What Is Civil War PDF

What Is Civil War?
Conceptual and Empirical Complexities of an Operational Definition Author(s): Nicholas Sambanis Reviewed work(s): Source: The Journal of Conflict Resolution, Vol. 48, No. 6 (Dec., 2004), pp. 814-858 Published by: Sage Publications, Inc. Stable URL: http://www.jstor.org/stable/4149797 . Accessed: 01/05/2012 14:25
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Sage Publications, Inc. is collaborating with JSTOR to digitize, preserve and extend access to The Journal of Conflict Resolution.
http://www.jstor.org
What Is Civil War?

CONCEPTUAL AND EMPIRICAL COMPLEXITIES OF AN OPERATIONAL DEFINITION
NICHOLASSAMBANIS
Departmentof Political Science YaleUniversity
on civil warhas seen tremendousgrowthbecauseof the compilationof quantitaThe empiricalliterature tive data sets, but there is no consensus on the measurementof civil war.This increases the risk of making inferencesfrom unstableempiricalresults.Withoutad hoc rules to code its startand end and differentiateit fromotherviolence, it is difficult, if not impossible, to define and measurecivil war.A wide rangeof variation in parameter estimates makes accuratepredictionsof war onset difficult, and differences in empirical results are greaterwith respect to war continuation. Keywords: civil war; Correlatesof War;data sets; coding rules
Advances in the empiricalliteratureon civil war depend critically on the collection and refinementof dataon civil waroccurrence.Most civil war lists rely heavily on the Correlatesof War(COW) project.'Since the first COW list was published, there has been little peer review of COW coding rules. Within the walls of the COW project, there has been substantialdebate on how to improve data quality.That debate, however, has not benefited from open scholarly discourse.2Given the project's preeminence, many large-N studies use COW data unquestioninglyor limit themselves to abouta dozen researchprojects Currently, makingonly minorchanges to COWdata.3
1. Singer and Small (1972, 1994), Small and Singer (1982), Sarkeesand Singer (2001), and Sarkees (2000). 2. Personalcommunication(July 31, 2001) with StuartBremer,foundingmemberof the Correlates of War(COW). COW recently initiatedan online forum for public debate,but the forum is rarelyused. 3. Walter(2002) andCollier and Hoeffler (2001) use COW datawithoutmakingany modifications. Mason and Fett (1996) make only minor changes. AUTHOR'S NOTE: For comments, I thank Keith Darden, Michael Doyle, James Fearon,Nils Petter Gleditsch,HAvard Hegre, StathisKalyvas,David Laitin,John Mueller,Monica Toft, Ben Vallentino,Elisain ComparativeEthnic Processes (DartmouthMeeting, October beth Wood,participantsin the Laboratory in the Universityof BritishColumbia'sworkshopon "Econometric 2001), andparticipants Analyses of Civil War." I also thankAnnalisaZinn, Douglas Woodwell, KatherineGlassmyer,Ana MariaArjona,and Anna Rose ArmentiaBordonfor researchassistance.All files necessaryto replicatethe analysis in this articlecan be foundathttp://pantheon.yale.edu/-ns237/index/research.html#Data andwww.yale.edu/unsy/jcr/jcrdata.htm.
JOURNALOF CONFLICTRESOLUTION,Vol.48 No. 6, December 2004 814-858 DOI: 10.1177/0022002704269355
O 2004SagePublications 814
IS CIVILWAR? Sambanis/ WHAT
815
have producedcivil warlists based on apparently divergentdefinitionsof civil war,but here one think. Most than thereis less pluralism might projectsdo not conductoriginal on COW. historical research and depend heavily The result may be replication of due to the COW rules and errors uncertaintyabout whetherdifferent original coding different results. definitions generate I take a close look at the operationaldefinitionof civil war and review severalcodand analyze theirimplicationsfor our understanding of ing rules used in the literature civil war.To make the discussion concrete, I referto specific cases. This exercise generatesinsights into our abilityto measurecivil warand distinguishit from other forms of political violence. Drawingon those insights, I propose a new coding rule and offer a new civil war list. I then measurethe impact of differences across coding rules by regressingthe same civil war model on 12 differentversions of war onset and prevalence. Althoughit may be impossible (and some would argueundesirable)to arriveat a consensus on a single definition and measurementof civil war, it is importantto of civil waris affectedby differentcoding rules. My analyknow if our understanding sis shows that some of our substantiveconclusions are sensitive to differences in coding rules, whereas others are remarkablyrobust. Significant differences across civil war lists are mainly due to disagreement on three questions:What thresholdof violence distinguishes civil war from other forms of internalarmedconflict? How do we know when a civil war startsand ends? How can we distinguish between intrastate,interstate,and extrastatewars? Answers to these questionsare not only relevantfor the purposesof accuratecoding, but they also of the concept of civil reveal the degree to which we share a common understanding war. I discuss these issues as a way to explore what differences across coding rules can tell us aboutthe meaningof civil war.I do not offer new theory thathelps distinguish civil war from other forms of violence.4 Rather,I want to see if availableoperational definitions(coding rules) allow us to measureand analyze civil war as a distinctcategory with a "naturalhistory"(cf. McAdam, Tarrow,and Tilly 2001), an "ontology" that is differentfrom thatof other forms of political violence (cf. Tilly 1978). Thus, I implicitly accepthere the premisethatcivil wars aredifferentfrom otherviolence and consider if the coding rules we have at our disposal are sufficient to make a clear empiricaldistinctionbetween civil war and other violence. I find thatit is not possible to arriveat an operationaldefinitionof civil war without adopting some ad hoc way of distinguishing it from other forms of armed conflict. Although a core set of "ideal"cases of civil war may exist, too many cases are sufficiently ambiguous to make coding the start and end of the war problematic and to question the strict categorizationof an event of political violence as a civil war as opposed to an act of terrorism,a coup, genocide, organized crime, or international war.5In the end, it may be difficultto study civil war withoutconsideringhow groups
4. I offer such a theoreticaldiscussion in a book-lengthmanuscript, currentlyin progress.For a discussion of why we should distinguishcivil war from other political violence, see Sambanis(2004). 5. Proceedingtheoretically,ratherthanempirically,Tilly (2003, 14) makes a similarargument,statdestruction"-a typology ing thatcivil wardoes not have a distinctcausal logic. It is a formof "coordinated thatincludesvariousotherforms of political violence thatgeneratesalient "short-run damage"and are perpetratedby highly coordinatedactors (the two dimensions that describe Tilly's typology of violence).
816
RESOLUTION JOURNALOF CONFLICT
in conflict shift fromone form of violence to another,or it may be profitableto analyze political violence in the aggregate,ratherthancut across thatcomplex social phenomdefinitions.This articleis a first step in exploring thatidea by proenon with arbitrary an viding analytical review of existing coding rules that highlights the difficulty of accuratelydefining and measuringcivil war.It also improvescurrentlyavailablecoding rules at the margin,so as to producea civil war list thatis more consistent with the core elements of most working definitions of civil war. HOW WOULD WE KNOW A CIVIL WAR IF WE SAW ONE? In theirseminal studyResort to Arms, Small and Singer (1982, 210) defined a civil war as "anyarmedconflict that involves (a) militaryaction internalto the metropole, of the nationalgovernment,and (c) effective resistanceby (b) the active participation both sides."The main distinctionthey drew between civil (internalor intrastate)war or extrastate(colonial and imperial)warwas the internalityof the warto and interstate of the governmentas a combatthe territory of a sovereign state and the participation ant.Civil warwas furtherdistinguishedfrom otherforms of internalarmedconflict by the requirementthat state violence should be sustainedand reciprocatedand that the war exceeds a certainthresholdof deaths (typically more than 1,000). It is, I will argue, difficult, if not This definition is deceptively straightforward. an definition of civil war without adopting some to impossible, develop operational ad hoc coding rules to distinguishcivil warsfromotherforms of political violence and accuratelycode war onset and termination.First, it is often difficult to distinguish wars:for example, the Russian civil war in Chechnyain the extrastatefrom intrastate 1990s might be consideredas a war of decolonization, similar to Cameroon'swar of independencein 1954. Second, it is unclearwhatdegree of organizationis requiredof the partiesto distinviolence. In some cases, a functional guish a civil warfromone-sided, state-sponsored to but we still code a civil war(e.g., Somalia after 1991). has ceased exist, government In othercases, the governmentmay be fighting a war by proxy using militias in seemingly disorganized intercommunalclashes (as in Kenya's Rift Valley from 1991 to 1993), but most would not classify such a case as a civil war.Elsewhere, rebel organizations are indistinguishablefrom criminalnetworksor ragtagmilitias. Third,if we focus on a numericalthresholdof deaths to identify wars, how do we deal with the problemof unreliablereportingand incompletedata?Even with reliable be coded only on the basis of significantlyreducedhostilities, data,shouldtermination or should we also focus on discrete, easily coded events, such as peace treaties? Fourth,given that violence during civil war is typically intermittent,how do we determinewhen an old warstops anda new one starts?And how can we distinguishthe end of a civil warfrom the beginningof a periodof politicide, terrorism,or otherform of violence? The COW projecthas provideda valuablepublic good to the profession by coding cases of war, but it has not offered much guidance on how to answer these difficult
Sambanis/ WHAT IS CIVILWAR?
817
questions. It may also have caused some confusion. One source of conceptual confusion is the lack of clarityon the thresholdof deaths used to distinguishcivil war from otherviolence. Small and Singer (1982, 213) used an annualdeaththresholdof 1,000 deaths in their early coding efforts, although they later seem to have abandonedthat criterion.Many authorsstill operateunderthe assumptionthatthe COW projectuses This uncertaintystems from the fact that COW codan annualbattle-deathcriterion.6 have at least times. The firstmajorchange was the adoptionof rules three changed ing an annualdeathcriterionfor civil warsin Resortto Arms,whereasno such requirement existed for internationalwars in The Wages of War 1816-1965 (Singer and Small 1972). Then the annualdeath criterionfor civil wars was relaxed (Singer and Small 1994), but it was still requiredfor extrasystemicwars, althoughnow deaths incurred by nonsystem members were counted (Singer and Small 1994, 5).7 In a third set of changes, Sarkees (2000, 129) and Sarkees and Singer (2001) classified some extrasystemic wars as civil wars while abandoning the annual death criterion for It is unclearwhy the definitionof extrasystemicwar kept changing or extrastatewars." if the new rules were used to recode all armed conflicts in empires and colonies if dataon annualdeathshad the periodcoveredby COW.9 More important, throughout been collected to code civil warsaccordingto the old (annual)criterionup to the 1994 not have made use of those data,insteadof losrevision, why would COWresearchers of that information much a by using binaryvariabledenoting the onset and termiing nationof civil war?To fully evaluatethe researchthat went into compiling the COW lists, one needs access to COW's raw data and coding rules.'0 COW'sefforts to constantlyrefine its dataand improveits coding rules are admirable. But this process raises the following question: were the new rules consistently reappliedto historicaldata?When the annualdeathcriterionwas abandoned,how did COW coders determinethe beginning and end of a civil war? Did they code the first year with 1,000 deaths as the onset of the war?" Or did they code the startat the year that the cumulative death toll surpassedthe 1,000 mark?An example that demon6. See Walter(2002), Gleditsch et al. (2002, 617), and Licklider (1995, 682). 7. Accordingto Singer and Small (see codebook, 1994), interstatesystem membershipis defined in minimalcriteria... at least 500,000 totalpopulationandeitherdiplomaticrecognitionby at termsof "certain least two majorpowersor membershipin the Leagueof Nationsor United Nations."Extrasystemicwars are fought between "a nationthatqualifies as an interstatesystem member... [and]a political entity that is not an interstatesystem member." They are subdividedinto imperialand colonial wars. I referto extrasystemic as they both refer to colonial and imperialwars. and extrastatewars interchangeably, 8. As I was working on final revisions of this study, I received an e-mail message from Professor Sarkees (March 15, 2004), stating that the 1,000 battle-deaththresholdhas never been abandoned in the COWprojectandthatpooreditingof the Sarkees(2000) articlegives readersthis mistakenimpression.Severalmembersof the COWteamhadreadearlyversionsof this study,andthey hadnevercorrectedme. Moreover, I have referredto codebooks and publicationsfrom COWthat counter Sarkees's recent claim. on deathsof colonial subjectsmighthave been systematicallyworse 9. A concernis thatinformation in empires than informationon deaths of insurgentsin nation-states,particularlyin the post-1945 period. 10. The Wagesof War,1816-1965 and Resort to Arms include appendiceson included and excluded wars, but little, if any, explanationis given for exclusions. Most cases were droppedbecause they did not meet the deaththreshold(Small and Singer 1982, 330). 11. This may be the implicit COW coding rule, accordingto a communicationwith StuartBremer (July 31,2001). However,Gleditschet al. (2001, 16) note that"anyconflict coded by COW as having more than 1,000 battle-deathsoverall is recordedas a civil war for all years (even years of inactivity and years before the cumulativedeath toll reached 1,000)"
818
JOURNALOF CONFLICT RESOLUTION
stratespotentialerrorsin the applicationof the new rulesis the Algeriancivil warin the 1990s. This war,which some datasets code as startingin 1992, is omittedfrom Singer and Small (1994), even thoughfive otherwarsthatstartedin 1992 were included (thus revealing that their coding extended through 1992). Perhapsthe war had not caused more than 1,000 deaths in 1992 (it actuallyhad), but the revised COWdata set, which goes up to 1997 (Sarkeesand Singer2001), includes the Algerianwarwith a 1992 start date.Because the coding rules were the same in the two COWdatasets, unless this was a coding errorin the 1994 version, the coded startof the war in the new version suggests thatthe warreachedthe 1,000 deathmarkonly after 1992, andthe startof the war was then back-coded to the startof the violence in 1992. The cumulativedeathcriterionaddedfew warsto the COWlist. Forthe periodfrom 1816 to 1979, 13 wars were addedto the 1994 list (107 wars were includedin the 1982 list). Only 5 of these 13 wars had been included in Appendix B of Resort to Arms (Small and Singer 1982), and 3 of these 5 were excluded because of lack of system or because the violence was characterized as a massamembershipof the participants cre. Thus, shifting to the cumulativedeath criterionadded only 2 wars to the list.'2 The cumulative death criterion introduces some problems. First, it is harderto know when to code the startof the war. If we code it the first year the killing begins, then we will not be able to study violence escalation because the outbreakof minor One researchgroupthattries to violence will be subsumedin the periodof "civil war." avoid this problemis Gleditschet al. (2001, 2002), who code a "war"when they count morethan 1,000 deathsin a single yearanddo not code the beginningof the warduring the firstyearwith morethan25 deaths.Buttheirdataset does not havea rulefor coding war termination,so we do not know precisely how to code their wars in a way that is compatiblewith otherdatasets thatuse the cumulativedeathcriterion.The problemis armed conflict category, in which there are created by the so-called "intermediate" more than 100 but fewer than 1,000 deaths in a given year. If we had a conflict with, say, 600 people killed in the first year, then 3 years with virtuallyno deaths, and then anotheryearwith 600 deaths,this mightqualify as a civil war accordingto the cumulaviolence tive deathcriterionbut would be coded as two distinctevents of intermediate in the Gleditschet al. (2001, 2002) data set.'"The annualdeath thresholdsolves this
12. See, again, Sarkees's comments noted earlier.She argues that COW never abandonedthe 1,000 annualdeaththreshold.But this raises new questions.Forexample, the Uppsaladataset andthe relateddata set by Gleditschet al. (2002), which use an annualdeaththresholdof 1,000 to code civil war,shouldhavefew differencesfromCOW lists if COW used an annualthreshold.But, as I show later,the differencesbetween these datasets are large.Takethe example of Cambodia:Singer and Small (1994) code two civil wars, one from 1970 to 1975 and anotherfrom 1979 to 1991. Gleditsch et al. (2002) and Strand,Wilhelmsen, and Gleditsch(2003) code a firstwarin 1967, a second warfrom 1970 to 1975, a thirdwarin 1978, anda fourthin 1989 (these all have different"conflict sub-IDs"in the Strand,Wilhelmsen, and Gleditsch data set; hence, they are consideredas different"cases"or war starts). 13. The dataset by Gleditschet al. (2001, 2002) has undergonemany revisions (one currentand four old versionscan be foundonline at http://www.prio.no/cwp/armedconflict [accessed June22, 2004]). Later versions haveaddressed(althoughnot entirely)the problemof coding waronset/termination by assigning a "conflict ID"and "sub-ID"to each conflict and consideringconflicts to be differentif or when they switch from an intrastate to an interstatewar (and vice versa), when the parties and issues are different,or when therearemorethan 10 yearswith fewer than25 deathsperyear.However,it is still notclearto me if a "same" and back to "war"should be considereda single "war"for the conflict thatswitches from "war"to "minor" purposesof comparisonwith war lists that use a cumulativedeath criterion.
819
problembecause an end to the war would be coded whenever violence drops below 1,000 deaths. But this creates the opposite problem of coding too many war startsin what is essentially the same conflict, if levels of violence fluctuatewidely. One way aroundthese problemsis to stop tryingto code and analyze civil war as a distinct phenomenon and, instead, to code levels of violence along a continuum (by countryand year). Warsmight then be identifiedas periods with many violent spikes, in additionto othercharacteristics of the violence, dependingon the authors'theoretical argument(e.g., the degree of organizationof the parties,the presenceof two-way violence, etc.). The analysis would then try to explain political violence first and, by refining the theory, explain why violence takes differentforms. A secondproblemwith the cumulative-death criterionis thatit could simply lead us to code as civil wars manysmall conflicts thatslowly accumulatedeaths.In effect, this criterioncreates inherentlyright-censoreddata, making it hardto code war termination. By consistently applying the cumulative death criterion, Sarkees and Singer (2001) could code as a civil warany minorconflict or terrorist campaignthatcauses 25 deathsperyear for more than40 years. Fearonand Laitin(2003, 76) attemptto correct this with a rule that 100 deaths must occur every year on average in an ongoing war. But, in combination with the 1,000 aggregate-deathsthreshold,this creates another logical problem.They would not code as a civil war a conflict thatcaused 900 deaths over 9 years,butthey would code a conflict thatcaused 1,000 deathsover 10 years.14 A useful example to consider is Northeast India (Nagaland), where Fearon and Laitin(2003) code an ongoing civil warsince 1952. I was not able to find evidence that many(say, more than 100) deathsper year occurredin armedconflict therefrom 1952 to 1961. According to Gleditsch et al. (2002, appendix) and Small and Singer (1982, 339), there was no war or intermediateviolence during any year of the conflict in Nagaland.This case illustratesnot only the problemof how to code war termination with the cumulativethresholdbut also the relateddifficulty of how to handle several chronologically overlappinginsurgenciesin the same country.Combiningregionally concentratedinsurgencies in India's Northeaststates may be reasonableand would thresholdin the periodconsideredby Fearonand probablysatisfy the aggregate-death Laitin.But a strictapplicationof the cumulative-deathrule in such cases is problematic, given that in other countries,chronologically and even geographicallyoverlapping insurgenciesare often treatedas separateconflicts. How to distinguishbetween these rebellions is not always easy. Burma, Chad, India, Ethiopia, and Zaire (in the 1960s) are all countries that pose difficulties in distinguishingamong various rebelof how to hanlious groupsandperiodsof violence.15 In the absenceof a clear standard dle suchcomplicatedcases, a ruleof thumbshouldbe to code a "civilwar"in countries with manyoverlappinginsurgencieswhen the violence escalates markedlyand not at the startof low-level hostilities. In the case of India,this meansthatif we were to com14. Moreover,that 100 is the average numberof deathsperyear means thata conflict with thousands of deathsin the firstyearcan be consideredongoing, even if annualdeathsafterthe firstyearfall to nearzero. 15. For example, Gleditsch et al. (2002) and Strand,Wilhelmsen, and Gleditsch (2003) distinguish between the Serb andCroatrebellionsin Bosnia, butmost otherscombine the two insurgenciesinto a single "Bosnianwar"event. At least five separaterebellionswere ongoing from 1960 to 1965 in Zaire(now Democratic Republicof the Congo), and all data sets typically combine these events into a single civil war.
820
bine the rebellionsin the Northeaststates, a civil war should be coded as startingin the 1980s, when violence escalated in Assam, Tripuras,and Manipur. thresholdbecomes more defensible if Despite these problems, a cumulative-death it is combinedwith a criterionthatmilitaryactivitybe sustainedfor the durationof the conflict. Both Sarkeesand Singer (2001) and FearonandLaitin(2003) apply this rule, but it is unclearhow sustainedmilitaryactivity is best defined. To focus on an average numberof deathsper year reintroducesthe problemsof counting annualdeaths.If we can accuratelycount 100 deaths per year, we might be able to accuratelymeasure higher numbersof deathsandbe betteroff using these numbersas our dependentvariable. We mightinsteadcount battles,requiringat least one battleper year between the would createa definitionalproblemmore same parties.But perhapscounting"battles" severe than counting deaths, and again we would need an arbitrary"number-ofbattles"threshold,which would exclude cases of low-level insurgencyin warsof attrition wherebattlesarenot easily distinguishablefromterrorist activity(as in the case of Peru's civil war). To reacha balancebetween the pros and cons of the absolute versusannualthresholds, we must consider a few issues relatedto how we measurewar magnitude: as civil war'? 1. Whatlevelof violencequalifies or relative level? thisbe an absolute 2. Should we onlycountbattle deaths or alsociviliandeaths? 3. Should
ARRIVING AT AN ABSOLUTE THRESHOLD OF DEATHS TO CODE A CIVIL WAR
A characteristic of civil warthatdistinguishesit fromotherformsof violence is that it causes large-scaledestruction.A high thresholdof deaths can set wars apartfrom andsome coups (althoughnot necessarilyfrompogromsor genocide). riots,terrorism, aboutthe 1,000-deathsthresholdused in the litBut thereis nothinginherently"right" erature,and strictapplicationof thatthresholdcan cause us to drop severalcases that satisfy all other characteristicsof civil war. A range of 500 to 1,000 deaths could, in of civil war as an event principle,be equally consistent with a common understanding thana single cutoff point may work thatcauses majordestruction.Using a rangerather better,given the highly skewed distributionof civil war deaths. In 145 civil wars that startedbetween 1945 and 1999, the meannumberof deathsis 143,883 (with a standard deviationof 374,065), andthe medianis 19,000. Despite the high average,11 conflicts have caused fewer than 1,500 deaths,and some barelyreached 1,000 deaths.But these cases-for example, the Taiwaneserevolt in 1947, the Dar ul-Islam rebellion in 1953 in Indonesia,or the fighting in Croatiaafterits independencein 1992-satisfied most or all of the other criteriafor civil wars: they were fought by well-organized groups with political agendas,challengingthe sovereign authority,and violence was reciprocal. Given the poor quality of our data (recall how hardit was to count the numberof dead afterthe Twin Towersattackon September 11, 2001, in New Yorkand consider how muchharderit would be to measurecivil wardeathsin Angola in 1990 or Tibet in 1951) and the skewed distributionof deaths, we should use a more flexible coding
821
We could provirule, such as a rangeof deaths,insteadof the 1,000-deaththreshold.'6 the we to code a war onset at that count 100 500 deaths and sionally year keep the event as a civil war if coders count more than 1,000 deaths in total within 3 years of onset. This incorporates the sustainedfightingrulebecause events with low deathsin the first be coded as wars only if armedconflict is sustainedand causes 1,000 or would year more deathswithin a 3-yearperiod.The value of this coding rule is thatit could easily be appliedto back-codeevents by integratingexisting databasesthathave information on lower level armedconflicts.
FOCUSING ON THE RELATIVE MAGNITUDE OF VIOLENCE
Anotherproblemwith the absolutethresholdis thatit does not reflect the conflict's magnitudeas well as a relative (per capita) measure would. If we added a per capita deathsmeasureto our coding rules, we would be less likely to miss armedconflicts in small nations that producedfew deaths but were nonetheless dramaticallyimportant for the history of those countries.A 1-yearconflict that killed 500 people (the lowest numberin the rangesuggestedaboveto code waronset) in a countrywith half a million inhabitants (the smallestpopulationsize allowed by most coding rules)would amount to deaths at the level of 0.001 of the population. A conflict of equal magnitude in a countryof 20 million people would have caused 20,000 deathsand would have been coded as a civil war in all data sets. We could use the 0.001 thresholdas a benchmark, and wars that do not exceed the 500 to 1,000 deaths markcould still be classified as civil wars if they meet the per capita deaths criterion. An example thatillustratesthe need for greaterflexibility in applyingdeaththresholds is the DhofarRebellion in Oman(1965-1976). This was an insurgencywaged by an ethnically organizedrebel group (the Dhofar LiberationFront [DLFI) against the Sultanof Oman.The rebelsrecruitedmainlyfrom the Qara,a smallethnic groupliving in the mountainsof the Dhofarprovince,andthe grouppresenteditself as CommunistMaoist.Youngpotentialrecruitswere sent to school to receive political indoctrination (Price 1975, 7; Connor 1998, 156-57). There was an element of Islamic fundamentalism in the group's ideology until mass defections of Islamist soldiers after the 1970 Bait Ma'ahshinimassacre(Connor 1998, 159). Some scatteredterrorism took place in the populousnorth,butfightingwas largelyconfined to mountainous areas.The rebels were supportedby the People's Republic of Yemen in their campaignto control the mountains(Price 1975, 3). The patternof armedconflict and the organizationof the rebellionareconsistentwith a common understanding of civil war,butthis case is typiexcluded from civil war lists because of a low death count, even though, in per cally this conflict caused more than capita terms, damage many others that are typically includedas civil wars. The 1,000-deathscriterionmay lead us to include more cases of civil wars in large countries, if more populous countries can more easily produce several insurgencies thatcan cause high levels of deaths. Ethnicrioting in Nigeria or state repressionand
16. In my dataset, I includecases thatmay falljust shortof the 1,000 markand identifythem as potentially ambiguouscases.
822
populardissent in China is much more likely to generate large numbersof casualties thana coup in Fiji or Cyprus.This may be one reason thata variablecontrollingfor a country'spopulationis among the most significantandrobustexplanatoryvariablesin models of civil war onset. Consider the omission from most data sets of the GrecoTurkishwarin Cyprusthatstartedin 1963. This warwas excludedby SingerandSmall (1972, 398; Small and Singer 1982, 340) due to insufficientdeathcount. The conflict caused around1,000 deathsandcertainlymet all othercriteriafor civil war:therewere battlesbetween organizedmilitaryunits, involving the governmentagainstrebel miliand a clearly articulated tias, andpartieswith local recruitment political agenda.Even if only 500 deathshad occurredin Cyprusin thatperiod, this would have amountedto 0.001 of the population.A war with the same intensity in a countrywith 100 million inhabitantswould have caused 100,000 deaths-a massive tragedy that would have been coded in all data sets. A percapitameasurewould capturethese cases, but we would also need to preserve a relativelyhigh absolute thresholdnot to run the opposite risk of selecting too many small conflicts in small countries(or dropping"relatively" small wars in largercounmeasure a is difficult and labor Whatwe mustnot intensive. tries). Creating per capita do is computepercapitadeathsin conflicts thathave alreadybeen selected by COWor other projects on the basis of an absolute-deaththreshold. Rather,coding must be based on primaryresearchto measuredeathson a per capitabasis in all countrieswith any politicalviolence. A startingpoint mightbe the Gleditschet al. (2001) list of minor armedconflicts. So far,no deathfigures are availablefor those conflicts, and we know only that they have caused more than 25 deaths per year and less than 1,000 for the durationof the conflict. Gettingbetterdatafor those conflicts might facilitatethe constructionof a per capita measure.An alternative(and less costly) approachis to code per capita deaths for a shorterperiod (e.g., from 1980 onwards),allowing us to constructa data sample based on the combinationof absolute and per capita death measures (post-1980) and another sample using only the absolute threshold.We could then estimate a model on the two samples of the post-1980 period and compare the results.We could thendo a sortof out-of-sampletest by estimatingthe same model on each sampleperiodto see if the model's predictionsareconsistentacrossperiodsusing eithersample. If the resultsare not influencedby our coding rules,then going through the troubleof coding a per capita measuremight not be necessary.
THE EFFECTIVE RESISTANCE CRITERION AND MEASURING BATTLE DEATHS VERSUS TOTAL DEATHS
The thirdcrucial question is whetherwe should code battle deaths or also include civilian deathsdue to the war.The battle-deaths criterionis the legacy of theCOWproject and usually refers to militarydeaths (the measure was initially used to measure deathsin interstatewars). Small and Singer(1982, 213) claim thatthey counted "civilian as well as militarydeaths"in civil wars,butit is not clear if they have, in fact, coded civilian deaths due to rebel attacks.Civilian deaths must be counted in deathtotals.'7
17. This position is gaining ground: see Sarkees and Singer (2001, 12) and Valentino, Huth, and Balch-Lindsay(2001).
Sambanis IS CIVIL / WHAT WAR? 823
Civiliansare targetedin civil war and are disproportionately affected by humanitarian disasterscreatedby combatantsto hold civilian populationshostage and gain control of territory."8 One might also considercounting refugees and internallydisplaced persons as a measure of the humancost of the war (cf. Doyle and Sambanis 2000). To get an accuratemeasureof civilian deaths,we must also determineif deaths due to politicide that occur in and aroundcivil war should be counted. Small and Singer (1982, 215) counted deathsdue to "actsof massacrecommitted ... duringthe course of a civil war."That seems reasonable, but a harderquestion is, how do we handle deathsdue to acts of massacrethatflank periodsof civil war?Should we code a war as having startedif civilian massacresoccur for a period (say, 1 year) and military confrontationensues with deaths that exceed 1,000 in the next year? What about the reverse situation, where a civil war gives way to civilian massacres?Distinguishing periodsof civil war fromperiodsof massacresclearly is veryhard,as the case of Cambodia illustrates.Most data sets code civil war onsets in 1970 and 1978 in Cambodia butno warfrom 1976 to 1977, which is a periodthatcorrespondsto civilian massacres (the killing fields). Small-scale battles between the Khmer and Vietnamese-backed troops did take place near the Thai borderin that period, but it is unclear how many people died as a resultof armedconflict. Hence, most datasets code this as a period of politicide thatis distinct from the civil war.Yet, if we could have establishedthat 100 or so people had been killed on the side of the strongerpartybetween 1976 and 1978, we would have coded an ongoing war in thatperiod (especially in data sets using the cumulative-deathcriterion and where effective resistance is not measured as a percentageof the total numberof deaths). In light of these difficulties,a conservativestrategyis to count deathsdue to massacres thatoccur rightbeforeor afterthe waras war-related deaths.The maindifficulty is to find evidence of effective resistance. According to Small and Singer (1982, 21415), effective resistanceimplies thatthe strongerside should suffer at least 5% of the casualties of the weaker side. Fearon and Laitin (2003) relaxed this coding rule and measuredeffective resistance by 100 state deaths. They do not, however, specify if these deathsmustoccur in the firstyear of the war or cumulativelythroughoutthe war. Without data on the distributionof state deaths throughout the war, we are left rudderless in tryingto distinguishperiodsof "civil war"fromperiodsof "politicide"or (Could we say with certaintythat50 governmenttroopswere not killed in "genocide." the jungles of Cambodiain 1976 and another50 in 1977?) Some find the distinction easier to make if massacresoccur before a "civil war"starts,as in the case of the 1966 massacresin Nigeria, which are typicallythoughtof as separatefrom the Biafrancivil It mightalso be easier to distinguishmassacresfrom civil war,which startedin 1967.19 warif the massacresoccur aftera formalend to the waris reached(i.e., a peace treaty). But if the war peters out and mostly one-sided violence starts(as in the 1984-1987
18. It might even be desirableto count civilian deaths that are indirectlythe resultof the war as, for example, in the case of war-relatedstarvationdeaths in Ethiopiaand Somalia. However,this is a slippery Huth,andRussett slope, as it is unclearhow farone could tracethe healtheffects of civil war(see Ghobarah,
2003forsuchan effort).
19. In May-July 1966, massacresof 30,000 Ibos in the Northeasttook place, includingmassacresdur-
leaders' of Ironsi andIboleaders. Biafra de factoindependence achieved ingtheNorthern coupandmurder in 1967. by theendof 1966,andwarstarted
824
massacres in Ndebele, Zimbabwe), the distinction is harderto make, particularly because war may restartwithin 1 or more years as a result of such massacres. Thus, the effective-resistancecriterioncannot help us establish with certaintythe difference between civil war and politicide, and we would need to use a sustained effective-resistance criterionto make a clear distinction. Just as Fearon and Laitin (2003) require 100 deathsper year to code an ongoing civil war, we could require 10 statedeathsper year to distinguisha civil warfrom a politicide.20 But this type of arithmetic seems inappropriate, not least because it trivializes and imposes a petty discion the thatoccurs duringcivil war.Whatif insurgentssufferviolence complex pline deaths kill 60 soldiersor police officers in one year,none in can ing, say, 1,000 peryear the next year, and then 5 per year for the next 8 years (for a total of 100 in 7 years)? Wouldthis be any moreor any less of a civil warthana conflict in which 20 soldiers are killed per year for 5 years? The problem of counting effective resistance each year might be avoided if we require 100 deaths per year on average for the durationof the conflict, but this would obviously requirea higher thresholdof the state deaths criterion andwould lead us to reclassify some civil warsas politicides or genocides.2'What if we forget about the question of distributionof state deaths over time? Should an absolute criterionof 100 state deaths be the single most importantfactor in deciding whetherto classify an insurgencyas a civil warif the insurgencymeets all othercriteria for a civil war? An example is the Gamaat Islamiya (Islamic Group) and Islamic Jihad's insurgency in Egypt from 1992 to 1997. According to one source, around 1,200 people were killed over 5 or 6 years, which meets the cumulative death criterion.22 The rebels were organizedandhada political ideology. Guerillainsurgencywas sustained,andattackswere directedagainstpolice, securityforces, governmentministers, tourists,and the Coptic minority.Whetherwe code this case as terrorismor civil war currentlyrests on whether the state suffered 100 or more deaths. This seems a weak criterionon which to base the classification of any case.23 Another difficult case is Argentina's"dirtywar,"in which Harff (2003) codes a politicide from 1976 to 1980, Fearonand Laitin(2003) code a civil war from 1973 to 1977, Singerand Small (1994) do not code a civil war,andGleditschet al. (2002) code a war in 1975 and "possibly"in 1976 and 1977. Official statisticsfrom the Argentine attacks"from militarycite 492 deaths (including civilian officials) due to "terrorist 1969 to 1980.24 Guerillassuffereddeaths startingin 1969 until 1972 (more than 100 peryear),and it is hardto know how manypeople were killed between 1973 and 1977, the periodsometimescoded as a "civil war." Totaldeaths,includingguerillasandcivil20. The 1,000 cumulative-death criteriondividedby the 100 deaths-per-year rulegives a 10-year window for minor conflicts to be labeled civil wars. Dividing the Fearon and Laitin (2003) 100 state-deaths effective-resistancerule by the same 10-yearwindow yields a 10 state-deaths-per-year rule. 21. Also, averagingstatedeathsmightnot resolve the problem,if all or most deathsoccurredearly in the conflict. 22. See my online supplementfor informationon this and all cases discussed in the text. 23. The Kenyan"shifta"war in the 1960s is anothersuch case thatno dataset includes as a civil war, most likely because of the difficulty of finding informationon the effective-resistancecriterion. I found sources listing dozens of killings of police and militarybutno clear evidence of morethan 100, althoughthe number and apparentintensity of battles described in several sources suggest that there was effective resistance. 24. See a more detailed discussion and a list of sources in the online supplement.
825
ians, in that period seem not to have reached 1,000, particularlyif we do not count deaths afterthe coup of 1976, which broughta new regime to power, and so an end to the "war" could be coded in 1976. The new regime engaged in a massive purgeof suspected guerillas and their supporters; startingin 1976, the new regime killed an estimated 9,000 to 30,000 people in the period from 1976 to 1983. During that period, deathsincurredby the state were few, so we could not code an ongoing civil war after the 1976 coup. In the absence of clear rules on how sustainedviolence and effective resistancemustbe coded over time (i.e., how many statedeathsmustoccurper year), it is unclear if and when we can code this conflict as a civil war or rathera low-level insurgency,combined with a coup, and followed by a politicide. This brief discussion should reveal thatmost of the coding rules currentlyused to measurecivil war are somewhat ad hoc and that the problem is often exacerbatedby the low quality of data on deaths due to armedconflict. There is large room for measurementerrorhere, and such errorwill makeit harderto establishempiricallythe differences between civil war andotherforms of political violence. Thus, in the presence of these problems, one might argue that coding rules should never be applied too strictlyand that,when we arefaced with an ambiguouscase, we shoulderron the side of caution,including such cases while makingit possible to identify them at the analysis stage. In the civil warlist thatI have compiled by applyinga new set of coding rules that try to address the issues raised here, I have included ambiguous cases but have flagged them both in the data set and in a supplementarydocumentthat explains the coding for each case. This allows analysts to make their own decisions about which conflicts to drop or explore furtherto confirm that they are accuratelycoded. CLASSIFYING AND ANALYZING EXTRASYSTEMIC WARS Anotherissue thatsometimesaccountsfor coding differencesacrossdatasets is the difficulty of how to code extrastate(colonial, imperial) wars. I argue that extrastate wars may reasonablybe consideredto be differentfrom othercivil wars and excluded from civil war lists. The concept of a territorial state is centralto the definition of civil war but creates some problemsin the classificationof the so-called extrastate(or extrasystemic)wars. The extrasystemic-wars categoryin COWincludedcolonial and imperialwars. Imperial wars were defined as wars between "system member[s] versus independentnonmember[s]of the interstatesystem"(Singer and Small 1972, 382). Colonial warswere defined as wars between a "system member versus an ethnically different, nonmemberof the interstatesystem" (Singer and Small 1972, 382). nonindependent, wars were classified as a distinct category due to a conceptualdistincExtrasystemic tion between warsthatare"peripheral to the centerof government(or the metropole)" andthereforequalitativelydifferentfrom wars thattake place within the core territory of the metropole(Sarkees2000, 126). This was a normativedefinitionbecause it interpretedthe natureof the relationshipbetweengovernmentsanddependentor independThis normativeelement stands in contrastto ent regions within the state's territory. mechanisticdefinitionsof civil war(such as the deaththreshold).Moreover,a consid-
826
erationof power relationshipswas absentfrom the definitionof interstatewars, which were defined entirely based on a territorialconception of the state. Over time, "the entity necessitateda re-thinkingand ultiemphasison the statememberas a territorial of the metropole[andperiphery]distinction"(Sarkees2000, 127), mate abandonment leading the COW project to redefine its coding rules and reclassify many extrasystemicwars as civil wars. Identifyingextrasystemicwarsis morecomplicatedas a resultof the way the COW coding rulesevolved. Initially,coding rulesfor extrasystemicwarswere differentfrom rules for intrastatewar:"whereasan interstatewar which qualifiedon the basis of battle deathswould be includedif the total fatalityfigurefor the protagonistson each side reachedthe 1000 mark,the memberitself(including system memberallies, if any) had to sustain 1000 battle fatalities in order for the extrasystemic war to be included" (Singer andSmall 1972, 36), and "if the war lasted longer thana year,its battledeaths had to reachan annualaverageof 1,000"(Small and Singer 1982, 56; see also Sarkees 2000, 129). Eventually,that criterionwas relaxed, and nonsystem memberdeaths were taken into consideration.But could we assume thatwe would get an accuratecount of colonized people's deaths in uprisings against the metropole given the imperial powers' contempt for human life in their colonies? If data measurement problems for extrasystemicwarsaresystematicallydifferentfromdataproblemsin othercivil wars, then combining the two categories may bias analyses.25 The Mau-Mau(1952-1955) rebellion in Kenya, for example, was excluded from COWbecause it caused only 591 deathson the side of the United Kingdom (Singer and Small 1972, 397). The war was addedto the 1994 and 2000 updatesof COWas a resultof countingKenyandeaths, but how manyother such warsfailed even to appearin the original"excluded-wars" list in the first COW publications?One such example is the Rwandanrevolution (against Belgium), which Fearonand Laitin(2003) code as a war between 1956 and 1961, but this case does not appearin any COW list. A lingering concern with extrastatewars, therefore,is if the original coding rules have hamperedour ability to compile a list according to the current definition without having to engage in painstaking and resource-intensivehistoricalresearch. Another importantquestion is if extrasystemic wars should be coded as taking of the system member)or in the territoryof place in the metropole(i.e., in the territory the nonsystemmember.Given our territorial conception of the state(and of civil war), it seems clear thatthe war should be coded in the territoryof the metropole,but this is not establishedpracticeacrossthe literature. Gleditschet al. (2001), for example, code wars under colonized states. extrasystemic Doyle and Sambanis(2000) and Licklider code some as civil wars wars(wherethe countrywas undertrust(1995) extrasystemic in case of as the or in the case of the warin Zimbabwe/Rhodesiafrom Namibia, eeship, 1972 to 1980, given Rhodesia's independentstatus). Otherscholarshave committed more obvious errors in coding, for example, Bangladesh's war of independence in 1971 as takingplace in Bangladeshratherthan Pakistan(Leitenberg2001), when it is
25. This may be an empirical question because governments everywhere have an incentive to the numberof their subjects that they kill. underreport
827
clear thatBangladeshdid not exist as an independentstateuntil afterthe conclusion of that war. In Sarkees and Singer (2001), the intention seems to be to code extrastate wars as having taken place in the territoryof the system member. This bringsup anotherdifficulty.On one hand,extrastatewarsthatarecoded as taking place in colonized states will have to be dropped from quantitativeanalyses because data for the "usual suspect" explanatory variables (e.g., democracy, gross domestic product[GDP]) are typically not availablefor dependentterritories.On the other hand, if the war is coded as taking place in the territoryof the system member, then explanatoryvariableshave to be adjustedfor the whole empire.Estimatingaverfor entireempires would be age GDP per capitaor the level of ethnic fractionalization a very difficult task and could only be done by taking shortcutsthat might seriously compromisethe qualityof the data.Consideralso thataveragingGDP per capitaover entire empires will have the effect of turning highly developed countries, such as Franceor England,into middle- to low-income countries.If GDP per capitais used as a proxy for state strength,then averagingGDP values for empires will have the effect state capacity in the metropoles (although it may accuratelycapof underestimating ture state capacity with respect to the rest of the empire). But distance from the metropole and military technology are importantmissing variables here that could explain the metropole'sreach.26 Empireswere uniquely autocraticregimes, in which subjects lived underdifferent forms of government,and one argumentfor setting extrastatewars apartis that the legal structureof empires preventedthe articulationof colonized peoples' demands (voice) and left them with rebellion as their only option. Might right-handside controls for the level of democracyor autocracycapturethis pressureto rebel in colonies? Consider the example of France,which, according to Fearon and Laitin (2003), had several civil wars from 1945 to 1960. France scores as a "deep"democracyover the relevant period in the Polity IV database (Marshalland Jaggers 2000). However, a standardnotion of democracy is that it provides checks and balances to resolve disto individuals putes peacefully and that it affordsthe right of political representation and groups. But this notion does not apply to the peripheryof an empire where disputes are resolved despotically, even when they are resolved by consensus in the metropole. Fearon and Laitin attemptto correct for this by using polity scores for empires thatare weighted by the shareof the colonized subjects to the populationof the metropole,therebyreducingthe polity score of countries, such as Franceor England. But it is not clear to me thatthe combinationsof regime characteristicsthat the Polity IV databaseuses are even applicableto cases of foreign dominationor empire. "Wateringdown" France's, Britain's, or Belgium's democracy score in this way is likely to make these countriesseem like so-called "anocracies"(i.e., regimes scoring in the middle of the Polity range).But one could arguethat as far as colonial subjects were concerned, they were living in autocracies, whereas citizens in the metropole were living in a democracy.Thus,consideringFranceor Englandas anocraciesis dou26. Moreover,the waves of decolonizationwars in the 1950s and 1960s were all relatedto systemic changes that affected entire groups of countries in similar ways. This, at a minimum,calls for the use of controls for systemic trends. appropriate
828
bly wrong, and the meaning of an anocracy is here inconsistent with the theoretical hypotheses that scholars usually test by using the anocracy measure.27 right-handside Perhapsall these problemscan be addressedby addingappropriate controls to the regressionsby denoting, for example, if a countrypossessed colonies. Regional inequality,for example, may be more significantin explainingpolitical violence in a world full of empires than in a world composed mainly of nation-states.28 But even tryingto measureinequalityin the Britishempireor in the Belgian Congo is likely to be systematicallymore difficult thanmeasuringit in the averagenonimperial nation-state.If so, estimates based on the variableswith systematic coding errorfor empires are likely to be biased. Thus, abandoningthe distinctionbetween metropole andperipherymay havebeen ill-advised. At its core, the earlierCOWdistinctionhighlighted that some peripherieswere constitutionallyexcluded from the political proobstacles to the peaceful resolutionof political cess, which presentedinsurmountable with the metropole. disputes Despite these arguments,in many respects, extrastatewars are indeed similar to civil wars:rebels are mostly locally recruited,the groups have a political agenda, and the governingauthorityis involvedin the fighting. Thus, analystscould add these wars controls to to their civil war lists, but doing so would necessitate adding appropriate the statisticalmodel, as outlined above. Yet, only colonial wars should be considered rule because for inclusion.29 Imperialwars of annexationdo not meet the territoriality the nonsystem member'sterritorywas never partof the metropole'sterritorypriorto the war. Wars in East Timor,Tibet, or the WesternSaharacould be considered civil warsbecause(a) the fightingthatstartedduringa warof annexationcontinuesafterthe territoryis annexed by an imperial power, and/or (b) the annexation is de facto community.;o accepted by the international CODING CIVIL WAR Drawingon the precedinganalysis, I propose an operationaldefinitionof civil war thatresolves some of the problemswe have encounteredso far.Partsof my coding rule are new, and partsarebased on other,widely used coding rules. I providea list of civil
27. One hypothesis (Hegreet al. 2001) is that anocraciesare proneto violent rebellion because they areneitherautocratic enoughto preemptor crushrebellionnordemocraticenoughto resolve conflicts peacefully. Manyimperialpowerswere bothautocraticenough (in theircolonies) anddemocraticenough (in their metropoles). 28. A common notion of empire is that it uses the peripheryto extractresourceswith no regardfor equality;this may also be truein some nation-states,althoughthereis likely to be a systematicdifferencein the degreeof regionalinequalitybetweenempiresand nation-stateswith the same level of democracyat the metropole. 29. Warsof decolonizationmay be combinedwith lists of secessionist warsbecause they are likely to sharesome of the same causal logic. However,the difficulties associated with the measurementof key variables in empires (see above) would still apply. 30. Such cases in my data set include Morocco (WesternSahara),Indonesia(East Timor),and Israel (West Bank and Gaza). Excludedis, for example, the Malaysian Insurgency,because the level of violence afterindependencewas very low, althoughdeathsduringthe phase of decolonizationwere high (see online thenthis should by anotherstateto preventannexationof the territory, supplement).If thereis an intervention be consideredan interstateconflict, even if local partiesjoin the fighting.
829
wars thatconformsto the coding rule proposedhere and lateruse the new civil war list in an empiricalanalysis. Extensive notes thatjustify the coding of each case are available in a supplementposted online."3 An armedconflict should be classified as a civil war if
of a statethatis a memberof the international (a) The wartakesplace withinthe territory sysof or with a 500,000 greater.33 population tem32 (b) The partiesarepolitically andmilitarilyorganized,andthey have publicly statedpolitical objectives.34 (c) The government(throughits militaryor militias) must be a principalcombatant.If there the governmentinternationally is no functioninggovernment,thenthe partyrepresenting and/orclaiming the state domestically must be involved as a combatant.35 (d) The main insurgentorganization(s)must be locally representedand must recruitlocally. need not imply thatthe waris not intraAdditionalexternalinvolvementandrecruitment state.36Insurgentgroups may operate from neighboring countries, but they must also have some territorial control(bases) in the civil warcountryand/orthe rebels mustreside in the civil war country.37 (e) The startyear of the war is the first year that the conflict causes at least 500 to 1,000 If the conflict has not caused 500 deaths or more in the first year, the war is deaths.38
moredemandingthanmost of the others.Sometimes, the informationto code each case with certaintyis not available, so I identify those cases in the notes. I usually err on the side of caution and include ambiguous can decide if they wantto includethemor dropthem cases, identifyingthemin the dataset so thatresearchers from the analysis. 32. This includes states that are occupying foreign territoriesthat are claiming independence(e.g., West Bank andGaza in IsraelandWesternSaharain Morocco). A strictapplicationof this coding rulecould community(throughthe UnitedNations) rejectsthe state'sclaims dropthose cases in which the international of sovereignty on the occupied territories. 33. Wecould includecountriesaftertheirpopulationreachesthe 500,000 markor, fromthe startof the period,if populationexceeds the 500,000 markat some point in the countryseries. If a civil war occurs in a countrywith populationbelow the threshold,we could include it andflag it as a marginalcase. Cases of civil warclose to the 500,000 markareCyprusin 1963 (578,000 population)and Djiboutiin 1991 (450,000 population). The per capita death measurewould allow us to relax the populationthreshold. 34. This should apply to the majorityof the partiesin the conflict. This criteriondistinguishesinsurgent groupsandpoliticalpartiesfromcriminalgangs andriotousmobs. But the distinctionbetweencriminal and political violence may fade in some countries (e.g., Colombia after 1993). "Terrorist" organizations would qualify as insurgentgroupsaccordingto this coding rule,if they caused violence at the requiredlevels for war(see othercriteria).Noncombatant populationsthatareoften victimized in civil warsare not considto the warif they arenot organizedin a militiaor othersuch form,able to applyviolence in purereda "party" suit of their political objectives. 35. Extensive indirect support (monetary,organizational,military) by the governmentto militias mightalso satisfy this criterion(e.g., Kenyaduringthe Rift Valleyethnic clashes), althoughhere it becomes harderto distinguishcivil warfromcommunalviolence. In some cases, where the statehas collapsed, it may not be possible to identify partiesrepresentingthe state because all parties may be claiming the state, and these conflicts will also be hardto distinguishfrom intercommunalviolence (e.g., Somalia after 1991). war can be taking place at the same time as interstatewar. 36. Intrastate The Bay of Pigs, for examconflicts with no local participation. 37. This weeds out entirelyinterstate ple, would be excluded as a civil war because the rebels did not have a base in Cuba priorto the invasion. Some cases stretchthe limits of this definitionalcriterion-for example, Rwandain the late 1990s, when exFAR(RwandanArmyForces)recruitswith bases in the DemocraticRepublicof the Congo engagedin incursions and borderclashes againstgovernmentarmyand civilians. If this is a civil war,then so is the conflict between Lebanon-basedHezbollah and Israel (assuming the other criteriaare met).. 38. This rulecan be relaxedto a rangeof 100 to 1,000 because fighting might startlate in the year (cf. Senegal or Peru).Given the lack of high-qualitydata to accuratelycode civil war onset, if we do not have a large-scalearmed good estimateof deathsfor the firstyear,we can code the onset at the firstyear of reported
Note thatthiscodingruleis 31. Go to http://pantheon.yale.edu/-ns237/index/research.html#Data.
830
RESOLUTION JOURNALOF CONFLICT coded as having started in that year only if cumulative deaths in the next 3 years reach
1,000.39
(f) Throughout its duration, the conflict must be characterized by sustained violence, at least at the minor or intermediate level. There should be no 3-year period during which the conflict causes fewer than 500 deaths.40 (g) Throughout the war, the weaker party must be able to mount effective resistance. Effective resistance is measured by at least 100 deaths inflicted on the stronger party. A substantial number of these deaths must occur in the first year of the war.4 But if the violence becomes effectively one-sided, even if the aggregate effective-resistance threshold of 100 deaths has already been met, the civil war must be coded as having ended, and a politicide or other form of one-sided violence must be coded as having started.42 (h) A peace treaty that produces at least 6 months of peace marks an end to the war.43 (i) A decisive military victory by the rebels that produces a new regime should mark the end of the war.44Because civil war is understood as an armed conflict against the government, continuing armed conflict against a new government implies a new civil war.45If the government wins the war, a period of peace longer than 6 months must persist before we code a new war (see also criterion k). conflict, providedthatviolence continuesor escalates in the following years. Note thatin the data set, I also code the start/endmonth,where possible. In some cases, my coding rules can be used to identify the start month (e.g., in cases where the war causes 1,000 deaths in the first month of armedconflict). But in most cases, the monthonly indicatesthe startof majorarmedconflict or the signing of a peace agreement,which can give us a point of referencefor the startor end of the war,respectively. 39. This rulealso suggests when to code war terminationif the 3-yearaveragedoes not add up to 500. In such a case, we can code the end of the war at the last year with more than 100 deaths unless one of the other rules applies (e.g., if there is a peace treatythat is followed by more than 6 monthsof peace). the durationof a con40. This criterionmakescoding verydifficultbecausedataon deathsthroughout flict arehardto find.However,such a coding ruleis necessaryto preventone fromcoding too manywarstarts in the same conflict or coding an ongoing civil war for years afterthe violence has ended. Three years is an The datanotes (see online cutoff pointbutis consistentwith otherthresholdsfound in the literature. arbitrary supplement)give severalexamples of cases in which the coding of warterminationhas been determinedby this criterion.A more lenient version would be a 5-year thresholdwith fewer than 500 deaths. to the war's intensity in the first years of the war. If the war's 41. This criterionmust be proportional onset is coded the firstyearwith only 100 deaths(as often happensin low-intensityconflicts), then we would not be able to observeeffective resistancein the firstyear of the war if we definedeffective resistanceas 100 deaths sufferedby the state. 42. This criteriondistinguishescases in which insurgentviolence was limited to the outbreakof the war and, for the remainderof the conflict, the governmentengaged in one-sided violence. A hypothetical example is a case in which insurgentsinflicted 100 deaths on the governmentduringthe first week of fighting, andthenthe governmentdefeatedthe insurgentsandengagedin pogromsandpoliticide for severalyears with no or few deathson the government'sside. If we cannot apply this rule consistently to all cases (due to data limitations),then periodsof politicide at the startor end of the war should be combined with war periods. This implies that civil wars will often be observationallyequivalent to coups that are followed by politicide or other such sequences of differentforms of political violence. 43. Treatiesthatdo not stop the fighting are not considered(e.g., the IslamabadAccords of 1993 in Afghanistan'swar;the December 1997 agreementamong Somali clan leaders). If several insurgentgroups areengaged in the war,the majorityof groupsmustsign. This criterionis useful for the studyof peace transiareinterestedin studyingcivil warduration,forexample. if researchers tionsbutmay not be as important 44. Thus, in secessionist wars that are won by the rebels who establish a new state, if a war erupts immediatelyin the new state, we would code a new war onset in the new state (an example is Croatiafrom 1992 to 1995), even if the violence is closely relatedto the precedingwar.A continuationof the old conflict between the old partiescould now count as an interstatewar, as in the case of Ethiopiaand Eritrea,which fought a war between 1998 and 2000 after Eritrea'ssuccessful secession from Ethiopiain 1993. to study the stabilityof militaryvictories. Analysis of the stabil45. This criterionallows researchers ity of civil waroutcomes would be biased if we coded an end to civil warthroughmilitaryvictory only when the victory was followed by a prolongedperiodof peace. This would bias the resultsin favorof findinga posfor analyzingwar This criterionis important itivecorrelationbetweenmilitaryoutcomesandpeace duration. recurrencebut not necessarily war prevalence.
Sambanis/ WHATIS CIVILWAR?
831
theendof a civilwarif they canalsomark anendto fighting orsimply truce, (j)A cease-fire, The periodof peacemustbe longerthanwhatis resultin at least2 yearsof peace.46 of theparwe donothaveclearsignals because inthecaseof a peaceagreement required in thecaseof a truce/cease-fire.47 an agreement ties'intentto negotiate to be coded,subject enter thewarovernewissues,a newwaronsetshould (k)If newparties to waroverthesameissues,we return If thesameparties thesameoperational criteria.48 forcoding unlessanyof theabovecriteria of theoldwar, codethecontinuation generally of fighting. beforetheresurgence a war'sendapplyfortheperiod Using these coding rules, I have coded 145 civil waronsets between 1945 and 1999 (2.08% of 6,966 nonmissing observations).Withoutcoding new war onsets in countries with alreadyongoing civil wars, the numberof civil wars is 119 (1.93% of 6,153 nonmissing observations).Out of these cases, 20 are "ambiguous"-that is, they may not meet one or more of the coding rules. DIFFERENCES IN THE CODING OF CIVIL WAR AND THEIR SUBSTANTIVE IMPLICATIONS The discussion thusfarhas establishedsome of the sourcesof disagreementamong civil warlists. Disagreementsover the coded yearof onset and terminationof civil war may matterfor the inferences drawn when we analyze civil war onset, duration,or recurrenceusing differentdatasets. The extentof disagreementover the coded yearof civil war onset is apparentin Table l a-d, which presents correlationsbetween war startsduringthe periodfrom 1960 to 1993-a periodcovered by most datasets.49The The dependentvariableis civil waronset, a binary unitof analysis is the country-year. variable.All yearsof no wararecoded equal to 0. Therearetwo versionsof coding war starts:in version (a), I code a 1 when a civil war startsand drop observationsof ongoing war in thatcountryuntilthe warends. Thus, if anotherwar startsin the same country while anotherwar is ongoing, we would not consider it. In version (b), I code a I whenevera war starts,even if anotherwar is ongoing. Country-yearswith no new war startsare coded 0, and in this way, we end up with more war starts.'5
deathsor, in a lenientversionof this criterion,fewerdeathsthanthe 46. Peace implies no battle-related lowest thresholdof deaths used to code waronset, that is, fewer than 100 deaths per year. 47. These situationsaredifferentfromthose in which thereis no violence as a resultof armiesstanding down without a cease-fire agreement,which would fall undercriterion(f). 48. These incompatibilitiesmust be significantly different,or the wars must be fought by different groupsin differentregions of the country.For example, we would code threepartiallyoverlappingwars in Oromo)betweenthe 1970s and the 1990s. New issues alone shouldnot be suffiEthiopia(Tigrean,Eritrean, cient to code a new war because there is no "issue-based"classification in the definition of civil war. We could applysuch a ruleif we classified civil warsinto categories-for example,secessionist warsversusrevolutions over controlof the state. In additionto having new issues, most partiesmust also be new before we can code a new war onset. 49. See the online supplementfor a summaryof the definitionsandoperationalcriteriaused by major dataprojects.The periodwas selected to includethe Collier and Hoeffler(2001) dataset (1960-1999) in the Most of the otherdata sets startin 1945. I comanalysis, given the prominenceof thatstudyin the literature. pare results from a smallernumberof data sets covering the entire post-1945 period. 50. Collier and Hoeffler(2001), Hegreet al. (2001), and Sambanis(2001) use version(a). Fearonand Laitin (2003) use version (b). (text continues on p. 835)
00
TABLE I
Correlationsamong Civil WarLists, 1960-1993
a. Version(a) of WarOnset (3,198 Obser Collier and COW1994 COW2000 warstla warstla warst2a warst3a warst4a warst5a warst7a warst8a warst9a warstl0a warstlla warstnsa 1.00 0.96 0.82 0.74 0.42 0.69 0.60 0.70 0.69 0.76 0.74 warst2a Hoeffler (2001) warst3a Licklider Gleditsch (1995) et al. (2001) warst4a warst5a Fearon and Laitin (2003) warst7a
Leitenbe (2000)
warst8
1.00 0.83 0.75 0.46 0.70 0.66 0.70 0.69 0.76 0.74
1.00 0.71 0.52 0.70 0.56 0.66 0.80 0.75 0.73
1.00 0.57 0.70 0.66 0.66 0.68 0.79 0.83
1.00 0.54 0.46 0.46 0.48 0.53 0.51
1.00 0.59 0.67 0.72 0.77 0.80
1.00 0.59 0.55 0.61 0.62
b. Version(b) of WarOnset (3,503 Obse Collier and Hoeffler (2001) warst3b Fearon and Laitin (2003) warst7b
COW1994 COW2000 warstlb warstlb warst2b warst3b warst4b warst5b warst7b warst8b warst9b warstl0 warstllb warstnsb 1.00 0.92 0.83 0.57 0.37 0.58 0.51 0.74 0.69 0.67 0.64 warst2b
Licklider Gleditsch (1995) et al. (2001) warst4b warst5b
Leitenb (2000
warst8
1.00 0.83 0.55 0.45 0.61 0.55 0.70 0.67 0.66 0.64
1.00 0.62 0.46 0.61 0.49 0.68 0.80 0.73 0.68
1.00 0.40 0.58 0.50 0.52 0.60 0.68 0.69
1.00 0.49 0.37 0.40 0.41 0.45 0.44
1.00 0.56 0.58 0.61 0.63 0.69
1.00 0.51 0.46 0.50 0.51
00
0o
44.
TABLE 1 (continued) c. Correlationsamong Lists of Civil WarPrevalence Collier and Hoeffler (2001) atwar3 Fearon and Laitin (2003) atwar7
COW1994 COW2000 atwarl atwarl atwar2 atwar3 atwar4 atwar5 atwar7 atwar8 atwar9 atwarl0 atwarll atwarns 1.00 0.94 0.88 0.75 0.62 0.67 0.63 0.62 0.70 0.69 0.69 atwar2
Licklider Gleditsch (1995) et al. (2001) atwar4 atwar5
Leitenbe (2001)
atwar8
1.00 0.90 0.76 0.69 0.70 0.67 0.66 0.73 0.73 0.74
1.00 0.82 0.70 0.73 0.66 0.64 0.77 0.75 0.77
1.00 0.68 0.75 0.70 0.66 0.74 0.78 0.79
1.00 0.65 0.59 0.53 0.64 0.66 0.66
1.00 0.70 0.70 0.76 0.78 0.83
1.00 0.62 0.66 0.67 0.69
d. WarOnsets and Prevalencein Data Set Singer and Small (1994) Sarkeesand Singer (2000) Collier and Hoeffler (2001) Licklider(1995) Gleditsch et al. (2001) Fearonand Laitin(2003) Regan (1996) Doyle and Sambanis(2000) Sambanis(currentstudy) Numberof WarStarts 61 73 70 58 83 79 116 100 102
NOTE:Yearsfor which the dependentvariableis coded missing (due to lack of stateindependence or otherreason)in civil wars in those years. Leitenberg(2000) is excluded from this table because his civil war list ends before the en
835
As is evident from Table la,b, the correlationbetween most pairsof civil war onset lists is low. Among the lowest correlations(0.42 and0.46) arethose between Gleditsch et al. (2001) and the two versions of the Correlatesof Wardataset. These correlations are even lower with version (b) of the dependent variable (0.37 and 0.45, respectively).5'The correlationbetween Gleditsch et al. (2001) and my data set is 0.44 [version (b)], and the correlationwith Fearonand Laitin (2003) is 0.69 [version (b)]. The most highly correlatedcivil war lists are the two versions of the COW data set.52 The correlationbetween war prevalence in any two lists would be higher if one comparedwars thatstartedwithin the same 2- or 3-yearperiod (see Fearonand Laitin 2003 for such a comparison).But that would be a comparison of war lists, not war onsets. A differenceof 2 to 3 yearsin the coding of waronset is significantbecause values of right-hand side variables, such as economic growth or political instability, would be immediatelyinfluencedby war,which would in turninfluence the empirical results. In Table Ic, I correlatecivil war prevalence(combined onset and duration)in the differentlists. If two lists included differentwar startsbut the war overlappedfor most of its duration,then the correlationbetween the two databaseswould be higher thanin Table 1a,b. Indeed,the correlationsbetween pairsof lists are now higher (e.g., between Gleditsch et al. 2001 and COW2, it rises to 0.69). Thus, there is more disagreementon the question of war onset and terminationthan there is over whethera war happenedat all. However, there is still considerable disagreement about which armed conflicts shouldbe classified as civil wars.Many wars arecoded in only one out of a dozen data sets. Table 1d lists the numberof war startsandtotalcountry-yearsof warin the different lists from 1960 to 1993.5"The highest number of war starts (116) is found in Regan's (1996) list and the lowest two in Licklider's(1995) list (58) and Singer and Small's (1994) list (61). Now that I have establishedthat there is substantialvariationin the coding of war to ask if these differenceshave in most datasets, it is appropriate onset andtermination in tractable To answer that a substantiveimplications. manner,I regress the question variable of the civil war onset and measurevariawar model on all versions civil same are due to To the differences that the coding of the estimates. isolate in tion parameter for the side variablesand I the same sources of data use right-hand dependentvariable, to observathe annual over same same set of countries the period(1960 1993, analyze in the in should be due to differences estimates all differences Thus, tions).54 parameter
51. HereI consideronly warsfromthe Gleditschet al. (2001, 2002) list. In the regressionanalysis later on, I add a second measurethat included minor and intermediatearmedconflicts from that list. 52. If we expand the comparisonto the entire period from 1945 to 1999, the correlationsare even smaller. My data set correlates with others as follows: 0.62 with Sarkees and Singer (2000), 0.34 with Gleditsch et al. (2001), 0.63 with Fearonand Laitin (2003), and 0.67 withDoyle and Sambanis (2000). 53. The numberof observationsacross models differsin Table2 because I dropobservationsof ongoing war.Thus, any differences are directly the result of coding war onset and continuation.In Table4, the numberof observationsfor the Collier-Hoefflervariableis slightly smallerbecause the lagged war variable is missing for the year 1960 since their data set startsin 1960, and we cannot know how they would have coded the precedingyear.SingerandSmall's (1994) dataset has fewer observationsbecause it does not code events in 1993, and Leitenberg's(2001) data set ends in 1990. Othermodels all have the same N. 54. Leitenberg's(2001) data set is an exception because it ends with the end of the cold war.
836
I also restrictthe analysis to a smaller numberof lists that cover the coding rules."5 entire period from 1945 to 1999. The model is very similarto those developed by Fearonand Laitin(2003) and Collier and Hoeffler (2001). Because these studies are by now well known, I do not discuss the theorybehindthe selection of variables.Briefly, civil waronset is expected to be less likely the higher the level of development,proxied by real per capita income (gdpll).56Anocracies (anoc211),statesat the mid-rangein the Polity IV series, should have a higherriskof waras they areneitheras effective as autocraciesin repressionnor as good as democraciesin peaceful conflict resolution(FearonandLaitin2003; Hegre et al. 2001). States with political instability(inst311)andregimetransitionshouldbe at higher risk of war onset.57Ethnic heterogeneity (ef) should increase the risk of war onset by pittinggroupswith differentpreferencesagainsteach other.58 Competingtheories expect a different relationshipbetween this variable and civil war onset, and empiricalresults on the link between ethnicity and civil war are mixed. I control for population (measured by the naturallog of population, 1popnsll), which has been shown to be significantandpositively correlatedwith waronset.59 I also controlfor the of Muslims in the as a of percentage population(muslim) proxy partof the "clash-ofcivilization" hypothesis and as a measure of religious division." I control for economic growth, measuredas annualpercentagechange in the level of real per capita income (groll), because Collier and Hoeffler (2001) find this to be significantly and negatively correlatedwith war onset. I control for countries that are significant oil exporters(oil211).6'Such countriesarethoughtto be at higherrisk of warfor a number of reasons-the most commonly encounteredhypothesis is thatoil corruptspolitical institutionsor thatit generatesincentives for secessionist war.I control for countries' mountainousterrain(mtnll), following Fearonand Laitin,who view terrainas partof the technology of insurgency(mountainsprovide hideouts for rebels). I also control for a variable measuringtime at peace since the last war (pwt). All right-handside variablesare lagged 1 year.62
55. A riskis thatthe degreeof multicollinearity in the datamay be differentacrosscivil warlists, influerrorsdifferentially.But this would be a resultof differences in the coding of the dependent encing standard variable,so it does not pose a problemfor the analysis. 56. I use data from Fearonand Laitin(2003) for this variable.According to civil war theories,higher state capacityshould discouragerebellionor allow the state to repressit in its early stages. Higherdevelopment levels should discourage rebellion by raising the economic opportunitycosts of violence. 57. I use the "Polity2" series from the Polity IV dataproject,version 2002, to constructthe anocracy and instabilityvariables. 58. Ethnic heterogeneitywas constructedby Fearon (2003) and is available in Fearonand Laitin's (2003) replicationdata set. This is a constant, so I do not lag it. 59. I used data from the WorldBank and other sources to complete the populationseries. See online supplementfor more information. 60. The source for this variableis Fearonand Laitin (2003). 61. This series has severaldifferencesfromthe respectiveseries in FearonandLaitin(2003), although the same underlying sources have been used (World Bank data). I discuss all differences in the online supplement. 62. Fearonand Laitin(2003) do not lag all variables(e.g., oil is not lagged) and use the value for the secondperiodin the countryseries to fill in missing values at the startof the countryseries for those variables that are lagged.
837
This model capturesthe basic logic of prominenttheories of civil war.63I did not include FearonandLaitin's(2003) dummy variablefor "newstates"because this variable perfectlypredictsthe outcome when I lag all independentvariables.6 Moreover, it is not clearto me whatthatvariablemeasuresthatis not alreadycapturedby the conand"income.""New"statesaremore likely to have a trolsfor "instability," "anocracy," civil war than "old"states, as Fearonand Laitin found, but they are also more ethnically diversethan"old"states.Controllingfor "newstates"may artificiallyreduce the significance of ethnic fractionalization(ef). I say "artificially"because the fact that these statesare "new"may well be causally linked to theirhigh ethnic fragmentation. thancounMost "newstates"are formercolonies with higherethnic fractionalization effective of in a has reduced the level ethnic which tries nation-building process fragmentation.It might thereforebe the case that the "newness"of the state explains the Alternatively,looking furtherupstreamin the high degree of ethnic fractionalization. chain of causality,the "newness"of the state may also be a consequence of demands for self-determination by ethnicallydistinctgroups thata distantmetropolehad never to integrate into a single nation in the predecessor state.65 or succeeded attempted Eitherway, if thereis a causal connection between new state and ef, we cannotcontrol for both in the regression,and ef is the more theoreticallyinterestingconcept. We can now check if the correlationsbetween these key variablesand civil war are influencedby the coding of the dependentvariable.In what follows, I estimate probit models correspondingto 12 differentcoding rules and two versions of the war onset variableand for overall war prevalence.I review these estimates, looking for changes in significance levels and very largechanges in coefficient estimates(more than 1 or 2 standarddeviations), because such volatility would have importantpolicy implications in assessing the link betweenchanges in the explanatoryvariablesandchanges in the probabilityof civil war onset. I present the results of these estimations in several Table2 includes estimationresults for 12 regressionsusing version (a) of the tables.66 dependentvariable,and Table 3 summarizesthe range of these parameterestimates. Table4 includes estimationresultsfor 12 regressionsusing version (b) of the depend63. This is, of course, only one of several possible specificationsof the model. I do not include some variablesthat others have found significant because I find the theoreticaljustification for including them or becauseothervariablesincludedin the model arelikely to capturemuchof whatthe excluded problematic because it is, by variablesaremeasuring.Forexample, I do not include a controlfor noncontiguousterritory itself, not a good controlof statestrengthwithoutalso controllingfor militarytechnology and ability to project militarypower.It may also be the case thatstrongerstatesonly havethe capacityto maintainnoncontiguous territories.Although this certainlydoes not apply to each case, an equality of means test reveals that countrieswith noncontiguousterritorieshave significantly higher averageincome ($5,855) than countries with only contiguous territories($3,223). The t statistic for this test is -17.78. 64. Forexample, in my dataset, thereareonly five warstartsthatoccurin the second year of the country series and for which income is nonmissing,but all five are droppeddue to missing values in the income growth variable. 65. A logit regressionof "newstate"on "ethnicfractionalization" (ef) yields a statisticallysignificant and large coefficient (2.69) for ef(z value = 5.52) with 316 observations(I restrictedthe sample to the first two observationsin each countryseries). 66. I do not reportfixed-effects resultsbecause fixed-effects estimation is very sensitive to measurementerror(hence we can expect the coding ruledifferencesto influencethe results).I presentthese resultsin differencesacrosswarlists, even among variablesthatarerobust the online supplement.Thereareimportant in ordinarylogit models. (text continues on p. 840)
00
Collier and Hoeffler (2001) warst3 -0.097 (0.033) 0.493 (0.810) 0.131 (0.153) 0.199 (0.133) 0.180 (0.119) 0.175 (0.200) 0.104 (0.032) 0.006 (0.003) Licklider (1995) warst4 -0.075 (0.030) 0.564 (0.760) 0.238 (0.154) 0.234 (0.147) 0.188 (0.149) 0.209 (0.217) 0.113 (0.037) 0.003 (0.002) Gleditsch et al (2001) (Wars) warst5 -0.039 (0.020) -1.218 (0.798) 0.263 (0.126) 0.217 (0.130) 0.204 (0.145) 0.317 (0.258) 0.111 (0.033) 0.004 (0.002)
TABLE2
ProbitModels of Civil WarOnset, 1960-1993

Gleditsch et al. (2001) (All) warst6 -0.047 (0.018) -0.560 (0.542) 0.228 (0.104) 0.287 (0.089) 0.170 (0.150) 0.524 (0.180) 0.090 (0.032) 0.002 (0.002) Fearon and Laitin (2003) warst7 -0.093 (0.032) -0.225 (0.787) 0.239 (0.143) 0.251 (0.136) -0.060 (0.142) -0.006 (0.210) 0.109 (0.033) 0.004 (0.003)
COW1994 COW2000 Variable GDP GDP growth Instability Anocracy Oil exporter Ethnicfractionalization Population (log) Terrain warst1 -0.109 (0.032) -0.384 (1.015) 0.339 (0.147) 0.212 (0.141) 0.274 (0.186) 0.118 (0.225) 0.102 (0.032) 0.005 (0.003) warst2 -0.114 (0.040) -0.237 (0.898) 0.321 (0.134) 0.233 (0.121) 0.222 (0.157) 0.144 (0.206) 0.090 (0.030) 0.005 (0.003)
Leiten (200
wa
-0.0 (0.0 0.8 (1.0 0.3 (0.1 0.2 (0.1 0.3 (0.1
0.2 (0.2
0.1 (0.0 0.0 (0.0
Percentage Muslim Peace duration
-0.001 (0.002)
0.001 (0.002)
0.002 (0.001)
0.003 (0.001)
0.002 (0.002)
0.003 (0.001)
0.003 (0.001)
0. (0.
0.000 -0.002 -0.012 -0.0 -0.006 0.000 -0.004 -0.003 (0.004) (0.005) (0.005) (0.006) (0. (0.006) (0.006) (0.005) -3.668 -3.903 -3.987 -4.147 -4.0 -4.014 -3.626 -3.844 Constant (0.587) (0.506) (0.574) (0.718) (0. (0.527) (0.582) (0.563) 3,611 3,926 3,530 Observations 3,770 3,372 3,765 3,861 3,815 Log -250.53 -290.74 -435.29 -224.91 -238.1 -263.55 -254.81 -229.21 likelihood 82.62 102.21 89.31 48. 73.59 72.15 75.69 67.38 WaldX2(10) 0.0877 0.0719 0.0934 0.0944 0. 0.0876 0.0995 0.0933 Pseudo-R2
afterstart.Bold indicatessignificanceat . NOTE:Coefficients(standard errors)arepresented.Ongoingwarsdropped domestic product.
00
840
TABLE 3
Summaryof ParameterEstimatesfrom Table 2

Model 1 Variables Observations 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 Mean -0.084 0.029 -0.031 0.776 0.273 0.136 0.217 0.127 0.177 0.146 0.255 0.211 0.101 0.033 0.003 0.003 0.001 0.001 -0.004 0.005 -3.820 0.582 Standard Deviation 0.024 0.006 0.550 0.152 0.066 0.016 0.048 0.014 0.104 0.018 0.139 0.019 0.012 0.002 0.002 0.000 0.001 0.000 0.003 0.001 0.224 0.053 Minimum Maximum -0.114 0.018 -1.218 0.542 0.131 0.104 0.082 0.089 -0.060 0.119 -0.006 0.180 0.076 0.030 0.001 0.002 -0.001 0.001 -0.012 0.004 -4.147 0.506 -0.039 0.040 0.816 1.026 0.382 0.154 0.287 0.147 0.371 0.186 0.524 0.258 0.116 0.037 0.006 0.003 0.003 0.002 0.000 0.006 -3.370 0.718
GDP coefficient GDP SE Growthcoefficient GrowthSE Instabilitycoefficient InstabilitySE Anocracycoefficient AnocracySE Oil coefficient Oil SE Ethnicfractionalizationcoefficient SE Ethnic fractionalization Populationlog coefficient Populationlog SE Terraincoefficient Terrain SE Muslim coefficient Muslim SE Peace durationcoefficient Peace durationSE Constantcoefficient ConstantSE
foreach variable. NOTE:Meancoefficientsandstandard errors arepresented GDP = grossdomesticproduct.
ent variable.Table4 models allow us to gauge the effect of war in the previousperiod Table 5 (warll) on the probability of a new war onset in the same country.67 estimates in Table 4. summarizesthe range of parameter One of the most robustvariablesis income per capita. Gdpll is always highly significant, although its coefficient as reportedin Table 3 varies considerably,ranging from-0.039 to -0.114. In Table4, the resultsare similar,with a slightly largerrangeof variation.We can see in Tables3 and5 thatthe coefficient moves by slightly morethan 2 standard deviations,dependingon the coding of the dependentvariable.But the relationship between income and civil war is extremely robust. Growth(groll) is never significantand switches sign in about half the regressions in Tables2 and4.6 Othernonsignificantvariablesdo not switch signs as frequentlyas growth, and the switching sign of growth may suggest a particularly strong
67. InFearonandLaitin's(2003) dataset, 14 warsstartwhile anotherwaris ongoing in the same countryfrom 1945 to 1999. In my dataset, thatnumberis 29 and 17 for the periodfrom 1960 to 1993. The warll variablecapturessome of the time dependenceof peace and war so it replacesthe peacetimevariable(pwt) that I use in Table 2 models. 68. Using a 3-year growthratedid not change this. By lagging growth,we lose the first two observations in the countryseries because we must lag gdpen by I year to calculate the growthrate(gdpgro),which cannotbe calculatedfor the firstyearof the series. I coded one version of the growthvariable,grollm, where (text continues on p. 843)
TABLE4

Collier and Hoeffler (2001) warst3b -0.092 (0.029) 0.195 (0.716) 0.130 (0.145) 0.226 (0.121) 0.194 (0.130) 0.295 (0.203) 0.094 (0.032) 0.006 (0.003) 0.003 (0.001) -0.447 (0.200) Gleditsch et al. (2001) (Wars) warst5b -0.056 (0.023) -1.087 (0.804) 0.284 (0.119) 0.204 (0.132) 0.240 (0.141) 0.391 (0.277) 0.143 (0.035) 0.003 (0.002) 0.002 (0.002) -0.005 (0.169) Gleditsch et al. (2001) (All) warst6b -0.054 (0.016) -0.895 (0.560) 0.134 (0.094) 0.247 (0.082) 0.302 (0.116) 0.591 (0.169) 0.133 (0.033) 0.002 (0.002) 0.002 (0.001) -0.103 (0.107) Fearon and Laitin (2003) warst7b -0.098 (0.029) -0.351 (0.774) 0.167 (0.131) 0.239 (0.124) 0.111 (0.140) 0.084 (0.188) 0.131 (0.031) 0.004 (0.002) 0.003 (0.001) -0.274 (0.124)
COW1994 COW2000 Variable GDP GDP growth Instability Anocracy Oil exporter Ethnic fractionalization warstIb -0.126 (0.035) -0.295 (0.914) 0.316 (0.161) 0.314 (0.139) 0.358 (0.180) warst2b -0.134 (0.039) -0.072 (0.791) 0.348 (0.145) 0.289 (0.127) 0.291 (0.155) 0.157 (0.199) 0.083 (0.029) 0.006 (0.003) 0.000 (0.002) -0.351 (0.176)
Licklider (1995) warst4b -0.077 (0.028) 0.584 (0.738) 0.217 (0.152) 0.215 (0.143) 0.222 (0.156) 0.260 (0.222) 0.129 (0.045) 0.002 (0.002) 0.003 (0.001) -0.677 (0.236)
Leitenberg (2001) warst8b -0.088 (0.030) 0.431 (0.896) 0.393 (0.143) 0.170 (0.123) 0.373 (0.153) 0.075 (0.228) 0.101 (0.028) 0.001 (0.003) -0.001 (0.001) -0.371 (0.215)
0.095 (0.212) Population(log) 0.090 (0.031) 0.005 Terrain (0.003) -0.001 Percentage Muslim (0.002) -0.351 Warat (t - 1) (0.192)
40
ti
TABLE4 (continued) Collier and Hoeffler (2001) warst3b Gleditsch et al. (2001) (Wars) warst5b Gleditsch et al. (2001) (All) warst6b Fearon and Laitin (2003) warst7b
COW1994 COW2000 Variable Constant warst1b warst2b
Licklider (1995) warst4b
Leitenberg (2001) warst8b
-4.289 -4.353 -4.733 -4.520 -3.747 -3.575 -3.942 -3.669 (0.530) (0.605) (0.545) (0.787) (0.530) (0.581) (0.516) (0.554) 4,179 4,179 4,179 Observations 4,045 4,178 3,782 4 4,099 4,179 Log -561.33 -303.22 -334.42 -237.15 -270.64 -266.58 -289.52 -248.10 likelihood 101.28 75.07 102.19 74.27 40.53 79.28 75.12 63.13 WaldX2(10) 0.0844 0.0908 0.0838 0.0952 0.0715 0.0863 0.1026 0.0936 Pseudo-R2
NOTE:Ongoingwarscoded 0 if no new warstarts.Coefficients(standard errors)arepresented.Bold indicatessignificanc GDP = gross domestic product.
843
TABLE 5
Estimates from Table 4 Summaryof Parameter

Model 2 Variables Observations 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 Mean -0.093 0.028 -0.199 0.705 0.249 0.131 0.243 0.121 0.238 0.140 0.280 0.206 0.104 0.033 0.003 0.002 0.001 0.001 -0.286 0.164 -3.915 0.587 Standard Deviation 0.024 0.006 0.489 0.137 0.085 0.020 0.043 0.016 0.080 0.021 0.155 0.027 0.026 0.005 0.002 0.000 0.001 0.000 0.212 0.041 0.493 0.081 Minimum -0.134 0.016 -1.087 0.529 0.130 0.094 0.170 0.082 0.111 0.106 0.075 0.169 0.050 0.028 0.001 0.002 -0.001 0.001 -0.677 0.107 -4.733 0.516 Maximum -0.054 0.039 0.584 0.914 0.393 0.161 0.314 0.143 0.373 0.180 0.591 0.277 0.143 0.045 0.006 0.003 0.003 0.002 0.101 0.236 -2.911 0.787
GDP coefficient GDP SE Growthcoefficient GrowthSE Instabilitycoefficient InstabilitySE Anocracycoefficient AnocracySE Oil coefficient Oil SE Ethnic fractionalizationcoefficient SE Ethnic fractionalization Populationlog coefficient Populationlog SE Terraincoefficient TerrainSE Muslim coefficient Muslim SE Lagged war coefficient Lagged war SE Constantcoefficient ConstantSE
GDP= grossdomesticproduct. foreach variable. errors arepresented NOTE:Meancoefficientsandstandard
endogeneity problemwith this variable.Growthis very sensitive to civil war, so getting the date of onset of civil war "wrong"in some datasets should influence the coefficient sign of this variable.69 A possible explanationfor the instability of growth is that,in the immediatepostwarperiod,growthis typically very high, but these are also periodsof high risk of war recurrencedue to other influences. Thus, some periods of war recurrencemight also be high-growthperiods, whereas the general trend should be for growth to reduce the risk of civil war. Politicalinstabilityis mostly significantandalways positive (see Table2), although in two regressions(CollierandHoeffler2001; Licklider 1995) it is nonsignificant,and in anothertwo cases, it is significantonly at the .10 level. In Table4, it appearsto be
lagging is delayed by one observation,so this restoresthe second lost observationin each countryseries. In Tables2 and 4, this adds only five observations,including one war (Rwandain 1963). Using this measure insteadof groll makesanocracyslightly more significantand terraina little more significantin two regressions in Table2 andefa little less significantin Table4 (Rwandahas a low level of ef, at 0.18). The resultson growthdo not change. 69. Given my concern with endogeneity here, I dropped the growth variable and reestimated all regressions.See versions of Tables 2, 3, 4, and 5 without growth in the online supplement.The results are very similarto the ones presentedhere.
844
less robust,being significantin only half of the regressions and nonsignificantin the other half. Anocracybordersstatisticalsignificance in six regressionsin Table2 and is clearly significant only in Gleditsch et al.'s (2001) list, which combines all armedconflicts. Similarly,it is significantin only six out of the twelve regressionsin Table4. Its coefficient rangesfrom a low of 0.170 (Leitenberg2001 list) to a high of 0.314 (Singer and Small 1994 list) in Table4, and the range widens in Table 2 from a low of 0.082 to a high of 0.287. Oil exports are basically nonsignificant in Table 2, although the picture is more mixed in Table4, where it is significantin four regressionsand borderlinesignificant in anotherthree.70Its coefficient varieswildly from-0.060 to 0.371 (Table3; although it is less unstablein Table5). This variableis not robustto changes in the coding of war onset. Ethnicheterogeneity(ef) is positive and almost always nonsignificantwith a large standarderror,except for regression6 (Gleditsch et al. 2001--all armedconflict) in Table 2. Efis nearly significant (at the .10 level) in Regan's (1996) model (warst9), where again a numberof additionalwars are included because Regan uses a lower deaththreshold(200 deaths).In Table4, efis more often significant(in two models at the .05 level and in three more at the .10 level), although its coefficient has a wide range,from0.075 to 0.591 (see Table5). Thatefis generally nonsignificantin Table2 and muchmore significantin Table4 may have something to do with those extrawars that we are able to include when we use version (b) of war onset.7' The coefficient of population (lpopnsll) is quite stable and always positive and highly significant,except for model 9 in Table4, where it is not significant.This interesting resultseems to justify my earlierclaim thatthe significance of populationsize may be an artifactof the high absolute thresholdof deaths imposed to classify a civil war. In model 9, Regan's lower death thresholdcaptures more small wars in small countries. In both Tables 2 and 4, mountainousterrain(mtnll) is relativelystable and is only significantin the Correlatesof Warand Collier and Hoeffler (2001) data sets. Mountainousterrainis a key variablein theoriesof civil war thatemphasizethe opportunity for rebellion (e.g. Fearonand Laitin2003; Collier and Hoeffler 2001), but it structure is generally not statisticallysignificant. The percent of Muslims in a country (muslim) is marginally significant in two regressionsand in anothertwo at the .10 level (see Table 2) and becomes somewhat more significant in Table 4. There, we find it significant in the Fearon and Laitin (2003) model (warst7b)with a positive sign, despite the findings of these authorsthat religious division is nonsignificantin models of civil war onset. However,the period analyzed seems to matter,as later (see Table 6) we will find that the significance of
70. I use the variableoil2, lagged once (oil211).If I were to use Fearonand Laitin's(2003) oilll series (lagging it once), it would be significantin model 6 and,at the . 10 level, in models 8 and9 andnonsignificant in all other models, including some of those in which oil211 was significant (COW 1994 and my data). 71. Countrieswith more thanone war at the same time have slightly higherethnic fractionalization. There are 26 such cases (of chronologicallyoverlappingwars in my data set), and a means test for ethnic differencebetween them and the other cases. shows a statisticallysignificant I10-point fractionalization
TABLE6

Gleditsch et al. (2001) (Wars) Variable GDP GDP growth Instability Anocracy Oil exporter Ethnic fractionalization Population(log) Terrain PercentageMuslim Warat (t - 1) Constant
~tl
Gleditsch et al. (2001) (Wars) 2 -0.079 (0.023) 0.279 (0.088) 0.244 (0.090) 0.310 (0.128) 0.268 (0.211) 0.137 (0.025) 0.002 (0.002) 0.016 (0.143) -4.500 (0.447)
Gleditsch et al. (2001) (Wars) 3 -0.078 (0.023) -1.350 (0.539) 0.278 (0.089) 0.214 (0.095) 0.301 (0.135) 0.236 (0.208) 0.144 (0.027) 0.002 (0.002) -0.007 (0.145) -4.571 (0.474)
Fearon and Laitin (2003) 4 -0.095 (0.025) -0.396 (0.537) 0.222 (0.101) 0.265 (0.101) 0.145 (0.128) 0.254 (0.175) 0.142 (0.029) 0.003 (0.002) 0.002 (0.001) -0.309 (0.097) -4.581 (0.504)
Fearon and Laitin (2003) 5 -0.104 (0.025) 0.176 (0.096) 0.332 (0.093) 0.192 (0.119) 0.209 (0.166) 0.141 (0.029) 0.003 (0.002) -0.342 (0.104) -4.472 (0.489)
Fearon and Laitin (2003) 6 -0.099 (0.024) -1.443 (0.617) 0.186 (0.097) 0.296 (0.096) 0.173 (0.124) 0.245 (0.164) 0.148 (0.030) 0.003 (0.002) -0.364 (0.102) -4.568 (0.514)
D S
1 -0.065 (0.022) -0.962 (0.581) 0.305 (0.095) 0.157 (0.108) 0.294 (0.127) 0.417 (0.217) 0.140 (0.028) 0.003 (0.002) 0.000 (0.002) 0.010 (0.150) -4.670 (0.487)
o00
TABLE6 (continued) Gleditsch et al. (2001) (Wars) Variable Observations Log likelihood WaldX2(d/) Pseudo-R2 1 Gleditsch et al. (2001) (Wars) 2 Gleditsch et al. (2001) (Wars) 3 Fearon and Laitin (2003) 4 Fearon and Laitin (2003) 5 Fearon and Laitin (2003) 6
D S
Ex
6,092 5,893 6,051 6,051 6,092 5,89 5,893 -408.14 -479.81 -456.90 -491.67 -442.08 -445.83 -47 84.87 70.12 111.37 112.73 77.65 112.68 6 0.0908 0.0943 0.0897 0.0870 0.0957 0.0910
NOTE:Ongoingwarscoded 0 if no new warstarts.Coefficients(standard errors)arepresented.Bold indicatessignificancea GDP = gross domestic product.
847
muslim disappearswhen we analyze the entire 1945 to 1999 period. But these new results suggest that newer wars (post-1960) may have some differentcharacteristics from older wars (1945-1959) with respect to the role of religion. Peace durationis significant only in a single regression (Gleditsch et al. 2001warst5)in Table2. We get a much more mixed picturein Table4, where warin the previous period (warll) is significant and negative in about half of the regressions and nonsignificantin the other half.72 Comparingthe high- andlow-deaththresholdsin the two regressionsthatarebased on Gleditsch et al.'s (2001) data reveals importantsubstantivedifferences between lower level violence and civil war.In model 6 (all armedconflict), we find thatethnic heterogeneity,percentageMuslim, and oil exporterstatus(in Table4) are all very significant, whereas they become nonsignificant when we restrictthe analysis to civil wars. Instability,by contrast,is significantonly for civil wars and does not influence the riskof lower level violence significantly.This comparisonis particularly informative becausethe same researchteamdid the coding of both lists, and any differences in to the death threshold.Also, these two regresestimates must be attributed parameter sions highlight the differences between the two versions of the dependent variable. Using version(a), instabilityenterssignificantlyin both regressions5 (civil war)and6 (all armedconflict), and oil exports are nonsignificant.Using version (b), oil exports become significantfor regression6, and the coefficient for instabilitydropsalmost by half (see Table4). InTable6, I ranthe same regressions,restrictingthe numberof civil warlists to four so thatI could analyze the entire 1945 to 1999 periodand use the most recentand welldocumented data sets. Here, again, we observe that income and population are robustlysignificant, and anocracybecomes more significant (althoughnot in regression 1). Instabilityis significantin most lists, though marginallynonsignificant,using Fearonand Laitin's(2003) list in regressions5 and 6, where I use theircoding scheme and do not lag the first observationin each country series. Anocracy is more robust thanpreviously.Growthis now significantandnegativein threeof fourlists, butonly if we do not lag the first observation,which results in artificialstartingvalues for each countryseries. Adding growthto regressions3, 6, and 10 causes us to lose some observations, which reduces the significance of some variables(e.g., ef in regression 10). Although there is more agreementhere than in the previous tables, we can also see some importantdifferences in the results on ethnic fractionalization(now significant in two lists and borderlinesignificantin regression 1 with Gleditschet al.'s 2001 list) and with respect to oil exporterstatus, which is positive and significant in two out of fourlists (Gleditschet al. 2001; Sambanis2004). Similarly,warin the previousperiod is significantand negativeonly in two out of four lists (Fearonand Laitin2003; Doyle and Sambanis 2000). Other variables-percentage Muslim and mountainous terrain-are consistently nonsignificant.
72. It would have been interestingto explorethe effects on waronset of interactions between this variable andotherright-handside variablesbut, because thereare only a few instancesof chronologicallyoverlappingcivil wars in the same country,these interactionterms might have had the effect of overfittingthe model to a few observations.
848
The resultspresentedin Tables2, 4, and 6 suggest wide-rangingagreementon the robustnessof income and populationand cast doubt on the robustnessof most other variablesused in civil war models, especially when we consider the truncatedperiod analyzed in Tables 2 and 4. The most significant differences are with respect to the impact of oil exports, ethnic heterogeneity,and war in the previous period. Overall, there seems to be agreementthatthe mountainousterrainvariable,which is a significant variablein Collier and Hoeffler's (2001) model, is not robustto alternativemeasures of civil war.There is also no evidence of a robustassociation between civil war and the percentageof Muslims in a country.GDP growth is generally nonsignificant and may well be endogenous to levels of violence. If we restrictour analysis to the three most recent data sets-Gleditsch et al. (2001), Fearon and Laitin (2003), and Sambanis (This Study)-then we find more agreement among key variables. But when we extend the analysis back to 1945, we must also consider if there was somethose thatstartedbefore 1945 thingdifferentaboutthese earliercivil wars,particularly are at few wars left-censored civil 1945). (a The significanceof ethnic heterogeneity(ef) in some of the regressionsin Table6 is worth exploring further,given the prominenceof that variablein the civil war literature.Both FearonandLaitin(2003) andCollier and Hoeffler(2001 ) makestrongstatements againstthe significanceof ethnic heterogeneityas a factorleadingto civil war.I have shown that there is a very strong relationshipbetween ethnic heterogeneityand an aggregateindicatorof armedconflict and much less so with civil war.To the extent that violence escalates from minor to high levels, we should find ethnic fractionalizationto be significant in a dynamic model of violence escalation. But when we look at civil warsalone, why might efnot be significantin the Fearon and Laitin(2003) dataset and analysis, given some of the resultspresentedhere? Lagging right-handside variablesmakes a difference, as we saw in Table 6. Fearon and Laitinincludein theiranalysis 10 warsthatoccur on the firstyearof the countryseries. Using this data and droppingthose wars by lagging all explanatoryvariablesin their replicationdata set brings efvery close to statisticalsignificance (see online supplement).Those warsaredroppedin my dataset (except where I note thatI replace missing observationsat the startof each country series with values for the second year of the country series, as Fearon and Laitin do). For some variables (e.g., GDP), this approachto preserveobservationsmakes sense and approximateswhat any imputation program would do. However, other variables, such as instability, economic growth, and anocracy,are harderto "impute"for the startof the countryseries, espeNote also that among those droppedobservations,the mean cially for new states.73 for cases of waronset (using my civil warlist) is higher level of ethnicfractionalization thanthatfor cases of no war.Thus, restoringthose observationsshould not reduce the significance of ef in my data.Indeed, I followed Fearonand Laitin's approachto preserve those observationsand recomputedthe model in Table 6 (regressions3, 6, 10). is nonsignificantin the Fearonand Laitinmodel, but it is still Ethnicfractionalization
73. In Fearonand Laitin's(2003) replicationdata set, for example, out of a total 966 country-yearsof political instability,only 4 were years of instabilityin a newly establishedstate, which sounds implausible almost by definition.In fact, the instabilityvariable,as defined by Fearonand Laitin(a greaterthan2 change in the Polity scale) cannot be measuredfor a country'sfirst year of independence.
849
significantusing my data(it is marginallyabove the .05 level), so differencesin coding rules influencethe resultson thatvariable.Thatsaid, efseems very sensitiveto the coding rules.Notice, for example,thatby keeping 158 observationsin regression3 (by not lagging the first observation)using Gleditsch et al.'s (2001) civil war list, the coefficient for efdrops from 0.42 in regression 1 to 0.24. This suggests thatthe resultson ef are fragile. But the question also hinges on whetherwe can use this arbitrary coding scheme to preservethose observations. Another,perhapsmore substantive,reason for droppingthose few cases of war in the first year of the country series may be that several of them are left-censored. For example, the wars in Greece or the USSR all startedbefore 1945, even though some datasets list them as startingon or after 1945. If violence was ongoing beforethatdate, side variablesin 1945. Two of those casesthen it would have affectedthe right-hand Greece and South Korea (also Philippines)-are particularly important for the nonsignificance of ef in the Fearonand Laitin (2003) model because they have very low efscores. In sum, theremay be somethingdifferentaboutthe warsthatstartedduring the few years afterWorldWarII. Most were communistinsurgencieswith no clear "ethnic"dimension, and the mean value of ef for countries at war from 1945 to 1949 was abouthalf thatof countriesat war in the 1950s, 1960s, or any otherdecade up to 1999. Thus, losing those cases by lagging explanatory variables explains how ef comes close to significance using Fearonand Laitin's model and data. I now turnto an analysis of civil warprevalence.Prevalenceis defined as the union of onset andcontinuationof war,so the dependentvariableis coded 1 for all periodsof war.I assume thatcivil war is a first-order Markovprocess (i.e., thereis no significant time dependencebeyond the firstperiod)andestimatea dynamicprobitmodel by adding to the previous model a set of interactionterms between a variabledenoting the prevalenceof civil war in the previous period and all right-handside variables.The estimatedcoefficients of the interactiontermscan, aftera minoradjustment,be interbetween the independentvariablesand warconpretedas estimatesof the relationship of war in the previousperiod.The adjustment conditional on the occurrence tinuation, involves correcting the coefficient and standarderrorsof the interactionterms: for example,the coefficient of gdpll with respectto warcontinuationis the sum of the two coefficients for gdpll and wlgdp: (-0.124 + 0.293). The standarderroris the square root of the variance of gdpll plus the variance of wlgdp plus two times their covariance.The estimates in Table 7 are already adjustedin this way; thus, asterisks next to coefficientestimatesfor the interactiontermsindicatesignificancewith respect to war continuation.Estimatesof the linear terms refer to civil war onset, as before. The overallpictureis one of substantialdifferences across civil war lists. Studying prevalencebrings us a step closer to analyzing war duration,and it is not surprising, given the largedifferencesin waronset andtermination highlightedin Table 1a-d, that we should see differences in estimates of civil war prevalence. GDP is againrobustandnegativein all regressions,althoughpopulationnow is not always significantwith respectto waronset. Most variables(except GDP) are significantin only one or a few models, andinstabilityis the most often significant(in 9 out of 12 models). Growthis not significant.74
74. Using the variablegrollm (insteadof groll) again addsonly five observationsanddoes not affect the results substantially.
(text continues on p. 853)
00
Collier and Hoeffler (2001) atwar3 -0.098 (0.031) 0.418 (0.778) 0.122 (0.157) 0.194 (0.135) 0.206 (0.127) 0.130 (0.196) 0.058 (0.029) 0.006 (0.003) 0.003 (0.002) Licklider (1995) atwar4 -0.103 (0.034) 0.460 (0.733) 0.259 (0.154) 0.226 (0.150) 0.269 (0.172) 0.077 (0.242) 0.038 (0.031) 0.003 (0.002) 0.003 (0.001) Gleditsch et al. (2001) (Wars) atwar5 -0.061 (0.024) -1.187 (0.854) 0.289 (0.130) 0.218 (0.135) 0.237 (0.164) 0.257 (0.261) 0.054 (0.034) 0.003 (0.003) 0.002 (0.002)
TABLE7
Dynamic ProbitModels of Civil WarPrevalence, 1960-19

Gleditsch et al. (2001) (All) atwar6 -0.051 (0.017) -0.692 (0.557) 0.258 (0.106) 0.307 (0.092) 0.201 (0.156) 0.483 (0.175) 0.057 (0.031) 0.002 (0.002) 0.002 (0.001) Fearon and Laitin (2003) atwar7 -0.100 (0.033) -0.307 (0.829) 0.259 (0.141) 0.262 (0.137) -0.051 (0.147) -0.064 (0.206) 0.077 (0.032) 0.004 (0.003) 0.003 (0.001)
COW1994 COW2000 atwar I GDP GDP growth Instability Anocracy Oil exporter Ethnic fractionalization -0.126 (0.038) -0.425 (0.995) 0.361 (0.149) 0.217 (0.142) 0.327 (0.206) atwar2 -0.135 (0.042) -0.281 (0.899) 0.348 (0.136) 0.241 (0.124) 0.273 (0.178) 0.062 (0.211) 0.043 (0.032) 0.005 (0.003) 0.001 (0.002)
Leitenberg (2001) atwar8 -0.087 (0.031) 0.720 (0.985) 0.403 (0.155) 0.197 (0.131) 0.421 (0.187) 0.170 (0.220) 0.028 (0.031) 0.002 (0.002) 0.000 (0.001)
0.033 (0.226) Population(log) 0.053 (0.033) 0.004 Terrain (0.003) -0.001 Percentage Muslim (0.002)
War(t- 1) * Muslim War(t- 1) * GDP War(t- 1) * Growth War(t- 1) * Instability War(t- 1) *Anocracy War(t- 1) * Oil War(t- 1) * Ethnic War(t- 1) * Population War(t- 1) *Terrain
-0.005 (0.004) 0.169 (0.139) -0.021 (0.745) -0.533 (0.241) 0.077 (0.238) -1.001 (0.555) 1.561 (0.717) 0.220 (0.043) 0.000 (0.006)
-0.003 (0.004) 0.147 (0.110) -0.369 (0.615) -0.371 (0.223) 0.144 (0.215) -0.643 (0.514) 1.379 (0.587) 0.200 (0.043) 0.000 (0.005)
-0.001 (0.003) 0.033 (0.099) 0.198 (0.653) -0.266 (0.211) 0.339 (0.193) -0.508 (0.421) 0.761 (0.541) 0.253 (0.039) 0.000 (0.004)
-0.007 (0.004) 0.201 (0.111) 1.395 (0.838) -0.164 (0.217) 0.115 (0.194) -0.286 (0.383) 1.001 (0.772) 0.209 (0.045) 0.008 (0.005)
-0.002 (0.003) -0.197 (0.066) -0.451 (0.828) -0.172 (0.190) 0.336 (0.175) -0.312 (0.436) 1.233 (0.484) 0.221 (0.040) 0.005 (0.004)
0.000 (0.003) 0.029 (0.053) 1.036 (0.644) -0.152 (0.179) 0.019 (0.189) -0.283 (0.334) 0.306 (0.443) 0.244 (0.036) 0.005 (0.004)
-0.003 (0.003) 0.082 (0.068) 1.200 (1.148) -0.256 (0.191) 0.112 (0.200) -0.088 (0.354) 1.330 (0.553) 0.257 (0.041) 0.002 (0.005)
0.001 (0.003) -0.003 (0.058) 0.419 (0.592) 0.158 (0.173) 0.405 (0.154) -0.198 (0.367) 0.562 (0.370) 0.195 (0.030) 0.004 (0.004)
C0
00
TABLE7 (continued) Collier and Hoeffler (2001) atwar3 Gleditsch et al. (2001) (Wars) atwar5 Gleditsch et al. (2001) (All) atwar6 Fearon and Laitin (2003) atwar7
COW1994 COW2000 atwarI _cons atwar2
Licklider (1995) atwar4
Leitenberg (2001) atwar8
-3.095 -3.201 -2.898 -3.381 -2.842 -3.245 -3.016 -2.683 (0.550) (0.512) (0.551) (0.549) (0.486) (0.525) (0.548) (0.573) Observations 4,045 4,179 4,179 4,179 4,178 4,179 4,099 3,782 4 988.45 WaldX2 1,037.77 1,240.66 1,169.87 1,270.07 1,701.62 952.18 1,050.19 Log -687.92 -431.94 -371.13 -366.71 -417.82 -398.73 -375.13 -337.62 likelihood 0.6134 0.6510 0.7465 0.7887 0.7199 0.7015 0.6834 Pseudo-R2 0.6996
NOTE:Coefficients(standard errors)arepresented.Bold indicatessignificanceat .05 or higher;italics indicatessignificance tion terms have been adjustedto referto warcontinuationonly. GDP = gross domestic product.
Sambanis / WHAT IS CIVIL WAR? 853
The picturebecomes much more unstablewhen we considerresultswith respectto war continuationin the shaded portion of Table 7. Here we see that the interaction between income and lagged warprevalenceis often nonsignificant,butoccasionally it is significantwith switching signs: in Gleditsch et al.'s (2001) civil war list (atwar5), high income significantlyreduceswarcontinuation,whereasin the two models based on Doyle and Sambanis's(2000) data(atwarlO,atwarl 1), higher income once a civil war is ongoing has the effect of increasingthe risk of the war continuing.In Regan's (1996) and Licklider's (1995) data,the same relationshipis borderlinesignificant (at
.10).
Instabilitywhile war is ongoing significantlyreduces the risk of war continuation in two models (Singer and Small 1994; Regan 1996) andis nonsignificantin the rest.75 Similarly,anocracyis significantin one model andborderlinein anothertwo butotherwise nonsignificant.Oil exportsaregenerallynonsignificant.Thereis wide agreement thatpopulouscountrieswill have longer wars and no evidence thatcountrieswith significant Muslim populationsor mountainousterrainwill have longer civil wars. Interestingresults emerge again with respect to ethnic fractionalization(ef). The interactionterm with efis significant and positively correlatedwith war continuation in four regressions(e.g., in regression7, using Fearonand Laitin's 2003 civil war list) and close to significantin one more. Thus, the results with respect to this variableare genuinely divided: about half the models would tell us that a very diverse country would have long wars, once a waractuallystarted,whereasthe otherhalf would point to no statisticallysignificantrelationshipbetween ethnicityand warduration.There is more supportfor the positive association between ethnic fractionalizationand war continuationin Table8, where I presentprevalenceresultsfrom the fourcivil war lists thatspanthe entireperiodfrom 1945 to 1999, restrictingthe analysis to four civil war lists. Here,we see the same sortof substantialdisagreementwith respectto the effects of income on war continuation:in one model, the coefficient is significantand negative; in another,it is significantandpositive;andin the othertwo, it is nonsignificant.76 The more we push the data, the more the differences in coding rules will matter. CONCLUSION thatthereare substantial This articledoes threethings. First,it demonstrates differences across civil war lists with respect to the coding of the onset and terminationof civil war.Exploringthose differencesanalyticallyreveals some conceptualconfusion
75. I am not concerned here with the theoreticalexplanationof these results. How instability can reducewardurationis unclear.It may be thata regime transitiontowardsignificantdemocratization (which would be coded as instability)satisfies some of the rebels'demandsandleads to wartermination. The coding of the instabilityvariablemay be affected by the occurrenceof war,but it is not my purposehere to resolve these questions of endogeneity. 76. The differences actually become greaterif we do not lag the first observationsin each country series (giving 6,051 observations).Withrespectto onset, growthis significantandnegativein two out of four regressions,instabilityis significantin one regression,anocracyis significantin all regressions,oil exports are significantin one regression,and populationis significant in one regression.With respect to continuation, there is no significant change from the differences noted in Table 8 (see the results in the online supplement).
854
TABLE8
Dynamic ProbitModels of Civil WarPrevalence, 1945-1999

Gleditsch et al. (2001) (Wars) GDP GDP growth Instability Anocracy Oil exporter Ethnic fractionalization Population(log) Terrain PercentageMuslim War(t- 1) * GDP War(t- 1) ' Growth War(t- 1) * Instability War(t- 1) * Anocracy War(t- 1) ' Oil War(t- 1) * Ethnic War(t- 1) e Population War(t- 1) * Terrain War(t- 1) * Muslim _cons Observations WaldX2 Log likelihood Pseudo-R2 -0.073 (0.025) -1.006 (0.626) 0.273 (0.108) 0.172 (0.112) 0.276 (0.146) 0.220 (0.218) 0.055 (0.027) 0.003 (0.002) 0.001 (0.002) -0.153 (0.073) 0.108 (0.615) -0.108 (0.184) 0.319 (0.196) -0.103 (0.323) 1.004 (0.417) 0.211 (0.035) 0.004 (0.005) -0.001 (0.003) -3.159 (0.436) 5,893 791.89 -596.74 0.5797 Fearon and Laitin (2003) -0.095 (0.027) -0.204 (0.473) 0.243 (0.108) 0.291 (0.114) 0.067 (0.151) 0.115 (0.192) 0.078 (0.027) 0.003 (0.002) 0.002 (0.001) 0.042 (0.048) 0.288 (0.791) -0.309 (0.178) 0.054 (0.200) -0.042 (0.291) 1.052 (0.349) 0.265 (0.032) 0.003 (0.003) -0.002 (0.002) -3.501 (0.448) 5,893 1,736.07 -548.11 0.7731 Doyle and Sambanis(2000, Expanded) -0.102 (0.026) -0.187 (0.411) 0.219 (0.103) 0.274 (0.115) 0.148 (0.147) 0.295 (0.199) 0.034 (0.026) 0.001 (0.003) 0.000 (0.001) 0.111 (0.041) -0.399 (0.495) -0.169 (0.153) 0.159 (0.146) -0.228 (0.302) 0.700 (0.303) 0.197 (0.029) 0.002 (0.003) 0.000 (0.003) -2.689 (0.429) 5,893 1,898.86 -694.25 0.6963 Sambanis (This Study) -0.085 (0.025) -0.113 (0.392) 0.213 (0.109) 0.294 (0.099) 0.148 (0.130) 0.221 (0.200) 0.043 (0.024) 0.002 (0.003) 0.002 (0.001) 0.022 (0.026) 0.580 (0.661) -0.040 (0.166) 0.021 (0.177) -0.087 (0.284) 0.639 (0.353) 0.226 (0.032) 0.003 (0.003) 0.000 (0.002) -2.880 (0.412) 5,893 2,022.78 -671.34 0.7185
NOTE:Coefficients(standard errors)arepresented.Bold indicatessignificanceat .05 or higher;italics indicates significanceat .10. Parameters for all interaction termshavebeen adjustedto referto warcontinuation only. GDP = gross domestic product.
855
aboutthe meaningof civil warandcan serve as the foundationfor a theoreticaldiscussion thatestablisheswhatcivil waris andhow it can be distinguishedfrom otherforms of political violence. Second, it proposes a new coding rule for civil war thatattempts to sidestepsome of theproblemsidentifiedin othercoding rules andoffers a new list of civil wars that is based on this new coding rule. Third, it measures the substantive implications of differences in coding rules by formally comparing the empirical results thatwe get from a civil war model when we use 12 differentrules to code civil War. on civil warrevealsa remarkable The quantitative literature degreeof disagreement on how to code the onset and termination of wars, and the literatureis fuzzy on how to distinguishamongdifferentformsof political violence. This implies the need for theorizing aboutcivil warand then for propermeasurementof the concept. Given the coding complexities I have identified,researchersshould conduct robustnesstests using differentcivil war lists andjustify theircoding decisions as transparently as possible. Differences in the coding of civil warhave substantiveimplications,butperhapsnot as large as one might have expected. The results presentedhere show thatthe estimated coefficients of most variablesvary widely as a resultof changes in the coded onset of civil war.In a few cases, the sign of the coefficient also changes, and significance levels also varyacross data sets. Severalvariablesare not robustto changes in the coding of civil war onset, most strikingly in the case of lagged war prevalence, ethnic fractionalization andoil exports,and,to a much lesser extent,anocracyand instability. Predictionsof when and where civil war might occur are likely to dependcritically on which coding rule we use. More important,estimates of the substantiveeffects of to reducethe riskof civil warby manipulating the level of income, policy interventions the in or of other variables the model will also "manipulable" political instability, any for on the rule war onset and termination. vary widely, depending coding At the same time, some variables are remarkablyrobust to coding differences. Income level andpopulationsize arevery robustand significantlyassociatedwith civil war.Thatsignificance levels for these variablesdo not change much despite largedifferences in coding rules is likely because these variableschange slowly over time. A problem that this highlights is that we cannot rely on them to make accuratepredictions of the timing of war onset. Other variables-especially mountainousterrain, Muslim population, and economic growth-are consistently nonsignificant (with minor exceptions). The results from models of war prevalence suggest that predictions of civil war durationwill be even less accurate than predictions of civil war onset. There was greaterinstabilityof empiricalresultsin the prevalencemodel, so analysesof civil war durationwill be much more affected by differences in the coding rules. Robustness tests are thereforeessential when analyzing war durationor termination. Overall,civil war models seem to be good at identifying countrieswith long-term proclivitiesto civil war.But if the models arefurther developed to be able to predictthe timingof waronset better,thenwe arelikely to see coding differencesin the dependent variableaffect the parameter estimatesmoresignificantly(the changes we foundin the coefficient sign of economic growth, which changes substantiallyover time, are an indicationof this). Thus, differences in coding will begin to matterif the theoretical
856
models "catchup"to the databy including more time-sensitive variables.As the theory develops more, it seems likely thatanalystsof civil warswill have to go back to the datato develop a higher degree of consensus on the meaning of civil war before they can produceaccurateandcrediblepredictionsof where andwhen civil warswill occur. Thatsaid, I have no clear answeron whetherit is betterto have a single definitionor coding rule for civil waror if it is more beneficial to have differingdefinitions. On one of our coding rules would reflect theoreticalconsensus on hand,some standardization would ruleout one possible source whatcivil waris, andhomogeneityin measurement of difference in empiricalresults. On the other hand, that we have many coding rules suggests that the concept of civil war may mean differentthings to differentpeople, and coding rules should reflect differences in subjective understandingsof that concept. Analyzing the differences across coding rules can be instructivebecause these differences map out the space within which we might be able to find a shared of civil war. understanding A substantiveresult from this analysis, and to which I must return,is that ethnic fractionalization may not be as nonsignificanta correlateof civil waras many scholars haveargued.Its significancehinges perhapstoo muchon the coding rules for civil war, but ethnic fractionalizationis clearly importantin explaining a broad category of armedconflict thatincludesminorinsurgency.This is an important result,considering how difficultit was to clearly distinguishbetween civil warand otherforms and levels of violence. Ethnic fractionalizationwas sometimes significantly correlatedto both war onset and war continuation,but these results were at times affected by the few wars that are droppedwhen we lag explanatoryvariablesby 1 year. However,differences in coding rules seem to explain the nonsignificance of that variable in Fearon and Laitin's(2003) results.In otherruns (see supplement),religious fractionalization andthe size of the largestconfession were significantandoften dominatedthe effect of ethnic fractionalization.77 Thus, ethnoreligiousidentitymay have been writtenoff too a of as correlate large-scale armed conflict-even "civil war"-by many quickly scholars. Several other substantive results are worth noting. Economic growth may be endogenousto civil war,which implies the need for a differentestimationstrategyfor models that include that variable.The political variablesof anocracy and instability, which are central to some models of civil war, are sensitive to the coding rules but much less so if we examine the entirepost- 1945 period.The significance of the population variablemay be tied to the high thresholdof violence used to identify civil war and distinguish it from minor violence. Mountainousterrain-a key measure of the technology of insurgencyin civil war in some models-is not robustto changes in the coding rules, and neither is the measure used to identify countries that are major exportersof oil. Peace durationor war in the previousyear is also sensitive to changes in coding. Moreover,the significanceof some of these variablesis affectedby whether observationsof ongoing war are dropped(e.g., anocracy is much more consistently significant in Table4 than in Table 2). Finally, the period analyzed influences some
77. Foran analysis of the sensitivityof empiricalresultsin the civil war literatureto small changes in the specificationof the model, see Hegre and Sambanis(2004).
857
results due to the small number of war starts: for example, instability is more consistently significant in Table 6 than in Table 4. Despite these differencesanddifficulties,the conclusion from this study shouldnot be thatcoding wars and analyzingthem quantitativelyis a futile exercise. Ratherthan abandonthese efforts, I favor redoublingthem by improvingthe coding rules, applyto the data,and studyingthe implicationsof differences across ing them transparently rules. The coding legacy of the Correlatesof Warprojectis thatwe now have atourdison civil warthatwe can analyze quantitatively data to point to the theposal replicable oretical and empiricalcomplexities of defining and measuringcivil war. This article suggests ways of building on that legacy. REFERENCES
Collier,Paul,andAnke Hoeffler.2001. Greedandgrievancein civil war.Policy ResearchPaper2355, World Bank. Connor,Ken. 1998. Ghostforce: The secret history of the SAS. London:Weidenfeldand Nicolson. Doyle, Michael, and Nicholas Sambanis.2000. International peacebuilding:A theoreticaland quantitative analysis. American Political Science Review94 (4): 779-801. JournalofEconomic Growth8:195-222. Fearon,JamesD. 2003. Ethnicandculturaldiversityby country. Fearon,James D., and David D. Laitin. 2003. Ethnicity,insurgency,and civil war.AmericanPolitical Science Review97 (1): 75-90. Ghobarah,H., PaulHuth,and Bruce Russett.2003. Civil wars kill and maim people, long afterthe fighting stops. AmericanPolitical Science Review97 (2): 189-202. Gleditsch,Nils Petter,HAvard Strand,Mikael Eriksson,Margareta Sollenberg,and PeterWallensteen.2001. Armed conflict 1945-99: A new dataset. Unpublishedpaper,PRIO, Oslo, Norway. Strand.2002. Gleditsch,Nils Petter,PeterWallensteen,Mikael Eriksson,Margareta Sollenberg,and Havard Armed conflict 1946-2001: A new dataset.Journal of Peace Research39 (5): 615-37. 2003. No lessons learnedfromthe Holocaust?Assessing risksof genocide andpolitical mass Harff,Barbara. murdersince 1955. AmericanPolitical Science Review97 (1): 57-73. Hegre, Havard,TanjaEllingsen, Scott Gates, and Nils PetterGleditsch. 2001. Towarda democraticcivil peace? Democracy, political change, and civil war, 1816-1992. American Political Science Review 95:33-48. litHegre,Havard,andNicholas Sambanis.2004. Sensitivityanalysis of empiricalresultsin the quantitative eratureon civil war. Unpublishedmanuscript,PRIO and Yale University. for the ConLeitenberg,Milton.2001. Deaths in warsandconflicts between 1945 and 2000. Paperprepared ference on Data Collection in Armed Conflict, June 8-9, Uppsala, Sweden. Licklider,Roy. 1995.The consequencesof negotiatedsettlementsin civil wars, 1945-1993.AmericanPolitical Science Review89 (3): 681-90. Marshall,Monty, and Keith Jaggers. 2000. Polity IV project [Codebook and data files]. Accessed from www.bsos.umd.edu/cidcm/inscr/polity/. Mason, David, and PatrickFett. 1996. How civil wars end: A rationalchoice approach.Journalof Conflict Resolution40:546-68. and CharlesTilly. 2001. Dynamics of contention.Cambridge,UK: CamMcAdam, Doug, Sidney Tarrow, bridge UniversityPress. Price, D. L. 1975. Oman:Insurgencyand development.Conflictstudies. London:Institutefor the Studyof Conflict. Journalof ConflictResolution40 Regan,Patrick.1996. Conditionsfor successful thirdpartyinterventions. (1): 336-59. Sambanis, Nicholas. 2001. Do ethnic and nonethnic civil wars have the same causes? A theoreticaland empirical inquiry(Part 1). Journal of ConflictResolution45 (3): 259-82.
858
. 2004. Expandingeconomic models of civil war using case studies. Perspectiveson Politics 2 (2): 259-79. Sarkees,MeredithReid. 2000. The Correlatesof Wardataon war:An updateto 1997. ConflictManagement and Peace Science 18 (1): 123-44. Sarkees,MeredithReid, andJ. David Singer.2001. The Correlatesof Wardatasets:The totalityof war.Paper preparedfor the 42nd Annual Convention of the InternationalStudies Association, February20-24, Chicago. Singer,J. David, andMelvin Small. 1972. Thewages of war, 1816-1965: A statistical handbook.New York: John Wiley. andcivil wardata, 1816-1992 [Computerfile, Study . 1994. Correlatesof Warproject:International Consortiumfor Political and Social Research [distributor]. #9905]. Ann Arbor,MI: Interuniversity Small, Melvin, and J. David Singer. 1982. Resort to arms: Internationaland civil war, 1816-1980. Beverly Hills, CA: Sage. Strand, HAvard,Lars Wilhelmsen, and Nils Petter Gleditsch, in collaboration with Peter Wallensteen, MargaretaSollenberg, Mikael Eriksson, HalvardBulhaug, and Jan Ketil Rod. 2003. Armed conflict dataset codebook,version1.2a. Accessed June21, 2004, fromhttp://www.prio.no/page/Project_detail// 12cd291b/. 9244/44532.html?PHPSESSID=e92e0a3b72738fae3ff8585c Tilly, Charles. 1978. Frommodernizationto revolution.New York:RandomHouse. . 2003. Thepolitics of collective violence. Cambridge,UK: CambridgeUniversityPress. Valentino,Benjamin,Paul Huth,and Dylan Balch-Lindsay.2001. Drainingthe sea: Mass killing, genocide, and guerilla warfare.Unpublishedmanuscript,StanfordUniversity. to peace: Thesuccessful settlementof civil wars. Princeton,NJ: Prince2002. Committing Walter,Barbara. ton University Press.

What Is Civil War PDF

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

What Is Civil War PDF

Transféré par

Droits d'auteur :

Formats disponibles

What Is Civil War?

What Is Civil War?

IS CIVILWAR? Sambanis/ WHAT

RESOLUTION JOURNALOF CONFLICT

Sambanis/ WHAT IS CIVILWAR?

JOURNALOF CONFLICT RESOLUTION

Sambanis/ WHAT IS CIVILWAR?

RESOLUTION JOURNALOF CONFLICT

Sambanis/ WHAT IS CIVILWAR?

JOURNALOF CONFLICT RESOLUTION

Sambanis IS CIVIL / WHAT WAR? 823

JOURNALOF CONFLICT RESOLUTION

Sambanis/ WHAT IS CIVILWAR?

JOURNALOF CONFLICT RESOLUTION

Sambanis/ WHAT IS CIVILWAR?

RESOLUTION JOURNALOF CONFLICT

Sambanis/ WHAT IS CIVILWAR?

Note thatthiscodingruleis 31. Go to http://pantheon.yale.edu/-ns237/index/research.html#Data.

Sambanis/ WHATIS CIVILWAR?

Correlationsamong Civil WarLists, 1960-1993

1.00 0.71 0.52 0.70 0.56 0.66 0.80 0.75 0.73

1.00 0.57 0.70 0.66 0.66 0.68 0.79 0.83

1.00 0.54 0.46 0.46 0.48 0.53 0.51

1.00 0.59 0.67 0.72 0.77 0.80

1.00 0.59 0.55 0.61 0.62

Licklider Gleditsch (1995) et al. (2001) warst4b warst5b

1.00 0.62 0.46 0.61 0.49 0.68 0.80 0.73 0.68

1.00 0.40 0.58 0.50 0.52 0.60 0.68 0.69

1.00 0.49 0.37 0.40 0.41 0.45 0.44

1.00 0.56 0.58 0.61 0.63 0.69

1.00 0.51 0.46 0.50 0.51

Licklider Gleditsch (1995) et al. (2001) atwar4 atwar5

1.00 0.82 0.70 0.73 0.66 0.64 0.77 0.75 0.77

1.00 0.68 0.75 0.70 0.66 0.74 0.78 0.79

1.00 0.65 0.59 0.53 0.64 0.66 0.66

1.00 0.70 0.70 0.76 0.78 0.83

1.00 0.62 0.66 0.67 0.69

Sambanis/ WHAT IS CIVILWAR?

JOURNALOF CONFLICT RESOLUTION

IS CIVILWAR? Sambanis/ WHAT

ProbitModels of Civil WarOnset, 1960-1993

0.1 (0.0 0.0 (0.0

Percentage Muslim Peace duration

afterstart.Bold indicatessignificanceat . NOTE:Coefficients(standard errors)arepresented.Ongoingwarsdropped domestic product.

RESOLUTION JOURNALOF CONFLICT

Summaryof ParameterEstimatesfrom Table 2

foreach variable. NOTE:Meancoefficientsandstandard errors arepresented GDP = grossdomesticproduct.

ProbitModels of Civil WarOnset, 1960-1993

COW1994 COW2000 Variable Constant warst1b warst2b

Licklider (1995) warst4b

Leitenberg (2001) warst8b

NOTE:Ongoingwarscoded 0 if no new warstarts.Coefficients(standard errors)arepresented.Bold indicatessignificanc GDP = gross domestic product.

IS CIVILWAR? Sambanis/ WHAT

Estimates from Table 4 Summaryof Parameter

GDP= grossdomesticproduct. foreach variable. errors arepresented NOTE:Meancoefficientsandstandard

RESOLUTION JOURNALOF CONFLICT

ProbitModels of Civil WarOnset, 1945-1999

NOTE:Ongoingwarscoded 0 if no new warstarts.Coefficients(standard errors)arepresented.Bold indicatessignificancea GDP = gross domestic product.

Sambanis/ WHAT IS CIVILWAR?

RESOLUTION JOURNALOF CONFLICT

Sambanis/ WHAT IS CIVILWAR?

Dynamic ProbitModels of Civil WarPrevalence, 1960-19