset records data until 1999, but dataavailability on the explanatory variables limits the periodunder investigation to 1997.


line conflicts are interminable.’ Conditionsexist under which the use of force is deterredor a ceasefire is enforced, but none is likely tolast for long as the lack of trust, the diffi-culties in mutual understanding, and theindivisibility of the issues under disputemake jointly satisfactory settlements of con-troversies improbable. Moreover, should acrisis occur, it is likely to spiral to higherlevels of violence: ‘[fault line] conflicts tendto be vicious and bloody, since fundamentalissues of identity are at stake’ (Huntington,1996: 252).

This argument might be interpreted as aform of cultural realism: the venerable realisttenet of zero-sum state competition iscoupled with a new understanding of thesources of state interests.4 General values,ethnic identities, and religious beliefs informthe goals states pursue, shape the perceptionsthe decisionmakers hold, and characterizethe patterns of friendship and enmity in theinternational arena. Civilizations are notdeemed to replace states as the primarymovers of world politics. Their influence isindirect, but not less pervasive: civilizationsshape a state’s fundamental interests by defin-ing its identity. In Huntington’s framework,‘who are you?’ has become the central ques-tion of 21st-century world politics. And thatquestion is increasingly answered in civilizational terms (Huntington, 1996: 125).

Ideational factors, therefore, play a centralrole in the CofC theoretical construct.Identification with a civilization is seen as anenduring and inalienable feature that statescannot transcend. Those who try – Australia,Mexico, Russia, and Turkey are Huntington’s(1996: 139–154) examples – embark them-selves on a process that is prolonged andpolitically costly, and that has to date failed.In this respect, Huntington’s (1996) argu-ment is reminiscent of the primordialistperspective on ethnic and nationalistic

violence that posits that cultural factors are a‘given’ that inevitably set peoples apart, andonto a conflictual path (for an overview, seeFearon & Laitin, 2000; Henderson, 1997;Mercer, 1995).

The behavioral implications that aredrawn from this scenario are gloomy: as anew civilizational awareness emerges, discordand conflict are deemed to be increasing. Thesources of conflict might indeed remainthose that have plagued humankind sincetime immemorial: control of people, terri-tory, and wealth (Huntington, 1996: 129).But the cultural dimension that is nowattached to those factors makes compromiseunlikely, if not impossible. Given the indi-visible nature of the issues at stake, intenseconflict becomes a pervasive characteristic ofrelations across the civilizational divide. Thisyields the following central hypothesis:

H1: Intercivilizational dyads are morelikely to be involved in internationalconflict.

A second central aspect of Huntington’sargument is that the civilization factor isbound to become a prominent element inthe redefinition of the pattern of rivalries inthe post-Cold War era. As Huntington(1996: 125) argues, ‘Spurred by moderniz-ation, global politics is being reconfiguredalong cultural lines . . . Cultural communi-ties are replacing Cold War blocs, and thefault lines between civilizations are becomingthe central lines of conflict in global politics.’

Not that the CofC perspective is entirelydevoid of explanatory power in other tem-poral domains: relations between the Islamiccivilization and the West, for example, havebeen punctuated with tensions, incompre-hension, and ghastly violence for centuries(Huntington, 1996: 209–218). But if Hunt-ington’s argument has decisive empiricalrelevance, it should emerge most clearly inthe aftermath of the Cold War, as the bipolarrivalries have started to thaw and new


4 This label is borrowed from Henderson (1997). Needlessto say, this usage is different from Johnston’s (1995).


controversies have come to the fore. Nolonger constrained by the ideological con-frontation of the Cold War, differences in thecivilizational heritage should play a key rolein the emerging pattern of international con-flict involvement. Thus,

H2: The probability of conflict involve-ment for different-civilization dyadshas increased since the end of the ColdWar.

‘Wars between clans, tribes, ethnicgroups, religious communities, and nations’,Huntington (1996: 252) writes, ‘have beenprevalent in every era and in every civiliz-ation because they are rooted in the identi-ties of people.’ The Cold War, though, posesan exception to this pattern: ideology, notidentity, was the key factor informing powerpolitics in that period. Civilizational factorswere trumped in the confrontation betweenliberal democracy and communism. Evenconflicts occurring across different civiliz-ations were usually viewed through the prismof the Cold War (Huntington, 1996: 255). Ifthese dynamics were at work, we should beable to observe that intercivilizational dyadsthat belonged to the same Cold War blocshould have set aside their cultural andreligious differences and managed to keeppeaceful relations, while countries sharingthe same civilizational membership shouldhave been brought into conflictual inter-actions by their contrasting ideological allegiances. Thus,

H3: The Cold War bipolar confrontationdampened the propensity of conflictinvolvement for the intercivilizationdyads that belonged to the same ideo-logical bloc, and increased it for thesame-civilization dyads that belongedto different blocs.

These basic hypotheses, however, shouldbe complemented in several ways. A coreaspect of the CofC argument is that

intercivilizational dynamics might interactwith other variables and affect the proba-bility of conflict involvement indirectly.Three factors play a central role in Hunting-ton’s (1996) framework, namely border con-tiguity, regime type, and modernization.

Intercivilizational Dyads and BorderContiguityTerritorial contiguity is usually a strong pre-dictor of conflict (Bremer, 1992). The close-ness of interactions is likely to offer both theopportunity and the willingness to engage inconflict. The issue at stake is usually somepiece of territory with high symbolic valuefor both countries. This situation exacerbatespotential clashes of interests and often leadsto fault-line conflicts, the specific form ofconflictual relations between civilizations(Huntington, 1996: 252–253). In addition,the presence of minority groups that sharecivilizational identities with the populationof bordering countries might offer strongincentives for the initiation of militarizeddisputes. In forceful words, Huntington(1996: 252) draws attention to the appallingviolence of conflicts whose goal is ‘to conquerterritory and free it of other people byexpelling them, killing them, or doing both,that is, by “ethnic cleansing”.’ Yet, the sheerbrutality that characterizes these conflictsshould not obscure the fact that they are notby any means ubiquitous (Fearon & Laitin,1996).

Indeed, this hypothesis has receivedmixed support in previous research. Huth’s(1996: 80–85) findings show that irredentistclaims and support of the right to self-determination for ethnic and linguisticminorities are not strong predictors for theinitiation of territorial disputes. However,they have a substantial impact on the likeli-hood of escalation in conflict levels and onthe emergence of an enduring rivalry overdisputed territory. The reason suggested isthat countervailing factors – such as the



presence of international norms that supportthe right to self-determination only when itdoes not question the territorial integrity ofsovereign states – might dampen the proba-bility of low-level disputes. In other words,only deeply committed leaders with strongdomestic support are willing to act uponirredentist claims and initiate military con-frontation over territory. As a consequence, ifthese conflicts erupt, they are likely to bemore violent and persist over time (Huth,1996: 100–111; Huth, 1999: 55–56).

While bringing into effect importantqualifications to Huntington’s argument,these tests do not address directly its centralaspects. Huth’s (1996: 256–257) explanatoryvariable measures linguistic and ethnic com-monality between people living within dis-puted territory, or along the border, andpeople living in the challenger country. Yet,while ethnic and linguistic characteristics aretightly tied with civilization identities, theydo not include religion, which is, as Hunt-ington (1996: 253) maintains, ‘the definingcharacteristic of civilization’.

The interaction between territorial conti-guity and civilizational membership is also ofcentral importance given the fact that neigh-boring states are usually more likely tobelong to the same civilization. Strategiclocation and the economic value of contigu-ous territory are often sources of clashinginterests, no matter what the civilizationalstatus of the conflicting parties is. That is,

H4: Sharing a land border is positivelyassociated with conflict involvement.

H5: The effect of sharing a border on thelikelihood of conflict involvement ismagnified for intercivilization dyads.

Intercivilizational Dyads and DomesticRegime TypeAs Levy (1988: 662) suggested in a pre-scient article, ‘The absence of war betweendemocracies comes as close as anything we

have to an empirical law in internationalrelations’ (Russett, 1993; Ray, 1995). Thecharacter of the democratic peace pro-position has approached that of a time-invariant, universally valid, empiricalgeneralization. But while the temporaldimension of this proposition has been sub-jected to some criticism (Farber & Gowa,1995; for a response, see Gochman,1996/97; Thompson & Tucker, 1997), itsgeographical component has remained vir-tually unchallenged.5

The question is how far free governmentcan travel in time and space and whether thedemocratic peace can take root in differentcultural domains (Sartori, 1995). Insofar asliberal norms of regulated political com-petition and of mutual respect for adversariesconstitute the causal mechanisms that drivethe democratic peace (Maoz & Russett,1993; Owen, 1994; Dixon, 1994), then themere presence of electoral institutions is notsufficient to guarantee the emergence ofpeace. Not only should democracy consoli-date and become ‘the only game in town’, butwidespread consensus about liberal valuesand norms should form among elites and thegeneral public as well.

That these conditions hold – at least fornow – for most of the countries that experi-enced a democratic transition in the ThirdWave of democratization is, however, ques-tionable. Zakaria (1997) has drawn attentionto the rise of illiberal democracy, a regimetype in which electoral politics is separatedfrom constitutional liberalism. As he argues,the rule of law, the separation of powers, andrespect for the basic civil liberties are not apart of the political realities of many coun-tries with elective leaders.

Similarly, the proliferation of adjectivesused to qualify democratic rule in severalpost-authoritarian regimes shows that


5 Henderson (1997, 1998) offers a first attempt to tacklethis question.


area-study investigators often have a less san-guine assessment of the nature of democraticpolitics in those countries. Democracy is notyet accomplished if it is defined as delegative,protected, electoral, or military-dominated(Collier & Levitsky, 1997). The discrepancybetween political rights and civil liberties inthe Freedom House indicators is yet anothersign of the fact that as democracy expands inthe world, its liberal basis often lags behind(Diamond, 1996). A cogent example of thesecontradictions is offered by Iran, probablythe Middle-Eastern country with the liveliestparliament after Israel’s, but hardly yet aliberal democracy (Huntington, 1997: 9).

The implications for the democratic peacecannot be neglected. If – as Russett (1993:35) argues – decisionmakers try to follow thesame norms of conflict resolution that char-acterize domestic political processes andexpect their foreign counterparts to do like-wise, then all the qualifiers and limitationsthat accompany most non-Western democ-racies bode ill for the emergence of peaceeven among elective governments. Whatparties to a dispute might infer from thesystematic infringement of basic liberties, thepersecution of minorities, the suppression ofdissent is that violence is still a viable foreign-policy tool, no matter how competitive elec-tions might be. Under these conditions,conflict remains a possible outcome, either asa deliberate gamble from an illiberal state oras a pre-emptive move from a liberal countrythat tries to avoid being exploited.

Moreover, as free institutions and elec-toral politics expand in non-Western coun-tries, anti-Western elites are given a chance toascend to power. Their electoral platformsdeliberately blame any current shortcomingson pro-Western elites that had been in powerin the 1960s and 1970s. Nativism andnationalism have often become the electoralbanners of the ambitious politician in suchcountries. As Huntington (1996: 94) argues,‘democracy is inherently a parochializing not

a cosmopolitanizing process’. The existenceof religious, ethnic, or linguistic cleavages –and the formation of political parties alongthose lines – leads to a factionalization ofcompetitive politics, narrows the space forthe democratic compromise, and dampensthe chances of democratic consolidation.

The external consequence of these dynam-ics is that democracy makes the juxtapositionbetween countries that do not share acommon cultural heritage even more salient,as leaders become committed to nationalistpolicy positions and fear electoral retributionif they do not live up to them. The CofCthesis thus challenges the universal characterof the democratic peace and limits its validityto the Atlantic security community that hasemerged in the post-WWII world. That is,

H6: The greater the democratic characterof the regimes in a dyad, the smallerthe probability of involvement ininternational conflict.

H7: Democratic institutions foster conflictinvolvement among countries thatbelong to different civilizations.

Modernization and the Clash ofCivilizationsA third factor that might interact with civiliz-ation status is socio-economic moderniz-ation. In Huntington’s vision, we would bequite misled if we thought that moderniz-ation fostered conditions for peace through aprocess of homogenization of values, norms,and identities. Difference, and not similarity,is the consequence of the spread ofmodernization over the civilizational divide.A key indicator of these dynamics is ‘larevanche de Dieu’, the global religious revivalthat has characterized the second half of thetwentieth century, and the concomitantresurgence of indigenous identities amongthe rising elites of modernizing societies(Huntington, 1996: 95–101).

This process unfolds both at the



individual and at the societal level (Hunting-ton, 1996: 76). At the individual level,modernization disrupts the old social order,causes alienation and dislocation, and, in sodoing, creates the need for new identities.The sources of these identities are to befound not in the accomplishments in themarket place, but in those common featuresthat people inherit from their birth and thatcannot be disrupted by the dynamics ofmodernization. Religion, ethnicity, andculture – the basic components of civilization– therefore acquire new prominence andaffect the patterns of association and politicallife. Far from fostering an ineluctable trendtowards the adoption of Western values,institutions and practices, modernizationengenders a cultural and religious resurgencethat exacerbates differences across the civi-lizational divide.

These individual-level processes becomeall the more relevant insofar as moderniz-ation increases the economic, military, andpolitical power of a country. As Huntington(1996: 78) writes, ‘Modernization strength-ens those [non-Western] cultures, andreduces the relative power of the West.’ As aconsequence, these countries are likely tofind the existing hierarchy of prestige, allo-cation of resources, and territorial distri-bution in contradiction with emergentpower realities. ‘The revival of non-Westernreligions’, Huntington (1996: 101) suggests,‘is the most powerful manifestation of anti-Westernism in non-Western societies. . . . Itis a declaration of cultural independencefrom the West.’

The combination of these two dynamicshas major implications for the patterns of con-flict: contrary to the liberal tenets onmodernization and war (Morse, 1970), theCofC argument predicts that modernizationshould lead to greater bellicosity in interstaterelations in intercivilizational dyads. Increasedcapabilities and keener civilizational awarenessmake conflicts of interests more, and not less,

intractable. The liberal idea that moderniz-ation generates a pattern of interdependenceamong countries, in which the use of forcebecomes inconceivable (Morse, 1970; Oneal& Russett, 1997), is likely to be nullified inintercivilization relations, because the levels ofeconomic and social integration remain shal-lower than is the case in countries of the samecivilization. In these circumstances, modernintercivilizational dyads are less likely to beable to solve their controversies in peacefulmanners. Thus,

H8: There is a negative relationshipbetween modernization and proba-bility of conflict involvement forsame-civilization dyads.

H9: Modernization fosters conflict involve-ment among countries that belong todifferent civilizations.

Control VariablesThree control variables are included that arelikely to be correlated with the main explana-tory variables. The balance of military forcesaffects political leaders’ decisionmaking cal-culus, tilting the odds of the use of force.Following the power preponderance tra-dition (Organski & Kugler, 1980), this factormight be expected to be negatively associatedwith dispute involvement. The secondcontrol variable is the presence of majorpowers in interstate interactions. The expec-tation is that major power status is positivelyassociated with dispute involvement. Finally,we might expect that the probability of con-flict should be inversely related to the dis-tance between the countries in a dyad. Allthree control variables are likely to be corre-lated with civilizational membership as mostWestern countries are also major powers, andare also endowed with preponderant militarycapabilities. Moreover, countries of differentcivilization are more likely to be located atgreater distances. Thus,



H10: Conflict is more likely to erupt whenthe ratio of military capabilities isbalanced.

H11: Major powers are more likely to getinvolved in conflict.

H12: Distance between countries isinversely associated with conflictinvolvement.

Research Design

To test these hypotheses, a dataset was con-structed in which the unit of analysis is thedyad-year: each observation represents a pairof countries in a given year. The temporaldomain under investigation is the periodfrom 1946 through 1997. This dataset con-tains about 400,000 observations. Only3,142 (.79%), however, are engaged in aninternational dispute.

Most of these dyads comprise states thatare either too weak or too far apart to haveany meaningful international relationship,which swamps dispute interactions in a largenumber of non-events. This raises two sets ofproblems. First, the quality of the data islikely to be rather poor as investigators tradeoff precision in measurement with spatialand temporal breadth. Second, standardstatistical techniques, such as logit or probit,generate biased coefficients when applied tofinite samples of rare events, systematicallyunderestimating the probability of thephenomena under investigation.

A new estimator, however, has recentlybeen made available that addresses and solvesthese problems (King & Zeng, 2001a,b). Thisestimator allows for a selection of all the con-flict observations along with a random sampleof non-event interactions. At the same time, itcorrects the biases in logit estimation of rareevents. In this way, investigators can studysmaller samples, improve the quality of theirdata, and build more informative models.

King & Zeng’s (2001a,b) method constitutes

an innovative approach in quantitative studiesof international conflict. The traditionalapproach has so far been that of reducingsample size to manageable proportions byselecting observations on the basis of values ofthe independent variables. Maoz & Russett(1993), in particular, introduced the concept of‘politically relevant dyads’ – pairs of countriesthat either contain a major power or are con-tiguous by land – as a viable way to come toterms with the problem of the proliferation ofobservations. This selection mechanism isindeed reasonable, because conflict is usually amatter for major powers and contiguous coun-tries. But sometimes it is not, and the historicalrecord contains instances in which odd pairs –non-contiguous minor powers – fought eachother. Cases in point would include Iraq–Israel,Turkey–Cyprus, and South Africa–Angola –politically meaningful dyads in many respects,despite the lack of a shared border and majorpower status. Therefore, while allowing forcorrect inferences, sampling on values of theindependent variables fails to account for someconflict events and limits the generalizability ofthe results.

The dataset for this study was prepared intwo steps. First, a dataset was created thatcontained the population of all dyad-yearsfor all the countries listed in the Polity IV(2000) dataset. These observations werecoded with a dummy variable denoting thepresence/absence of international conflict ina given year. Next, the conflict observationswere combined with a random sample of thezero observations. The number of non-eventobservations selected is 10% of those existingin the entire population (King & Zeng,2001b: 707).6


6 Particular care was taken in trying to minimize thenumber of missing data points. Once the data sample wastaken, an extensive search was conducted to collect theadditional information required to integrate the missingdata. While it is always potentially true that there are data‘out there’ for all countries and all years, working with asample about 10% the size of the whole population madethis task more manageable and less daunting.


Two options are available to correct forthe bias introduced by subsampling the zerocases: prior correction and weighting. Whileboth perform in practically identical ways inthe Monte Carlo simulations run by King &Zeng (2001a), weighting seems to besuperior when the odds of model misspecifi-cation are high and efficiency is not aconcern. In line with King & Zeng’s advice(2001a: 145), this study opted for weighting,because this approach is more robust againstmodel misspecification. Huber/White stan-dard errors are also estimated, as is requiredwith the weighting option (King & Zeng,2001a: 154), while also clustering obser-vations by dyad, as Beck & Tucker (1997:11) suggest.

A second feature of the current researchdesign is the use of interaction effects in aregression model with a dichotomous depen-dent variable. This modeling strategy allowsfor the testing of the differential impact thatgeographic contiguity, regime type, modern-ization, and the Cold War and post-ColdWar periods have had in same-civilizationand intercivilization dyads, as posited inhypotheses H2 through H9.

The functional form of non-linearmodels, such as logit and probit, assumesthat the effect of each predictor on the proba-bility of an event is dependent upon thevalues of the remaining explanatory variables(Long, 1997). These interactive effects bydefault, however, are not sufficient to testsubstantive hypotheses on the conditionaleffects of a set of variables (Nagler, 1994:252). Therefore, should we posit that a vari-able X1 has a conditional impact on thedependent variable depending upon anotherpredictor X2, then this relationship should beexplicitly modeled including a term equal tothe product of X1 and X2 in the statisticalspecification (Friedrich, 1982; Gill, 2001).

This has a major implication for theinterpretation of the regression coefficients:once interaction terms are introduced in a

regression specification, coefficients nolonger represent the general impact of onevariable on the dependent variable, all elsebeing equal, but yield estimates that areconditional on the values of the interactingvariable – the civilization indicator in thiscase. More precisely, ‘if a variable interactswith others in the model specification, thenthe main effect coefficient is just the con-tribution of that variable assuming that allof the other interacting variable coefficientsequal zero’ (Gill, 2001: 13). Interactionterm coefficients, on the other hand, esti-mate the changes in the regression co-efficients associated with a shift from same-civilization to different-civilization. Thisshift might either depress or magnify thebaseline effect of a variable depending uponthe sign of the interaction term coefficient(Gill, 2001).7

Finally, this study controls for the effectsof the duration of the spells of peace usingBeck, Katz & Tucker’s (1998) procedure.The basic intuition underlying theirapproach is that time-series cross-sectiondata with binary dependent variables areidentical to duration data. Temporal depen-dence is, then, detected by measuring howlikely an event, in this case peace, is to ter-minate at a particular time, given that itreached that time.

Data and Measurement

Dependent variable The presence/absenceof conflict between a pair of countries in agiven year is the dependent variable for thisstudy. Ongoing confrontations are alsoincluded in the dataset, consistently with thecoding procedure advocated by Maoz &Russett (1993). The number of conflict


7 This basic rationale can also be extended to the interpre-tation of the standard errors. If we want to assess the vari-ability of a coefficient for different-civilization dyads, weshould add up the relevant standard errors, using the rulesof covariance algebra (Friedrich, 1982: 810, 828–829; Gill,2001: 14).


dyad-years is 3,142 in a sample of 42,844observations.8

This variable was coded using the set ofinterstate conflicts listed in the Kosimodataset (Pfetsch & Rohloff, 2000). Thisdataset defines conflict ‘as the clashing ofoverlapping interests (positional differences)around national values and issues (indepen-dence, self-determination, borders and terri-tory, access to or distribution of domestic orinternational power); the conflict has to be ofsome duration and magnitude of at least twoparties (states, groups of states, organizationsor organized groups) that are determined topursue their interests and win their case’(Kosimo online manual, www.kosimo.de).

The Kosimo dataset distinguishes amongfour levels of conflict intensity: war, violentcrisis, nonviolent crisis, and latent crisis.While the first three categories identifypatterns of conflict that are analogous tothose usually covered in the MID (Jones,Bremer & Singer, 1996) and the ICB(Brecher & Wilkenfeld, 1997) datasets, thecategory ‘latent crisis’ broadens the empiri-cal realm under investigation to sets ofinteractions in which ‘groups, parties, orstates question existing values, issues orobjectives that pertain to an issue ofnational interest’ (Pfetsch & Rohloff, 2000:386–387). Confrontations of this kindremain completely nonviolent. Examples oflatent crises are the UK–Argentina interac-tions over the Falklands/Malvinas islandsbefore and after the 1982 war, the US–Cubarelations since the crisis of 1962, and theterritorial dispute between Spain andBritain over Gibraltar from 1964. Thedependent variable was coded as 1 whenevera conflict of any intensity occurred. TheKosimo dataset also distinguishes betweendirect and external participants, and stateand non-state actors. External participants

and non-state actors and are not consideredin this study.9

Two advantages can be obtained fromusing the Kosimo data: first, the empiricalanalysis covers a much larger portion of thepost-Cold War period. Second, by includinglatent crises, ‘Kosimo traces the nonviolentroots of violent conflicts and checks whetherthe end of fighting was indeed the end of theconflict’ (Kosimo online manual, www.kosimo.de). If what characterizes inter-civi-lizational interactions is the lingering sense ofan ongoing crisis and the difficulty of findinga definitive solution to the conflict, then theset of events included in the Kosimo data cap-tures this dimension of the CofC thesis.Therefore, both from a temporal and from asubstantive point of view, this study takes upHuntington’s (2000) rejoinder to Russett,Oneal & Cox (2000) and engages it in a con-frontation against the empirical record.10

Intercivilization dyad This is a dummyvariable, and is equal to 1 for all the dyadsthat contain states belonging to differentcivilizations, and equal to 0 when the dyadcomprises states in the same civilization.

One of the major challenges of any studythat attempts to shed empirical light on theCofC thesis is to elaborate an operationaldefinition of civilization. As several reviewershave pointed out, Huntington’s con-ceptualization and selection of civilizations isone of the most controversial elements of hisargument (Jervis, 1997; Buzan, 1997).Katzenstein (1996: 533) makes this pointsuccinctly: ‘Although they are real, the defin-ing characteristics of civilizations (history,language, culture, tradition, religion) cannotbe grasped easily.’ As a matter of fact, Hunt-ington himself is sometimes ambivalent


8 A total of 383 observations are dropped from the analy-sis because of missing values. Four of these observations areconflicts.

9 This study excludes external participants because, whilethese actors have influence on the course of the conflict,‘the external party has to refrain from direct participationin the conflict e.g. involvement in combat’ (Kosimo onlinemanual, www.kosimo.de).10 For further discussion, see web appendix.


about the civilizational status of certaincountries. The main difficulty is that thefactors Huntington mentions in his owndefinition – religion, ethnicity, commonhistory and institutions – can yield contra-dictory country-dyad civilizational codings.The Iran–Iraq war, for example, is an intra-civilization war in Huntington’s terms.However, Iranians and Iraqis belong todifferent ethnic groups and speak differentlanguages. One might wonder whether theyidentify themselves as members of an encom-passing Islamic civilization. They are, afterall, members of two different subciviliza-tions, Arab and Persian, within the mainIslamic civilization (Huntington, 1996: 45;Welch, 1997: 202–203).

Moreover, given the emphasis on the stateas the key player in the world arena, a civiliz-ation needs to be attributed to entities thatallegedly comprise groups of differentcultural heritage. Huntington (1996: 128)glosses over this problem when he claims that‘narrower identities . . . do not necessarilyconflict with broader ones . . . a person canidentify culturally with his or her clan, ethnicgroup, nationality, religion, or civilization’.

But, despite all the ambiguities that sur-round the concept of civilization, Hunting-ton (1996: 26–27) meets the challenge byproviding a map that groups countries on thebasis of their civilization. This study reliesupon Huntington’s map, and the classifi-cation list that Henderson & Tucker (1999,2001) have elaborated from that map. Ninecivilizations are thereby identified: (1)African; (2) Buddhist; (3) Hindu; (4)Islamic; (5) Japanese; (6) Latin American; (7)Orthodox; (8) Sinic; (9) Western. Hender-son & Tucker (1999, 2001) have created aresidual category for a few countries that arenot clearly placed in any of these categories.These are mostly ‘cleft countries’, that is,countries where large portions of the popu-lation belong to different civilizations: thePhilippines, Nigeria, and Kenya are the most

prominent examples of this condition. Israelis also coded in the residual category giventhe fact that Huntington (1996: 48) neverexplicitly defines Judaism as a civilization. Inmany respects, this might appear as a roughclassification of the key explanatory concept,but it has the advantage of testing the CofCargument on its own terms.11

Post-Cold War period This is a dummyvariable that is coded as 1 for all the yearsafter 1989, and 0 otherwise.

Different Cold War bloc This is a dummyvariable that is coded as 1 for all the dyadsthat contain countries that were members ofdifferent Cold War blocs, and 0 otherwise.Post-Cold War dyad-years are also coded as0. Consistent with Huntington’s (1996:24–25) Cold War map, three blocs are dis-tinguished: the Free World, the Communist,and the Non-Aligned.

Border This is a dummy variable that iscoded as 1 when two countries share acommon border, and 0 otherwise. Data aretaken from the COW contiguity datasetfrom Bennett & Stam’s (2001) EUGeneprogram.12

Regime type This variable is measuredusing the 21-point indicator from the PolityIV (2000) dataset. To ease interpretation, itwas rescaled from 0 to 20. The democraticcharacter of a dyad is then measured by thelesser of the two countries’ scores on theregime type variable. The rationale of thiscoding rule – the weak-link assumption – is


11 The civilization indicator takes on the value of 1 when-ever the civilization of both countries in a dyad is coded inHenderson & Tucker’s (1999, 2001) residual category. Theanalysis was replicated using Russett, Oneal & Cox’s(2000) classification of civilizations. The results remainunchanged. See web appendix.12 The dyad-years after 1993, when the COW contiguitydata ends, were coded using the value the dyad had in1993.


that the character of interstate interactions ismostly determined by the country that hasfewer domestic constraints (Oneal &Russett, 1997).

Countries that are undergoing politicaltransformations are given a score of +10, themid-point on the scale. That is, polities intransition are given the benefit of the doubtby decisionmakers. Polities that are eitherexperiencing a period of interregnum orinterruption, on the other hand, are given ascore of 0, as if they were perfect autocracies.The rationale is that the presence of coun-tries in these conditions affects the wholecharacter of the dyad: decisionmakers areunlikely to grant any concession or to expectany overture. In other words, they are likelyto assume the worst and behave accordingly.

Modernization Modernization is a multi-faceted concept encompassing social, econ-omic, and political dimensions that cannotbe captured adequately with single-indicatorvariables (for an overview, see Przeworski &Limongi, 1997; Lipset, 1994). Therefore, inorder to have a more comprehensivemeasure, a factor analysis model was esti-mated using four different indicators: the logof energy consumption per capita, the per-centage of the total population living incities, the percentage of students enrolled inprimary schools, and the number of radioreceivers per 10,000 inhabitants. All thesefour indicators capture important dimen-sions of modernization, namely economicactivity, urbanization, education, and thedegree of exposure to the mass media. Theexpectation is that a modern society shouldscore high on all those dimensions.

The results of the factor analysis model arethen used to estimate factor scales, that is,estimates of the latent modernization con-struct for each observation. From this, anoperational measure of modernization wasderived that varies from –2.024 to 3.688. Toease interpretation, this variable was rescaled

so that its minimum is equal to 0 and itsmaximum is equal to 1. The level ofmodernization in a dyad is again measuredusing the weak-link assumption.13

Data on the four indicator variables weretaken from several sources: the COW capa-bility dataset from Bennett & Stam (2001),the OECD Statistical Compendium CD-ROM, the World Bank 2000 World Develop-ment Indicators CD-ROM, Mitchell’s(1998a,b,c) International Historical Statistics,Banks’s (1976) Cross-National Time-Series,and various issues of the United NationsEnergy Statistical Yearbook, the UnitedNations Demographic Yearbook, the UnitedNations Statistical Yearbook, and the UnescoStatistical Yearbook.14

Balance of military forces This variablewas coded using a procedure developed byGelpi (1999: 126). There are three indi-cators for state military capabilities: (a)number of troops, (b) military expenditures,(c) military expenditures per soldier. Foreach country and each year, these three indi-cators were standardized using its respectiveglobal total. These percentages were dis-counted using Bueno de Mesquita’s (1981:105) power projection discount factor.15 Foreach of the three indicators, the ratio of stateA’s capabilities over the sum of the capabili-ties of both countries in the dyad was com-puted. Then, the average of the threeresulting elements was taken. Finally, .50was subtracted from the average value andthe absolute value of this difference wasgenerated. This procedure creates a variable


13 Dyads with missing values on either country were treatedas if the unique value available were the minimum. Thisaffected 14 observations.14 A more detailed discussion of the measurement of themodernization variable can be found in the web appendix.Several robustness checks on the results reported in Table Iwere performed using alternative specifications of themodernization variable. Results remain substantivelyunchanged. See web appendix.15 This computation was performed for each dyad-year.Only contiguous pairs of countries were excluded.


with values between 0 and .5: a score of 0 onthis indicator reflects a balance of militaryforces, while a score of .5 means that eitherone or the other state enjoys completemilitary superiority. In other words, thegreater the score, the larger the militarypower preponderance of one of the countriesin the dyad.

Data were taken from the COW capa-bility dataset from Bennett & Stam (2001).Missing values were integrated wheneverpossible using various issues of WorldMilitary Expenditures and Arms Transfersfrom the US Arms Control and Disarma-ment Agency.

Major power This is a dummy variablethat is coded as 1 if a dyad contains at least amajor power (as defined in the COWproject), and 0 otherwise. Data are takenfrom Bennett & Stam’s (2001) EUGeneprogram.

Distance This variable measures the log ofthe number of miles between the capitalcities of the countries in a dyad. Data aretaken from Bennett & Stam’s (2001)EUGene program.16

Interaction terms To test hypotheses H2,H3, H5, H7, and H9, the intercivilizationdummy was interacted with (a) the post-Cold War dummy; (b) the Cold War blocdummy; (c) border; (d) regime type; and (e)modernization.

Peace years This variable counts thenumber of years that have elapsed since thelast international crisis between two coun-tries. Given the fact that there is no a prioritheoretical reason for expecting a linearimpact of time on the probability of conflict,the co-efficients for a natural cubic spline

with three knots were included (Beck, Katz,& Tucker, 1998: 1270; Tucker, 1999).

Data Analysis

Table I presents the estimates of two models:Model 1 is the baseline model which servesas a benchmark for comparison, while Model2 presents the specification with the inter-action effects which directly tests the hypoth-eses elaborated above.17

The first thing that might be noted aboutModel 1 is that the coefficient for the inter-civilization variable is negative: contrary toH1, the civilizational status of a dyad doesnot increase the likelihood of internationalconflict, but it even seems to reduce it. Thisfinding undermines the main thrust ofHuntington’s argument: international con-flict is not a pervasive feature of all inter-actions across the civilizational divide. Quitethe contrary, international conflict is associ-ated with the close connections and inter-actions that occur between countriesbelonging to the same civilization.

The coefficient on the post-Cold Wardummy variable is negative and statisticallysignificant. This shows that, in general, theyears since the end of the Cold War have witnessed a decrease in the likelihood ofinterstate conflict. With the exception ofmodernization, the remaining variablesincluded in Model 1 behave as expected:membership in different Cold War ideo-logical blocs and border contiguity are associ-ated with an increase in the probability ofinternational conflict, while higher levels ofdemocracy induce a reduction in the chanceof conflict. The coefficient on the moderniz-ation variable, on the other hand, is positiveand significant: modern countries are morelikely to be involved in international conflict,which contradicts the liberal case posited in


16 The dyad-years after 1993, for which no data were avail-able from Bennett & Stam’s (2001) EUGene, were codedusing the value the dyad had in 1993.

17 Models estimated using the prior correction option,instead of weighting, yield similar results. See web appen-dix.


hypothesis H8. All the parameters for tem-poral dependence turn out to be statisticallysignificant as well. From a substantive pointof view, the coefficients on the four variablesaddressing temporal dependence imply that,as the years since the previous conflictualconfrontation go by, the probability of con-flict decreases. The results on the threecontrol variables are consistent with theexpectations elaborated in hypotheses H10,H11, and H12: countries that are unequal inmilitary power and countries separated bylarge distances are less prone to conflictinvolvement, while the opposite is true fordyads that contain at least one major power.

A more comprehensive assessment of theCofC thesis can be obtained from Model 2.The coefficient on the intercivilizationdummy variable continues to be negativeand significant. But we can also notice thatthe temporal pattern of intercivilizationalconflict depicted in the CofC thesis is notborne out in the empirical realm. Thecoefficient on the interaction between thepost-Cold War and the intercivilizationdummy indicators is positive, but of no sub-stantive size and statistically undistinguish-able from zero. If we combine the maineffect and the interaction effect coefficients,we obtain a parameter equal to –.804 (se(b)


Table I. Rare Events Logit Analysis Predicting Involvement in International Conflict

Model 1 Model 2

Variables b se(b) p-value b se(b) p-value

Intercivilization dyad –0.643 0.267 0.016 –2.203 0.640 0.001Post-Cold War period –0.853 0.261 0.001 –0.874 0.378 0.021Different Cold War bloc 0.805 0.209 0.000 0.428 0.351 0.223Border 2.800 0.313 0.000 2.325 0.351 0.000Regime type –0.056 0.016 0.000 –0.103 0.023 0.000Modernization 4.703 0.928 0.000 4.667 1.557 0.003

Interaction effects with intercivilization dyadPost-Cold War period – – – 0.070 0.516 0.892Different Cold War bloc – – – 0.738 0.434 0.089Border – – – 1.160 0.469 0.013Regime type – – – 0.094 0.029 0.001Modernization – – – 0.765 1.854 0.680

Temporal dependencePeace years –1.741 0.094 0.000 –1.743 0.095 0.000Spline(1) –0.017 0.001 0.000 –0.017 0.001 0.000Spline(2) 0.012 0.001 0.000 0.012 0.001 0.000Spline(3) –0.004 0.0004 0.000 –0.004 0.0004 0.000

Control variablesBalance of military forces –1.498 0.595 0.012 –1.625 0.583 0.005Major power 2.679 0.225 0.000 2.690 0.231 0.000Distance –0.310 0.130 0.017 –0.258 0.132 0.050Constant –0.963 1.064 0.365 –0.612 1.123 0.586

Number of observations 42,461 42,461

Parameters estimated using King & Zeng’s (2001a,b) Relogit program. Weighting option used with robust standarderrors and clustering by dyad. All tests are two-tailed.


= .357, p-value < .026). That is, both fordifferent and same-civilization dyads thetransition to the post-Cold War period isassociated with a reduction in the likelihoodof conflict. The first eight years of the post-Cold War period, therefore, reveal a patternof state interactions that is at odds withCofC hypothesis H2.

The coefficient on the different-Cold Warbloc variable is positive, but statisticallyinsignificant, while the interaction termbetween the different-Cold War bloc and theintercivilization dummy indicators is alsopositive, and significant at the .09 level in atwo-tailed test. This means that, for the statesof different civilization, being on opposingideological sides entailed greater chances ofconflict during the Cold War. This is hardlysurprising if we think that the principalrivalry of the Cold War – the US–USSRdyad – is of that kind.

Turning to the other explanatory variablesin Model 2, we observe that sharing a landborder again turns out to be a strong predic-tor of conflict. Territorial contiguity is associ-ated with an increased propensity forconflictual interactions among states, ashypothesis H4 posits. Different civilizationalstatus in a dyad magnifies this effect: thecoefficient associated with the interactionterm between the intercivilization dummyand border contiguity is positive and statisti-cally significant.

This finding is indeed consistent with theCofC argument, but it is not by itself sufficient to establish that bordering inter-civilization dyads are more conflictual in allcircumstances. A cursory analysis of theresults in Model 2 shows that the sum of allthe relevant coefficients – that for inter-civilization dyad, that for border, and theirinteraction term – is about 1.282, a numbersmaller than that of the estimated effect forbordering countries from the same civiliz-ation. To clarify the interactive effects of civilizational status, Table II presents the

estimates of the predicted probabilities ofinternational conflict under different con-figurations of the explanatory variables (Gill,2001: 13).18

The focus in Table II is on the temporalpattern of conflict involvement for same- anddifferent-civilization dyads, and on bordercontiguity, while setting the level of regimetype at 10 – the mid-point on the regimetype scale – and the level of modernization atits mean value. In other words, Table IIidentifies a set of scenarios and presents thecounterfactual estimates of the probability ofconflict involvement based upon the co-efficients of Model 2.

Hypothesis H1 predicts that different-civilization dyads should be more conflictprone no matter what the values of the othervariables might be. But that is only the casewhen countries shared a border. And even inthat case, there is only one scenario in whichthe impact is substantial, that is, when thecountries in the dyad belonged to opposingideological blocs during the Cold War.Otherwise, state interactions across civiliz-ations were either more peaceful or as con-flict prone as they were within civilizations.It is worth noting that this result also obtainsin the first eight years of the post-Cold Warperiod: the probability of conflict fordifferent-civilization dyads was practicallyzero when countries did not have a landborder in common, and just slightly higherwhen they shared a border, a major contra-diction for Huntington’s (1996) thesis.

During the Cold War, countries ofdifferent civilization but on the same side ofthe ideological divide were less likely to fighteach other than countries of same civilizationbut different ideology: as Table II shows,these probabilities were equal, respectively, to.117% and .502% when there were nocommon borders, and to 3.685% and


18 These values, as well as those reported in Figure 1 andFigure 2, were computed using King & Zeng’s (2001a)method to account for estimate uncertainty.


4.891% when there were common borders.This result implies that, during the ColdWar, ideological differences were much moresalient in generating conflict than civiliza-tional ones. Moreover, Table II shows that forcountries that had common borders, a hypo-thetical shift from same-civilization todifferent-civilization was associated with a5.92% increase in the probability of conflictwhen countries were ideological opponents,but with a meager .38% increase when theywere in the same Cold War bloc.

This pattern is indeed consistent with theexpectation of CofC hypothesis H3. But,when we shift our attention to the countries

that do not share a border, a different pictureemerges: countries of different civilizationwere less likely to be embroiled in conflict irre-spective of their Cold War allegiances. Whatunderlies this result is the general reduced like-lihood of conflict among different civilizationdyads rather than any constraining effect ofthe Cold War. Therefore, the analysis of theestimated probabilities under differentalternative scenarios shows only partialsupport for the general dynamics Huntington(1996) posits with respect to the Cold Warand intercivilizational conflict.

Turning to the coefficients for the regime-type variables, we see that democracy


Table II. Probabilities of Conflict Involvement (%)

Other variables were set as follows:(1) regime type: 10 (4) major power: 1(2) modernization: mean (5) distance: mean(3) balance of military forces: mean (6) peace years: 18Values computed using King & Zeng’s (2001a,b) Relogit program.

DIFF. BLOC

DIFFERENT 0,117 0,375 0,053

SAME 0,134

DIFF. BLOC

DIFFERENT 3,685 10,810 1,675

SAME 3,302 4,891 1,422

LAND

BORDER

NO

YES

CIVILIZATI

ON

SAME BLOC POST-CW

CIVILIZATI

ON

COLD WAR

COLD WAR

POST-CWSAME BLOC

0,5020,326


dampens the probability of crisis involve-ment for countries that share the same civilizational heritage. This effect, however, isattenuated when countries in a dyad belongto different civilizations: the interaction termbetween regime type and the intercivilizationdummy variable is positive and significant, ashypothesis H7 posits. The linear combi-nation of the main effect and the interactioneffect coefficients on the regime type variableyields a parameter that, while still negative, isof much smaller magnitude, and statisticallyinsignificant (b = –.010, se(b) = .018, p-value< .598).

It would seem that democracy no longerexerts its restraining effects on the chances of conflict involvement when the countriesdo not share the same civilizational status,

contrary to the vast literature on the demo-cratic peace (Russett, 1993; Ray, 1995). Butthis finding should be interpreted in combi-nation with the general result that inter-civilization dyads are less conflict prone. AsFigure 1 clarifies, the probability of conflictinvolvement for countries of different civiliz-ation remains practically constant as thedemocratic character of a dyad increases. Forsame-civilization dyads, however, regimetype makes a large difference: as the democ-racy scores increase, the probability of con-flict drops rather quickly. Nonetheless, thebaseline probability of conflict for countriesof different civilization is much lower tobegin with. The effects of regime type on thepatterns of interstate behavior in the inter-national arena are of much lower magnitude


Pr(

Con

flict

)

Regime type

0 2 4 6 8 10 12 14 16 18 20

0%

1%

2%

3%

4%

5%

Same civilization dyads

Different civilization dyads

Figure 1. Regime Type and Conflict Involvement across Civilizations

Other variables were set as follows: (1) post-Cold War: 1 (5) balance of military forces: mean(2) different Cold War bloc: 0 (6) major power: 1(3) border: 1 (7) distance: mean(4) modernization: mean (8) peace years: 18Range of variation of regime type: 5th percentile = 0; 95th percentile = 19Values computed using King & Zeng’s (2001a,b) Relogit program.


for countries of different civilization notbecause there is anything pathological aboutdemocracy in non-Western civilizations, butbecause countries of different civilization aregenerally less likely to be embroiled in inter-national confrontations.

The final factor considered in Hunting-ton’s (1996) framework is modernization.Model 2 replicates the result found in Model1 with respect to modernization: dyads thatcomprise modern countries experiencehigher risks of conflict involvement. This istrue among both same-civilization anddifferent-civilization pairs of states: thecoefficient on the interaction term betweenmodernization and intercivilization dummyis positive, as hypothesis H9 expects, but farfrom any acceptable level of statistical signifi-cance. No matter what the civilizationalstatus of a dyad might be, as countriesdevelop economically and socially, theybecome more likely to get involved in con-flictual relations in the international arena.The impact of the modernization variableemerges from Figure 2, where the probabilityof conflict involvement for same- anddifferent-civilization dyads is plotted as afunction of modernization. The two curvesidentifying same- and different-civilizationdyads are practically undistinguishable as themodernization index moves from its 5th per-centile to its 95th.19

Finally, hypotheses H10, H11, and H12continue to be supported in Model 2: con-sistently with power preponderance theory,imbalances of military capabilities reduce theprobability of conflict involvement, whilemajor power status is a strong predictor ofconflict. Greater distances between countries

reduce the likelihood of conflictual inter-actions.

Specification ChecksThe findings discussed so far have shownlittle support for the CofC thesis. The ques-tion remains whether these results are con-tingent upon the specification of thestatistical model, or can be generalized to abroader set of conflict interactions. To thisend, the analysis was replicated on threealternative dependent variables that identifydifferent thresholds of conflict intensity, anddifferent participants. The coding of theseadditional variables is based upon the ICBand the MID data, two of the most commondatasets in quantitative InternationalRelations.20 The third additional dependentvariable is coded using a subset of the con-flict events in the Kosimo dataset thatexcludes the latent conflicts.

Table III shows that, regardless of the con-flict type under investigation, dyads ofdifferent civilizations are less conflict pronethan those of the same civilization. Across allthree models, the coefficient on the inter-civilization dummy variable is negative, andstatistically different from zero. But it is alsoimportant to notice that the overall picturethat emerges from Table III bears closeresemblance to the one obtained while ana-lyzing the Kosimo data. With the exceptionof three variables, namely the post-Cold Wardummy, the interaction effect associatedwith it, and modernization, the signs of thecoefficients remain unchanged.

It is probably not surprising that datasetsthat cover different portions of the post-ColdWar period yield slightly different resultsregarding the incidence of conflict since theend of the bipolar confrontation. But thenegative sign on the interaction termcoefficient for the post-Cold War period


19 The effect of modernization appears to be substantial asthe modernization index gets larger than .5. This finding,however, should not be overstated. Very few countries scoreso high on that variable. The number of data points onwhich the predicted probabilities are computed is, there-fore, limited. This makes the estimated effects stronglydependent on the features of the model more than on anyunderlying causal relationship (King & Zeng, 2001c).

20 The datasets employed were Grieco’s (2001) version ofthe ICB dataset and Maoz’s (2001) version of the MIDdataset.


variable shows that if there is any differencein the likelihood of conflict before and after1989, dyads of different civilizations havebecome even less prone to conflict. Themodernization variable, on the other hand,fares differently with different dependentvariables. The MID model yields results thatare fairly similar to those of Model 2 in TableI, the difference being an interaction effectcoefficient of larger magnitude, and signifi-cant at the .1 level in a two-tailed test. TheICB model and the Kosimo model excludingthe latent crises, however, show thatmodernization has no impact on conflict forcountries of same civilization, and that itmight have a positive impact – at least in theICB model – for countries of differentcivilizations. This result shows that caution

should be exercised in interpreting the effectsof modernization across different empiricaldomains. But, even granting these differ-ences, expanding the class of conflict inter-actions under investigation does not alter theoverall conclusion: the evidence in support ofHuntington’s thesis remains tenuous at most.

Conclusions

Three major conclusions follow from thisstudy. First, intercivilizational dyads are notmore likely to find themselves embroiled ininternational conflict, not even in the post-Cold War period, when civilizational conflictdynamics should be more prominent. Theyare usually less conflict prone than dyadscomprising countries of the same civilization.


Pr(

Con

flict

)

Level of modernization

.15 .20 .25 .30 .35 .40 .45 .50

0%

2%

4%

6%

Different civilization dyads

Same civilization dyads

Figure 2. Modernization and Conflict Involvement Across Civilizations

Other variables were set as follows: (1) post-Cold War: 1 (5) balance of military forces: mean(2) different Cold War bloc: 0 (6) major power: 1(3) border: 1 (7) distance: mean(4) regime type: 10 (8) peace years: 18Range of variation of modernization: 5th percentile = .145; 95th percentile = .516Values computed using King & Zeng’s (2001a,b) Relogit program.


Giacom

o Ch

iozzaI

ST

HE

RE

AC

LA

SH

OF

CIV

ILIZ

AT

ION

S?731

Table III. Alternative Specifications of the Dependent Variable

Kosimo model ICB model MID model(excluding latent crises) 1946–94 1946–92

Variables b se(b) p-value b se(b) p-value b se(b) p-value

Intercivilization dyad –1.643 0.684 0.016 –2.533 0.680 0.000 –1.112 0.410 0.007Post-Cold War period 0.353 0.356 0.322 –0.243 0.484 0.615 0.203 0.335 0.545Different Cold War bloc 0.695 0.433 0.108 0.381 0.347 0.272 0.682 0.213 0.001Border 2.540 0.385 0.000 2.440 0.348 0.000 1.826 0.211 0.000Regime type –0.123 0.029 0.000 –0.104 0.029 0.000 –0.118 0.023 0.000Modernization –0.205 1.449 0.888 –0.124 1.525 0.935 1.580 0.915 0.084

Interaction effects with intercivilization dyad

Post-Cold War period –0.936 0.508 0.065 –0.662 0.669 0.323 –0.766 0.472 0.105Different Cold War bloc 0.143 0.517 0.782 0.121 0.445 0.785 0.126 0.305 0.680Border 0.339 0.570 0.552 0.982 0.450 0.029 0.235 0.329 0.475Regime type 0.110 0.039 0.005 0.075 0.037 0.041 0.064 0.028 0.023Modernization 2.624 1.905 0.168 4.415 1.991 0.027 2.010 1.188 0.091

Temporal dependence

Peace years –1.454 0.088 0.000 –0.552 0.053 0.000 –0.658 0.037 0.000Spline(1) –0.014 0.001 0.000 –0.005 0.001 0.000 –0.007 0.001 0.000Spline(2) 0.010 0.001 0.000 0.004 0.001 0.000 0.005 0.001 0.000Spline(3) –0.003 0.0004 0.000 –0.001 0.0003 0.001 –0.002 0.0002 0.000

Control variables

Balance of military forces –0.810 0.656 0.217 –0.718 0.882 0.415 –1.842 0.458 0.000Major power 1.845 0.301 0.000 2.063 0.323 0.000 2.110 0.167 0.000Distance –0.309 0.146 0.034 –0.246 0.136 0.069 –0.533 0.087 0.000Constant –0.587 1.148 0.609 –2.353 1.151 0.041 0.903 0.699 0.197

Number of observations 42461 36364 35216

Coefficients represent estimates of a rare events logit model. Parameters estimated using King & Zeng’s (2001a,b) Relogit program. Weighting option used with robust standard errorsand clustering by dyad. All tests are two-tailed.


This finding adds to the systematic literatureto date that is unsupportive of the empiricalplausibility of the CofC thesis: interciviliza-tion interactions are not placed on a path toineluctable conflict. Second, the article mightgive special strength to the doubts in theliterature about the thesis insofar as a discon-firming pattern was identified in the post-Cold War era, that is, precisely the timeperiod in connection to which Huntingtonhimself is most confident that the CofC argu-ment should find evidentiary support. Third,while the civilization factor modifies theeffects of border contiguity and regime type,these conditional effects are not sufficient togenerate conditions under which differencesin religious, ethnic, and cultural heritage areassociated with greater risks of controversiesand conflict. At most, countries of differentcivilization are as conflict prone as countriesof the same civilization.

Other conflict dimensions central toHuntington’s (1996) argument, namelyintrastate violence and the breakup ofnations, have not been addressed in thisstudy. Civilizational and religious differencesmight indeed play a prominent role in thosecontexts, as the sadly familiar events in manyparts of the world do not fail to remind us.Nonetheless, insofar as the CofC thesisaspired to pinpoint the likely trajectory ofstate relations in the 21st century, it has notyet garnered the empirical support necessaryto make it a basis for foreign policy.

This does not necessarily mean thatrelations with China, Iran, and the Islamiccountries – a fundamental theme in Hunt-ington’s (1996) analysis – will be rosy andsmooth. While tensions and contrastsmight arise, though, the civilizational factoris unlikely to be their main underlyingcause. Civilizational differences seemunlikely to become the dominant factorthat shapes the patterns of enmity andfriendship in the international arena in theyears to come.

References

Banks, Arthur S., 1976. Cross-National TimeSeries, 1815–1973 [computer file]. ICPSRStudy 7412. Ann Arbor, MI: Inter-UniversityConsortium for Political and Social Research.

Beck, Nathaniel & Richard Tucker, 1997. ‘Con-flict in Time and Space’, Working Paper 97–8.Weatherhead Center for International Affairs,Harvard University (http://www.ciaonet.org/wps/tur01/).

Beck, Nathaniel; Jonathan N. Katz & RichardTucker, 1998. ‘Taking Time Seriously inBinary Time-Series-Cross-Section Analysis’,American Journal of Political Science 42(4):1260–1288.

Bennett, D. Scott & Allan C. Stam, 2001.EUGene: Expected Utility Generation and DataManagement Program. Version 2.013 (http://eugenesoftware.org).

Brecher, Michael & Jonathan Wilkenfeld, 1997.A Study of Crisis. Ann Arbor, MI: University ofMichigan Press.

Bremer, Stuart A., 1992. ‘Dangerous Dyads:Conditions Affecting the Likelihood of Inter-state War, 1816–1965’, Journal of Conflict Res-olution 36(2): 309–341.

Bueno de Mesquita, Bruce, 1981. The War Trap.New Haven, CT: Yale University Press.

Buzan, Barry, 1997. ‘Civilizational Realpolitik asthe New World Order?’, Survival 39(1):180–183.

Collier, David & Steven Levitsky, 1997. ‘Democ-racy with Adjectives: Conceptual Innovationin Comparative Research’, World Politics49(3): 430–451.

Diamond, Larry J., 1996. ‘Is the Third WaveOver?’, Journal of Democracy 7(3): 20–37.

Dixon, William J., 1994. ‘Democracy and thePeaceful Settlement of International Conflict’,American Political Science Review 88(1):14–32.

Farber, Henry S. & Joanne Gowa, 1995. ‘Politiesand Peace’, International Security 20(2):123–146.

Fearon, James D. & David D. Laitin, 1996.‘Explaining Interethnic Cooperation’, Ameri-can Political Science Review 90(4): 715–735.

Fearon, James D. & David D. Laitin, 2000.‘Violence and the Social Construction of



Ethnic Identity’, International Organization54(4): 845–877.

Fox, Jonathan, 2001. ‘Two Civilizations andEthnic Conflict: Islam and the West’, Journalof Peace Research 38(4): 459–472.

Friedrich, Robert J., 1982. ‘In Defense of Multi-plicative Terms in Multiple Regression Equa-tions’, American Journal of Political Science26(4): 797–833.

Gelpi, Christopher, 1999. ‘Alliances as Instru-ments of Intra-Allied Control’, in Helga Haftendorn, Robert O. Keohane & CelesteWallander, eds, Imperfect Unions: SecurityInstitutions over Time and Space. Oxford:Oxford University Press (107–139).

Gill, Jeff, 2001. ‘Interpreting Interactions andInteraction Hierarchies in Generalized LinearModels: Issues and Applications’, paper pre-sented at the 2001 Annual Meeting of theAmerican Political Science Association, SanFrancisco, CA, 30 August–2 September (http://pro.harvard.edu).

Gochman, Charles S., 1996/97. ‘Correspon-dence: Democracy and Peace’, InternationalSecurity 21(3): 177–186.

Grieco, Joseph M., 2001. ‘Repetitive MilitaryChallenges and Recurrent International Conflicts, 1918–1994’, International StudiesQuarterly 45(2): 295–316.

Gurr, Ted Robert, 1994. ‘Peoples Against theStates: Ethnopolitical Conflict and the Chang-ing World System’, International Studies Quarterly 38(3): 347–377.

Henderson, Errol A., 1997. ‘Culture or Contigu-ity: Ethnic Conflict, the Similarity of States,and the Onset of War, 1820–1989’, Journal ofConflict Resolution 41(5): 649–668.

Henderson, Errol A., 1998. ‘The DemocraticPeace Through the Lens of Culture,1820–1989’, International Studies Quarterly42(4): 461–484.

Henderson, Errol A. & Richard Tucker, 1999.The Clash of Civilizations Data Project. Version2.0 (http: //www.vanderbilt.edu/~rtucker/data/clash/member.html).

Henderson, Errol A. & Richard Tucker, 2001.‘Clear and Present Strangers: The Clash of Civilizations and International Conflict’,International Studies Quarterly 45(2):317–338.

Huntington, Samuel P., 1993a. ‘The Clash ofCivilizations?’, Foreign Affairs 72(3): 22–49.

Huntington, Samuel P., 1993b. ‘If not Civiliz-ations, What? Paradigms of the Post-ColdWar’, Foreign Affairs 72(5): 186–194.

Huntington, Samuel P., 1996. The Clash ofCivilizations and the Remaking of World Order.New York: Simon & Schuster.

Huntington, Samuel P., 1997. ‘After TwentyYears: the Future of the Third Wave’, Journalof Democracy 8(4): 3–12.

Huntington, Samuel P., 2000. ‘Try Again: AReply to Russett, Oneal & Cox’, Journal ofPeace Research 37(5): 609–610.

Huth, Paul K., 1996. Standing Your Ground:Territorial Disputes and International Conflict.Ann Arbor, MI: University of Michigan Press.

Huth, Paul K., 1999. ‘Enduring Rivalries andTerritorial Disputes 1950–1990’, in Paul F.Diehl, ed., A Road Map to War: TerritorialDimension of International Conflict. Nashville,TN: Vanderbilt University Press (37–72).

Jervis, Robert, 1997. ‘Book Review of The Clashof Civilizations and the Remaking of WorldOrder’, Political Science Quarterly 112(2):307–308.

Johnston, Alastair I., 1995. Cultural Realism:Strategic Culture and Grand Strategy in MingChina. Princeton, NJ: Princeton UniversityPress.

Jones, Daniel M.; Stuart A. Bremer & DavidSinger, 1996. ‘Militarized Interstate Disputes,1816–1992: Rationale, Coding Rules, andEmpirical Patterns’, Conflict Management andPeace Science 15(2): 163–213.

Katzenstein, Peter J., 1996. ‘Conclusion:National Security in a Changing World’, inPeter J. Katzenstein, ed., The Culture ofNational Security: Norms and Identity in WorldPolitics. New York: Columbia University Press(498–537).

King, Gary & Langche Zeng, 2001a. ‘LogisticRegression in Rare Events Data’, PoliticalAnalysis 9(2): 137–163.

King, Gary & Langche Zeng, 2001b. ‘ExplainingRare Events in International Relations’, Inter-national Organization 55(3): 693–715.

King, Gary & Langche Zeng, 2001c. ‘How Factualis Your Counterfactual?’ Working Paper, 2 August (http://gking.harvard.edu).



Levy, Jack S., 1988. ‘Domestic Politics and War’,Journal of Interdisciplinary History 18(4):653–673.

Lipset, Seymour Martin, 1994. ‘The Social Req-uisites of Democracy Revisited’, AmericanSociological Review 59(1): 1–22.

Long, J. Scott, 1997. Regression Models forCategorical and Limited Dependent Variables.Thousand Oaks, CA: Sage.

Maoz, Zeev, 2001. Dyadic Militarized InterstateDisputes Dataset. Version 1.1 (ftp://spirit.tau.ac.il/~zeevmaoz/dyadmid60.xls).

Maoz, Zeev & Bruce Russett, 1993. ‘Normativeand Structural Causes of Democratic Peace1946–1986’, American Political Science Review87(3): 624–638.

Mercer, Jonathan, 1995. ‘Anarchy and Identity’,International Organization 49(2): 229–252.

Mitchell, B. R., 1998a. International HistoricalStatistics: Africa, Asia & Oceania, 1750–1993.London: Macmillan.

Mitchell, B. R., 1998b. International HistoricalStatistics: The Americas, 1750–1993. London:Macmillan.

Mitchell, B. R., 1998c. International HistoricalStatistics: Europe, 1750–1993. London: Mac-millan.

Morse, Edward L., 1970. ‘The Transformation ofForeign Policies: Modernization, Interdepen-dence, and Externalization’, World Politics22(3): 371–392.

Nagler, Jonathan, 1994. ‘Scobit: An AlternativeEstimator to Logit and Probit’, AmericanJournal of Political Science 38(1): 230–255.

Oneal, John R. & Bruce M. Russett, 1997. ‘TheClassical Liberals Were Right: Democracy,Interdependence, and Conflict 1950–1985’,International Studies Quarterly 41(2):267–294.

Organski, A. F. K. & Jacek Kugler, 1980. The WarLedger. Chicago, IL: University of ChicagoPress.

Owen, John M., 1994. ‘How Liberalism ProducesDemocratic Peace’, International Security19(2): 87–125.

Pfetsch, Frank R. & Christoph Rohloff, 2000.‘KOSIMO: A Databank on Political Conflict’,Journal of Peace Research 37(3): 379–389.

Polity IV Project, 2000. Polity IV Dataset [com-puter file; version p4v2000]. College Park,MD: Center for International Developmentand Conflict Management, University ofMaryland.

Przeworski, Adam & Fernando Limongi, 1997.‘Modernization: Theory and Facts’, WorldPolitics 49(2): 155–183.

Ray, James Lee, 1995. Democracies and Inter-national Conflict: An Evaluation of the Demo-cratic Peace Proposition. Columbia, SC:University of South Carolina Press.

Russett, Bruce, 1993. Grasping the DemocraticPeace: Principles for a Post-Cold War World.Princeton, NJ: Princeton University Press.

Russett, Bruce; John R. Oneal & MichaeleneCox, 2000. ‘Clash of Civilizations, or Realismand Liberalism Déjà Vu? Some Evidence’,Journal of Peace Research 37(5): 583–608.

Sartori, Giovanni, 1995. ‘How Far Can FreeGovernment Travel?’, Journal of Democracy6(3): 101–111.

Thompson, William R. & Richard Tucker, 1997.‘A Tale of Two Democratic Peace Critiques’,Journal of Conflict Resolution 41(3): 428–454.

Tucker, Richard, 1999. BTSCS: A Binary Time-Series-Cross-Section Data Analysis Utility.Version 3.0.4 (http: //www.vanderbilt.edu/~rtucker/programs/btscs/).

Welch, David, 1997. ‘The “Clash of Civiliz-ations” Thesis as an Argument and as aPhenomenon’, Security Studies 6(4): 197–216.

Zakaria, Fareed, 1997. ‘The Rise of IlliberalDemocracies’, Foreign Affairs 76(6): 22–43.

GIACOMO CHIOZZA, b. 1968, PhDcandidate in Political Science, Duke Uni-versity. Specializing in international relationsand political methodology.



Documents

Is There a Clash of Civilizations? Evidence from Patterns ... · PDF fileSamuel Huntington’s The Clash of Civiliz- ... Is There a Clash of Civilizations? Evidence from Patterns of