What Is Civil War.pdf

7/29/2019 What Is Civil War.pdf

http://slidepdf.com/reader/full/what-is-civil-warpdf 1/46

What Is Civil War? Conceptual and Empirical Complexities of an Operational DefinitionAuthor(s): Nicholas SambanisReviewed work(s):Source: The Journal of Conflict Resolution, Vol. 48, No. 6 (Dec., 2004), pp. 814-858Published 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 [email protected].

Sage Publications, Inc. is collaborating with JSTOR to digitize, preserve and extend access to The Journal of

Conflict Resolution.

http://www.jstor.org

http://www.jstor.org/action/showPublisher?publisherCode=sage

http://www.jstor.org/stable/4149797?origin=JSTOR-pdf

http://www.jstor.org/page/info/about/policies/terms.jsp

http://www.jstor.org/page/info/about/policies/terms.jsp

http://www.jstor.org/stable/4149797?origin=JSTOR-pdf

http://www.jstor.org/action/showPublisher?publisherCode=sage



What Is Civil War?

CONCEPTUAL AND EMPIRICAL COMPLEXITIESOF AN OPERATIONAL DEFINITION

NICHOLASSAMBANIS

Departmentof Political Science

YaleUniversity

Theempirical iterature n civil war has seen tremendous rowthbecauseof thecompilationof quantita-tive datasets, but thereis no consensus on the measurementof civil war.This increases the risk of makinginferencesfromunstableempiricalresults. Withoutad hoc rules to code its startand end anddifferentiate tfrom otherviolence, it is difficult,if not impossible,to define and measurecivil war.A wide rangeof varia-tion in parameter stimatesmakes accuratepredictionsof waronset difficult,and differences in empiricalresults aregreaterwith respectto war continuation.

Keywords: civil war; Correlatesof War;data sets; coding rules

Advances in the empiricalliteratureon civil wardepend criticallyon the collectionand refinementof dataon civil war occurrence.Most civil war lists rely heavilyon theCorrelatesof War(COW)project.'Since the first COW list was published,there hasbeen little peer review of COWcoding rules. Withinthe walls of the COW project,there has been substantialdebate on how to improvedataquality.Thatdebate,how-

ever, has not benefited fromopen scholarlydiscourse.2Given the project's preemi-nence, many large-Nstudies use COW dataunquestioninglyor limit themselves to

makingonly minorchangesto COW

data.3Currently, bout adozen researchprojects

1. SingerandSmall(1972, 1994),Small andSinger(1982), SarkeesandSinger(2001), andSarkees

(2000).2. Personalcommunication July31, 2001) with StuartBremer, oundingmemberof theCorrelates

of War COW).COWrecentlyinitiatedan online forum for public debate,but the forumis rarelyused.3. Walter 2002) andCollierand Hoeffler(2001) use COWdatawithoutmakinganymodifications.

Mason and Fett (1996) make only minorchanges.

AUTHOR'S NOTE: Forcomments, I thank KeithDarden,Michael Doyle, James Fearon,Nils Petter

Gleditsch,HAvardHegre,StathisKalyvas,DavidLaitin,JohnMueller,MonicaToft,Ben Vallentino,Elisa-bethWood,participants n theLaboratoryn ComparativeEthnic Processes (DartmouthMeeting,October

2001), andparticipantsntheUniversityof BritishColumbia'sworkshopon "EconometricAnalysesof CivilWar." also thank AnnalisaZinn,Douglas Woodwell,KatherineGlassmyer,Ana MariaArjona,andAnnaRose ArmentiaBordon or researchassistance.All files necessarytoreplicate he analysisinthis articlecanbe foundathttp://pantheon.yale.edu/-ns237/index/research.html#Datandwww.yale.edu/unsy/jcr/jcrdata.htm.

JOURNALOF CONFLICTRESOLUTION,Vol. 48 No. 6, December 2004 814-858

DOI: 10.1177/0022002704269355

O2004SagePublications

814



Sambanis WHAT S CIVILWAR? 815

haveproducedcivil war ists based on apparentlydivergentdefinitionsof civil war,but

there s less pluralismherethanone mightthink.Mostprojectsdo notconductoriginalhistorical research and depend heavily on COW.The result may be replication of

errorsdue to the originalCOWcoding rules anduncertaintyabout whether differentdefinitionsgeneratedifferentresults.

I take a close look at theoperationaldefinitionof civil warand review severalcod-

ing rules usedin the literatureandanalyzetheirimplicationsforourunderstanding f

civil war.To make the discussionconcrete,I referto specificcases. Thisexercisegen-erates nsightsintoourabilityto measurecivil waranddistinguish t fromother forms

of politicalviolence. Drawingon thoseinsights,Iproposea newcodingrule and offer

a new civil warlist. I then measure the impactof differencesacross coding rules by

regressingthe samecivil warmodel on 12 differentversionsof waronset andpreva-

lence. Although t maybe impossible(andsome wouldargueundesirable) o arriveata consensus on a single definitionand measurementof civil war, it is important o

knowif ourunderstanding f civil war is affectedby differentcodingrules.My analy-sis shows that some of our substantive conclusions are sensitive to differences in

coding rules,whereasothers areremarkably obust.

Significantdifferences across civil war lists are mainly due to disagreementon

threequestions:What thresholdof violence distinguishescivil war from otherforms

of internalarmedconflict? How do we know when a civil war startsandends? How

can we distinguishbetween intrastate, nterstate,and extrastatewars? Answers to

thesequestionsare notonly relevant or thepurposesof accuratecoding, buttheyalsorevealthedegreeto which we share a common understanding f the concept of civil

war.I discuss these issues as a way to explorewhatdifferences acrosscoding rules can

tell us aboutthe meaningof civil war.I do notoffernew theorythathelps distinguishcivil war from other formsof violence.4Rather, want to see if availableoperationaldefinitions(coding rules)allow us to measureandanalyzecivil war as a distinct cate-

gory with a "naturalhistory"(cf. McAdam,Tarrow,andTilly 2001), an "ontology"thatis different rom thatof otherformsof politicalviolence (cf. Tilly 1978). Thus, I

implicitlyacceptherethepremisethatcivil wars are different rom otherviolence andconsider if the coding rules we have at our disposal are sufficient to make a clear

empiricaldistinctionbetweencivil warand otherviolence.

I find that t is notpossibleto arriveatanoperationaldefinitionof civil war without

adoptingsome ad hoc way of distinguishingit from other forms of armed conflict.

Althougha core set of "ideal"cases of civil warmayexist, too manycases are suffi-

ciently ambiguousto make coding the start and end of the war problematicand to

question the strictcategorizationof an event of political violence as a civil war as

opposed to an act of terrorism,a coup, genocide, organizedcrime, or international

war.5 ntheend, it maybe difficultto studycivil warwithoutconsideringhow groups4. I offer such a theoreticaldiscussion in a book-lengthmanuscript, urrently n progress.For a dis-

cussion of why we shoulddistinguishcivil warfromotherpolitical violence, see Sambanis(2004).5. Proceeding heoretically, ather hanempirically,Tilly (2003, 14)makes a similarargument, tat-

ing thatcivil wardoes not havea distinctcausallogic. It is a formof "coordinated estruction"-a typologythat ncludesvariousotherforms of politicalviolence thatgeneratesalient"short-run amage"and areper-petratedby highly coordinatedactors(the twodimensions thatdescribeTilly's typology of violence).



816 JOURNALOF CONFLICTRESOLUTION

inconflictshift from one form of violence toanother,or itmaybe profitable o analyze

politicalviolence in theaggregate,rather hancut across thatcomplex social phenom-enon witharbitrary efinitions.This article s a firststepinexploringthat deaby pro-

viding an analyticalreview of existing coding rules thathighlights the difficulty ofaccuratelydefiningandmeasuringcivil war.It also improvescurrentlyavailablecod-

ingrules at the margin,so as to producea civil war list that s more consistent with the

core elements of most workingdefinitions of civil war.

HOW WOULD WE KNOW A

CIVIL WAR IF WE SAW ONE?

In theirseminalstudyResort toArms,Small andSinger(1982, 210) defined a civilwaras "anyarmedconflict that involves (a) militaryaction internal o the metropole,(b) the activeparticipation f the nationalgovernment,and(c) effective resistancebybothsides." The main distinctionthey drew between civil (internalor intrastate)war

andinterstateor extrastate colonialandimperial)war was the internalityof the war to

the territory f a sovereignstateandtheparticipation f thegovernmentas a combat-

ant. Civil warwas furtherdistinguished rom otherforms of internalarmedconflictbythe requirement hat state violence should be sustained andreciprocatedand that the

war exceeds a certain thresholdof deaths(typically more than 1,000).

This definition is deceptively straightforward. t is, I will argue, difficult, if notimpossible, to develop an operationaldefinition of civil war withoutadoptingsome

adhoccodingrules todistinguishcivil wars from other formsof politicalviolence and

accuratelycode war onset and termination.First, it is often difficult to distinguishextrastate romintrastatewars:forexample,the Russian civil warin Chechnya n the

1990s mightbe consideredas a war of decolonization,similarto Cameroon'swar of

independence n 1954.

Second,it is unclearwhatdegree of organization s requiredof theparties odistin-

guisha civil war fromone-sided,state-sponsored iolence. In somecases, a functional

governmenthas ceasedtoexist, but we still code a civil war(e.g., Somalia after1991).In othercases, thegovernmentmaybe fightinga warby proxy using militiasin seem-

ingly disorganizedintercommunal lashes (as in Kenya's Rift Valley from 1991 to

1993),but most would notclassify such a case as a civil war.Elsewhere,rebelorgani-zationsareindistinguishable rom criminal networksor ragtagmilitias.

Third, f we focus on a numerical hresholdof deaths to identifywars,how do we

deal with theproblemof unreliablereportingandincompletedata?Evenwith reliable

data,shouldterminationbe codedonly on the basis of significantlyreducedhostilities,or should we also focus on discrete,easily coded events, such as peace treaties?

Fourth,given that violence duringcivil war is typically intermittent,how do wedeterminewhenan old warstopsandanew one starts?And how can we distinguish he

end of a civil warfrom thebeginningof aperiodof politicide,terrorism,or otherformof violence?

The COWprojecthasprovideda valuablepublic good to theprofession by codingcases of war,but it has not offered muchguidanceon how to answer these difficult




questions.Itmay also have caused some confusion. One sourceof conceptualconfu-

sion is thelackof clarityon the thresholdof deaths used to distinguishcivil warfrom

other violence. Small andSinger(1982, 213) used anannualdeath thresholdof 1,000

deathsin theirearly coding efforts,althoughthey later seem to have abandonedthatcriterion.Manyauthorsstill operateunderthe assumption hat the COWprojectuses

an annualbattle-death riterion.6This uncertainty temsfrom the factthat COW cod-

ing ruleshavechangedat leastthreetimes. Thefirstmajorchangewas theadoptionof

an annualdeathcriterion orcivil wars nResort toArms,whereasnosuchrequirementexisted 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), butit was still required or extrasystemicwars,althoughnow deaths incurred

by nonsystemmemberswere counted(Singer and Small 1994, 5).7 In a third set of

changes, Sarkees (2000, 129) and Sarkees and Singer (2001) classified someextrasystemicwars as civil wars while abandoningthe annual death criterion for

extrastatewars." tis unclearwhythedefinitionof extrasystemicwarkeptchangingor

if the new rules were used to recode all armed conflicts in empires and colonies

throughoutheperiodcoveredby COW.9Moreimportant,f dataon annualdeaths had

been collectedto code civil warsaccordingto the old (annual)criterionupto the 1994

revision,whywouldCOWresearchersnothave made use of thosedata, nsteadof los-

ing muchof that informationby usinga binaryvariabledenotingthe onset and termi-

nation of civil war?Tofully evaluatethe research hat went intocompiling the COW

lists, one needs access to COW'sraw data andcoding rules.'0COW'sefforts toconstantlyrefine its dataandimprove tscodingrules are admira-

ble. But this process raises the following question:were the new rules consistently

reapplied o historicaldata?Whenthe annualdeathcriterionwas abandoned,how did

COW coders determine he beginningand end of a civil war? Did they code the first

yearwith 1,000 deaths as theonset of the war?" Ordid theycode the startat the yearthat the cumulativedeath toll surpassedthe 1,000 mark?An example that demon-

6. See Walter 2002), Gleditschet al. (2002, 617), andLicklider(1995, 682).7. According o SingerandSmall (see codebook, 1994),interstate ystemmembership s definedin

termsof "certainminimalcriteria .. at least500,000 totalpopulationandeitherdiplomaticrecognitionbyatleasttwomajorpowersormembershipntheLeagueof Nationsor United Nations."Extrasystemicwars are

foughtbetween"anationthatqualifiesas an interstate ystemmember .. [and]a political entitythatis notan interstate ystemmember."Theyaresubdivided nto imperialand colonial wars.I referto extrasystemicand extrastatewars interchangeably, s theyboth referto colonial and imperialwars.

8. As I was workingon final revisions of this study,I received an e-mail message from ProfessorSarkees(March 15, 2004), statingthatthe 1,000 battle-death hresholdhas neverbeen abandoned n theCOWprojectandthatpooreditingof theSarkees 2000) articlegives readers hismistaken mpression.Sev-eral membersof theCOWteam hadreadearlyversionsof thisstudy,andtheyhad nevercorrectedme. More-

over,I have referred o codebooks andpublications rom COWthat counterSarkees'srecentclaim.9. A concern s that nformation ndeathsof colonial subjectsmighthave beensystematicallyworse

in empiresthaninformationon deathsof insurgents n nation-states,particularlyn the post-1945 period.

10. TheWagesof War,1816-1965 andResort to Armsincludeappendiceson included andexcludedwars,butlittle, if any, explanation s given for exclusions. Most cases were droppedbecause they did notmeet the death threshold(Small andSinger 1982, 330).

11. This may be the implicitCOWcoding rule, accordingto a communicationwith StuartBremer

(July31,2001). However,Gleditsch et al. (2001, 16)note that"anyconflict codedby COWas havingmorethan 1,000battle-deathsoverall is recordedas a civil war for all years (even yearsof inactivityand yearsbefore the cumulativedeathtoll reached1,000)"




stratespotentialerrors n theapplicationof the new rules is theAlgeriancivil war n the

1990s. Thiswar,which somedatasetscode as starting n 1992, is omittedfromSingerandSmall(1994), eventhoughfive other wars thatstarted n 1992 were included(thus

revealingthattheircoding extendedthrough1992). Perhapsthe war had not causedmorethan1,000 deathsin 1992 (itactuallyhad),butthe revised COWdataset, which

goes upto 1997(SarkeesandSinger2001), includes theAlgerianwarwith a 1992start

date.Because thecodingrules were the samein the two COWdatasets,unlessthis wasa coding error n the 1994 version,the coded startof the warin the new versionsug-

gests that he warreached he 1,000deathmarkonly after 1992,and the startof thewar

was thenback-coded to the startof the violence in 1992.

The cumulativedeathcriterionaddedfew warstothe COWlist.Fortheperiodfrom

1816to 1979, 13 wars were added o the 1994 list (107 wars were included n the 1982

list). Only 5 of these 13 wars had been included in Appendix B of Resort to Arms(Small andSinger 1982), and 3 of these 5 were excluded because of lack of system

membershipof theparticipants r because the violence was characterized s a massa-

cre. Thus, shiftingto the cumulativedeath criterionaddedonly 2 wars to the list.'2The cumulativedeath criterion introduces some problems. First, it is harder to

know when to code the startof the war.If we code it the firstyearthe killing begins,then we will not be able to studyviolence escalation because the outbreakof minor

violence will be subsumed ntheperiodof "civilwar."One researchgroupthattriesto

avoidthisproblem s Gleditschet al. (2001, 2002), who code a "war"whentheycount

morethan1,000deaths nasingle yearand do notcode thebeginningof the warduringthefirstyearwith more than25 deaths.Buttheirdatasetdoes nothave a ruleforcodingwartermination, o we do not knowpreciselyhow to code their warsin a way that is

compatiblewith otherdatasets thatuse thecumulativedeath criterion.Theproblem s

createdby the so-called "intermediate" rmed conflict category, in which thereare

morethan 100 but fewer than 1,000 deathsin a given year.If we hada conflict with,

say,600 people killed in the firstyear,then 3 yearswith virtuallyno deaths,and then

anotheryearwith 600 deaths,thismightqualifyasa civil waraccording othe cumula-

tive deathcriterionbut would be coded as two distincteventsof intermediate iolence

in the Gleditschet al. (2001, 2002) dataset.'"The annualdeath thresholdsolves this

12. See, again,Sarkees's comments notedearlier.She arguesthat COW never abandoned he 1,000annualdeath hreshold.But this raisesnewquestions.Forexample,theUppsaladatasetand therelateddatasetbyGleditschet al. (2002), whichuse anannualdeath hresholdof 1,000tocode civil war,should havefewdifferencesfromCOWlists if COW used an annual hreshold.But,as I show later, he differences betweenthese datasets arelarge.Take theexampleof Cambodia:SingerandSmall (1994) code two civil wars,onefrom 1970 to 1975 and anotherfrom 1979 to 1991. Gleditsch et al. (2002) and Strand,Wilhelmsen,andGleditsch(2003)code a firstwar n 1967,a secondwarfrom 1970 to 1975,a thirdwar n 1978,andafourth n1989(theseall havedifferent"conflictsub-IDs" n the Strand,Wilhelmsen,and Gleditsch dataset; hence,

they areconsideredas different"cases"or warstarts).13. The data set by Gleditschet al. (2001, 2002) has undergonemanyrevisions(one currentandfour

old versionscan be found online athttp://www.prio.no/cwp/armedconflictaccessedJune22, 2004]). Laterversionshaveaddressed althoughnotentirely)theproblemof coding waronset/termination y assigninga"conflictID"and"sub-ID" o each conflict andconsideringconflicts to be different f orwhen theyswitch

from an intrastate o an interstatewar(andvice versa),when the partiesand issues aredifferent,or whentherearemore han10yearswith fewer than25 deathsper year.However, t is still not clear to me if a "same"conflict thatswitchesfrom "war" o "minor" nd back to "war" hould be considereda single "war" or the

purposesof comparisonwith warlists that use a cumulativedeath criterion.




problembecause an end to the war would be coded whenever violence dropsbelow

1,000 deaths. But this creates theoppositeproblemof coding too manywar starts n

what is essentially the sameconflict, if levels of violence fluctuatewidely.

Oneway around heseproblems s to stop tryingto code andanalyzecivil war as adistinctphenomenonand, instead,to code levels of violence along a continuum(by

countryandyear).Warsmightthen be identifiedas periodswithmanyviolent spikes,in addition o othercharacteristics f the violence, dependingon the authors' heoreti-

cal argument e.g., the degreeof organizationof the parties,the presenceof two-wayviolence, etc.). The analysiswould thentryto explain politicalviolence firstand, by

refiningthe theory, explain why violence takes different forms.

A secondproblemwiththe cumulative-death riterion s that t couldsimplylead us

to code ascivil warsmanysmall conflictsthatslowly accumulatedeaths.Ineffect, this

criterioncreatesinherentlyright-censoreddata,makingit hard to code wartermina-tion. By consistently applying the cumulative deathcriterion, Sarkees and Singer(2001) couldcode as a civil waranyminorconflictor terrorist ampaign hatcauses 25

deathsper yearformorethan40 years.Fearonand Laitin(2003, 76) attempt o correct

this with a rule that 100 deathsmust occurevery year on average in an ongoing war.

But, in combinationwith the 1,000 aggregate-deaths hreshold,this creates another

logical problem.They wouldnot code as a civil wara conflict thatcaused900 deaths

over 9 years,buttheywould code a conflict thatcaused 1,000deathsover 10years.14A useful example to consider is Northeast India (Nagaland),where Fearon and

Laitin(2003) code anongoingcivil war since 1952.Iwas not able to findevidence thatmany(say,morethan100)deathsperyearoccurred n armedconflicttherefrom 1952

to 1961.Accordingto Gleditsch et al. (2002, appendix)and Small andSinger(1982,

339), therewas no war or intermediateviolence during any year of the conflict in

Nagaland.This case illustratesnot only the problemof how to code wartermination

with the cumulativethresholdbutalso the relateddifficultyof how to handle several

chronologically overlapping nsurgencies n the samecountry.Combining regionallyconcentrated nsurgenciesin India's Northeast states may be reasonable and would

probablysatisfytheaggregate-deathhreshold n theperiodconsideredby Fearonand

Laitin.But a strictapplicationof the cumulative-death ulein suchcases is problem-atic, given thatin othercountries,chronologically and even geographicallyoverlap-

ping insurgenciesareoften treatedas separateconflicts. How to distinguishbetween

these rebellions is not always easy. Burma, Chad, India,Ethiopia,and Zaire (in the

1960s) areall countries thatpose difficulties in distinguishingamong various rebel-

liousgroupsandperiodsof violence.15 Inthe absence of a clear standard f how to han-

dle suchcomplicatedcases, a rule of thumbshould be to code a "civilwar" ncountrieswith many overlapping nsurgencieswhenthe violence escalatesmarkedlyand notat

the startof low-level hostilities. In the case of India,this meansthat f we were to com-

14. Moreover, hat 100 is theaveragenumberof deathsperyearmeans thataconflict with thousandsof deaths n the firstyearcan be consideredongoing,evenif annualdeathsafter he firstyearfallto nearzero.

15. Forexample, Gleditschet al. (2002) and Strand,Wilhelmsen,and Gleditsch(2003) distinguishbetween the SerbandCroatrebellions n Bosnia,butmostotherscombinethe two insurgencies nto a single"Bosnianwar"event.At leastfive separate ebellionswereongoingfrom 1960 to 1965in Zaire(now Demo-craticRepublicof the Congo), and all data sets typicallycombine these events into a single civil war.




binetherebellions n theNortheast tates,a civil warshouldbe coded as starting n the

1980s, when violence escalatedin Assam, Tripuras,andManipur.

Despite theseproblems,a cumulative-death hresholdbecomes moredefensible if

it is combinedwith a criterion hatmilitaryactivitybe sustained orthedurationof theconflict.Both SarkeesandSinger(2001) andFearonand Laitin(2003) applythisrule,butit is unclearhow sustainedmilitaryactivityis best defined.Tofocus on anaveragenumberof deathsperyearreintroduces heproblemsof countingannualdeaths.If we

can accuratelycount 100 deathsper year, we might be able to accuratelymeasure

highernumbersof deaths andbe betteroff usingthesenumbersas ourdependentvari-

able. Wemightinsteadcountbattles,requiringat leastone battleper yearbetween the

sameparties.Butperhapscounting"battles"would createa definitionalproblemmore

severe than counting deaths, and again we would need an arbitrary"number-of-

battles" hreshold,which would excludecases of low-level insurgency nwars of attri-tion where battlesare noteasily distinguishable rom terrorist ctivity(as inthecase of

Peru's civil war).Toreachabalancebetweentheprosandcons of the absoluteversus annual hresh-

olds, we mustconsidera few issues relatedto how we measure warmagnitude:

1. What evelof violencequalifies s civil war'?2. Shouldhisbe an absolute rrelativeevel?3. Shouldweonlycountbattledeaths ralsociviliandeaths?

ARRIVING AT AN ABSOLUTE THRESHOLD

OF DEATHS TO CODE A CIVIL WAR

A characteristic f civil warthatdistinguishes tfrom otherforms of violence is that

it causes large-scaledestruction.A high thresholdof deaths can set warsapartfrom

riots, terrorism, ndsomecoups (althoughnotnecessarilyfrompogromsorgenocide).But there s nothing nherently"right" boutthe 1,000-deaths hresholdused inthe lit-

erature,andstrictapplicationof thatthresholdcan cause us to dropseveral cases that

satisfy all othercharacteristicsof civil war.A rangeof 500 to 1,000 deathscould, in

principle,be equallyconsistentwitha commonunderstanding f civil war as aneventthatcausesmajordestruction.Usingarangerather hanasinglecutoffpointmaywork

better,given the highly skewed distributionof civil war deaths. In 145 civil wars that

startedbetween 1945and 1999,the meannumberof deaths s 143,883(withastandard

deviationof 374,065), andthe median s 19,000.Despitethehigh average,11conflicts

have causedfewer than1,500 deaths,and somebarelyreached1,000deaths.But these

cases-for example,theTaiwaneserevoltin 1947, the Dar ul-Islam rebellionin 1953

in Indonesia,or thefightingin Croatiaafter ts independence n 1992-satisfied most

or all of the othercriteriafor civil wars:they were fought by well-organized groupswithpoliticalagendas,challengingthesovereignauthority,and violence was recipro-cal. Given thepoor qualityof ourdata(recallhow hard t was to count the numberof

dead aftertheTwinTowersattackon September11, 2001, in New Yorkand consider

how much hardert wouldbe to measurecivil wardeaths n Angola in 1990 or Tibetin

1951) and the skewed distributionof deaths,we should use a more flexible coding




rule,such as a rangeof deaths, nsteadof the 1,000-death hreshold.'6We couldprovi-sionallycode a waronset attheyearthatwe count 100to 500 deathsandkeeptheevent

as a civil war if coders count more than 1,000 deathsin total within 3 years of onset.

Thisincorporates hesustained ightingrulebecauseevents with low deaths n the firstyearwould be coded as warsonly if armed conflict is sustainedandcauses 1,000 or

more deathswithin a 3-yearperiod.Thevalue of thiscodingruleis that t couldeasilybe applied o back-codeeventsby integrating xistingdatabases hathave informationon lower level armedconflicts.

FOCUSING ON THE RELATIVE MAGNITUDE OF VIOLENCE

Anotherproblemwiththe absolutethreshold s that t does not reflect theconflict's

magnitudeas well as a relative(per capita)measure would. If we addeda per capitadeathsmeasureto ourcodingrules,we would be less likely to miss armedconflicts in

small nations thatproducedfew deaths but were nonethelessdramatically mportantfor thehistoryof those countries.A 1-yearconflict thatkilled 500 people (the lowestnumber n therangesuggestedabove to code waronset) in acountrywith halfa millioninhabitantsthesmallestpopulationsize allowedby mostcodingrules)would amountto deaths at the level of 0.001 of the population.A conflict of equal magnitudein a

countryof 20 million people would have caused20,000 deaths andwould have beencoded as acivil war in all data sets. Wecould use the 0.001 thresholdas a benchmark,

andwars that do not exceed the 500 to 1,000 deaths markcould still be classified ascivil warsif they meet the percapitadeaths criterion.

Anexamplethat llustrates he need forgreater lexibility inapplyingdeath thresh-

olds is the Dhofar Rebellionin Oman(1965-1976). This was an insurgencywaged byan ethnicallyorganizedrebelgroup(the DhofarLiberationFront[DLFI)againsttheSultanof Oman.The rebelsrecruitedmainlyfromtheQara,a small ethnicgroup ivingin the mountainsof the Dhofarprovince,andthegroup presented tself asCommunist-Maoist.Young potentialrecruitswere sent to school to receivepoliticalindoctrination

(Price 1975,7; Connor1998, 156-57).Therewas an element of Islamicfundamental-

ism in the group's ideology until mass defections of Islamistsoldiers after the 1970Bait Ma'ahshinimassacre Connor1998, 159).Some scattered errorism ookplace in

thepopulousnorth,butfightingwas largelyconfinedto mountainous reas.Therebelswere supportedby the People's Republicof Yemen in theircampaignto control themountains Price 1975, 3). The patternof armed conflict and the organizationof therebellionareconsistentwith a commonunderstanding f civil war,butthiscase is typi-cally excluded from civil war lists because of a low deathcount, even though, in percapitaterms, this conflict caused more damage thanmany others that are typicallyincluded as civil wars.

The 1,000-deathscriterionmaylead us to includemore cases of civil wars in largecountries,if more populouscountriescan more easily produceseveral insurgenciesthat can cause high levels of deaths. Ethnicrioting in Nigeria or staterepressionand

16. Inmydataset,I includecases thatmayfalljustshortof the 1,000markandidentifythemas poten-tially ambiguouscases.




populardissent in China is much morelikely to generate largenumbersof casualtiesthana coup in Fijior Cyprus.This maybe one reason thata variablecontrollingfor a

country'spopulation s amongthemostsignificantandrobustexplanatoryvariables n

models of civil war onset. Consider the omission from most data sets of the Greco-Turkishwar nCyprus hatstarted n 1963. Thiswarwasexcludedby SingerandSmall

(1972, 398; SmallandSinger 1982, 340) due to insufficientdeathcount. Theconflictcaused around1,000deathsandcertainlymet all othercriteria orcivil war: here werebattlesbetweenorganizedmilitaryunits,involvingthegovernmentagainstrebel mili-

tias,andpartieswithlocal recruitment nd a clearlyarticulatedpolitical agenda.Evenif only 500 deaths hadoccurred n Cyprus n thatperiod,thiswould haveamounted o0.001 of the population.A war with the same intensityin a countrywith 100 millioninhabitantswould have caused 100,000 deaths-a massive tragedythat would have

been coded in all data sets.A percapitameasurewouldcapture hesecases, but we wouldalsoneed topreservea relativelyhigh absolute thresholdnot to runthe opposite riskof selecting too manysmall conflicts in smallcountries(ordropping"relatively" mallwarsin largercoun-

tries).Creatinga percapitameasure s difficult and labor ntensive.What we mustnotdo is computepercapitadeaths nconflicts thathavealreadybeenselectedby COWorother projectson the basis of an absolute-death hreshold.Rather,coding must bebasedon primary esearch o measuredeathson a percapitabasis in allcountrieswith

any politicalviolence. A startingpoint mightbe theGleditschetal. (2001) list of minor

armedconflicts. So far,nodeathfiguresareavailablefor thoseconflicts,and we knowonly thatthey have caused morethan25 deathsper year and less than 1,000 for thedurationof theconflict. Gettingbetterdataforthose conflicts mightfacilitatethecon-structionof a percapitameasure.An alternative andless costly) approach s to code

percapitadeathsfor a shorterperiod(e.g., from 1980 onwards),allowing us to con-struct a datasamplebasedon the combinationof absoluteandpercapitadeathmea-sures (post-1980) and anothersample using only the absolute threshold.We couldthenestimate a model on the two samplesof the post-1980 periodandcomparethe

results.Wecould thendo a sort of out-of-sampletestby estimating hesamemodel on

eachsampleperiodtosee if themodel'spredictionsareconsistentacrossperiodsusingeithersample.If theresultsare not influencedby ourcoding rules,thengoing throughthe troubleof coding a percapitameasuremightnot be necessary.

THE EFFECTIVE RESISTANCE CRITERION AND

MEASURING BATTLE DEATHS VERSUS TOTAL DEATHS

The thirdcrucialquestionis whetherwe should code battledeathsor also includecivilian deathsduetothe war.Thebattle-deaths riterion s thelegacyof the COWpro-

ject andusually refers to militarydeaths (the measure was initiallyused to measuredeaths n interstatewars).SmallandSinger(1982, 213) claim thattheycounted"civil-ian aswell as militarydeaths" ncivil wars,but t is notclearif theyhave,infact,coded

civilian deathsdue to rebelattacks.Civiliandeaths mustbe countedin deathtotals.'7

17. This position is gaining ground:see Sarkees and Singer (2001, 12) and Valentino, Huth,and

Balch-Lindsay 2001).



Sambanis WHATS CIVILWAR? 823

Civilians aretargeted n civil warand aredisproportionately ffectedby humanitarian

disasterscreatedby combatants o hold civilian populationshostageandgain controlof territory."8ne mightalso considercounting refugeesandinternallydisplaced per-

sons as a measureof the humancost of the war (cf. Doyle andSambanis2000).Toget anaccuratemeasureof civiliandeaths,we mustalsodetermine f deathsdue

to politicide that occur in and aroundcivil warshould be counted. Small and Singer(1982, 215) counteddeathsdue to "actsof massacrecommitted... duringthe courseof a civil war."That seems reasonable,but a harderquestion is, how do we handle

deathsdueto acts of massacre hatflankperiodsof civil war?Should we code a war as

havingstarted f civilian massacresoccur for a period(say, 1 year)andmilitarycon-frontationensues with deaths that exceed 1,000 in the next year? What about the

reversesituation,where a civil wargives way to civilian massacres?Distinguishing

periodsof civil war fromperiodsof massacresclearlyis veryhard,as thecase of Cam-bodia illustrates.Most data sets code civil war onsets in 1970 and 1978 in Cambodiabut nowarfrom 1976to 1977,whichis aperiodthatcorresponds ocivilian massacres

(the killing fields). Small-scale battles between the Khmerand Vietnamese-backed

troopsdid takeplace near the Thaiborder n thatperiod,butit is unclear how manypeopledied as a resultof armedconflict. Hence, mostdatasets code thisas aperiodof

politicidethat s distinctfrom the civil war.Yet,if we could have established that 100or so people had been killedon theside of thestrongerpartybetween 1976 and 1978,we wouldhave coded anongoing warin thatperiod(especially in datasets using the

cumulative-deathcriterion and where effective resistance is not measured as apercentageof the total numberof deaths).

Inlightof thesedifficulties,aconservativestrategy s tocount deathsdue to massa-cresthatoccurrightbeforeorafter he waraswar-relateddeaths.The maindifficultyisto find evidence of effective resistance.According to Small and Singer (1982, 214-

15),effective resistance mplies thatthe strongerside should suffer at least 5% of thecasualties of the weakerside. FearonandLaitin(2003) relaxedthis coding rule andmeasuredeffective resistanceby 100 statedeaths.They do not, however, specify ifthese deathsmustoccurin thefirstyearof thewar orcumulatively hroughout he war.

Without data on the distribution of state deaths throughoutthe war, we are leftrudderless n trying odistinguishperiodsof "civil war" romperiodsof "politicide"or

"genocide." Couldwe say withcertainty hat50 government roopswerenotkilled inthejungles of Cambodia n 1976 andanother50 in 1977?) Some findthe distinctioneasierto makeif massacresoccur before a "civilwar"starts,as in thecase of the 1966massacres n Nigeria,which aretypically thoughtof as separate romthe Biafrancivil

war,which started n 1967.19tmightalsobeeasierto distinguishmassacres romcivilwarif themassacresoccur afteraformalend to the waris reached i.e., apeace treaty).But if the war petersout and mostly one-sided violence starts(as in the 1984-1987

18. It mighteven be desirableto count civilian deaths thatareindirectly he result of the waras, forexample, in the case of war-related tarvationdeaths in Ethiopiaand Somalia.However,this is a slipperyslope,as it is unclearhow farone could tracethehealtheffects of civil war(see Ghobarah,Huth,and Russett2003forsuchaneffort).

19. InMay-July1966,massacresof 30,000 Ibos inthe Northeast ookplace, includingmassacresdur-

ingtheNortherneaders'oupandmurderf Ironsi nd bo eaders.Biafra chieved efacto ndependencebytheendof 1966,andwar tartedn 1967.




massacres in Ndebele, Zimbabwe), the distinction is harderto make, particularlybecause warmay restartwithin 1 or moreyears as a result of such massacres.

Thus, the effective-resistancecriterioncannothelp us establish with certaintythe

difference between civil war and politicide, and we would need to use a sustainedeffective-resistance criterion to make a clear distinction. Just as Fearon and Laitin

(2003) require100 deathsperyearto code anongoing civil war,we could require10statedeathsperyeartodistinguisha civil warfromapoliticide.20 But thistypeof arith-metic seems inappropriate,not least because it trivializes andimposes a petty disci-

pline on thecomplex violence thatoccursduringcivil war.Whatif insurgentssuffer-

ing, say, 1,000deathsperyearcankill 60 soldiers orpolice officersinone year,noneinthe next year,and then5 peryear for the next 8 years (for a total of 100 in 7 years)?Wouldthisbe anymoreoranyless of a civil war thana conflictinwhich 20 soldiers are

killed per year for 5 years?The problemof counting effective resistanceeach yearmightbe avoidedif we require100 deathsper yearon average for the durationof the

conflict, but this wouldobviously requirea higherthresholdof the statedeaths crite-rion and would leadustoreclassifysome civil warsaspoliticidesorgenocides.2'Whatif we forget about the question of distributionof state deaths over time? Should anabsolute criterionof 100 statedeathsbe the single most important actorin decidingwhether oclassify aninsurgencyas acivil war f theinsurgencymeets all othercriteriafor a civil war? An example is the GamaatIslamiya (Islamic Group) and IslamicJihad's insurgencyin Egypt from 1992 to 1997. According to one source, around

1,200 people were killed over 5 or 6 years, which meets the cumulativedeath crite-rion.22 he rebels wereorganizedandhad apolitical ideology.Guerilla nsurgencywas

sustained,and attacksweredirectedagainstpolice, securityforces,governmentminis-

ters, tourists,and theCopticminority.Whetherwe code thiscase as terrorismor civilwar currentlyrests on whetherthe state suffered 100 or more deaths.This seems aweak criterionon which to base the classificationof anycase.23

Another difficult case is Argentina's "dirtywar," n which Harff (2003) codes a

politicidefrom 1976to 1980,FearonandLaitin(2003) code a civil warfrom 1973 to

1977,Singerand Small(1994) do not code acivil war,andGleditschet al. (2002) code

a warin 1975 and"possibly" n 1976 and 1977. Official statisticsfromthe Argentinemilitarycite 492 deaths (includingcivilian officials) due to "terrorist ttacks"from1969 to 1980.24Guerillassuffered deathsstarting n 1969 until 1972 (morethan 100

per year),and it is hard o know how manypeoplewere killed between 1973and 1977,theperiodsometimes coded as a "civil war."Totaldeaths, ncludingguerillasandcivil-

20. The 1,000cumulative-death riteriondividedby the 100deaths-per-yearulegives a 10-yearwin-dow for minor conflicts to be labeled civil wars. Dividing the Fearon and Laitin(2003) 100 state-deathseffective-resistanceruleby the same 10-yearwindowyields a 10 state-deaths-per-yearule.

21. Also, averagingstate deathsmightnot resolve the problem, f all ormost deathsoccurredearlyinthe conflict.

22. See my online supplement or informationon this andall cases discussed in the text.23. The Kenyan"shifta"warin the 1960s is another uchcase thatnodataset includes as a civil war,

most likely because of the difficultyof finding informationon the effective-resistancecriterion.I found

sourceslistingdozens of killingsof police andmilitarybut no clear evidence of more than100,although henumber and apparent ntensity of battles described in several sources suggest that there was effectiveresistance.

24. See a moredetailed discussion and a list of sources in the online supplement.




ians, in thatperiod seem not to have reached 1,000, particularly f we do not countdeaths afterthecoup of 1976, whichbroughta new regimeto power,and so anend tothe "war" ould be codedin 1976. The newregime engagedin a massivepurgeof sus-

pected guerillasand theirsupporters; tarting n 1976, the new regimekilled an esti-mated9,000 to 30,000 people in the periodfrom 1976 to 1983. Duringthatperiod,deaths ncurredby the state were few, so we could notcode anongoing civil war afterthe 1976 coup. In the absence of clearrules on how sustained violence andeffectiveresistancemustbecodedover time(i.e., howmanystatedeathsmust occurper year),itis unclearif and when we can code this conflict as a civil war or rathera low-level

insurgency,combinedwith a coup, and followed by a politicide.This brief discussion should reveal that most of the coding rulescurrentlyused to

measurecivil wararesomewhatad hoc and that the problemis often exacerbatedby

the low qualityof dataon deathsdueto armedconflict. Thereis largeroom for mea-surement rrorhere,and sucherrorwill make t harder o establishempirically hedif-ferencesbetween civil war and otherforms of politicalviolence. Thus,in thepresenceof these problems, one might argue thatcoding rules should never be applied too

strictlyandthat,when we are faced with anambiguouscase, we should erron the sideof caution, ncludingsuch cases while making tpossible to identifythem at theanaly-sis stage.Inthecivil war list thatIhavecompiledby applyinganew setof codingrulesthattry to address the issues raisedhere, I have includedambiguouscases but have

flagged them bothin the data set andin a supplementarydocument thatexplains the

coding for each case. This allows analyststo make theirown decisions about whichconflicts to dropor explore further o confirm thatthey areaccuratelycoded.

CLASSIFYING AND ANALYZING EXTRASYSTEMIC WARS

Another ssue thatsometimes accountsforcodingdifferencesacross datasets is the

difficultyof how to code extrastate colonial, imperial)wars. I arguethatextrastatewarsmayreasonablybe considered o be different romothercivil warsandexcluded

from civil war lists.The concept of a territorial tateis centralto the definitionof civil warbut createssomeproblems n theclassificationof the so-called extrastate orextrasystemic)wars.Theextrasystemic-wars ategory n COWincludedcolonial andimperialwars.Impe-rial wars were defined as warsbetween"systemmember[s]versus independentnon-

member[s]of theinterstate ystem" Singerand Small 1972,382). Colonialwars weredefined as wars between a "system member versus an ethnically different,

nonindependent,nonmemberof the interstate ystem"(Singerand Small 1972, 382).

Extrasystemicwarswere classified as a distinctcategorydue to a conceptualdistinc-

tionbetweenwarsthatare"peripheralo thecenterof government orthemetropole)"andthereforequalitativelydifferent rom warsthattakeplacewithin the core territoryof themetropole Sarkees2000, 126).This was a normativedefinitionbecauseit inter-

preted he natureof therelationshipbetweengovernmentsanddependentorindepend-ent regions within the state's territory.This normativeelement stands in contrasttomechanisticdefinitionsof civil war(suchas the deaththreshold).Moreover,a consid-




erationof powerrelationshipswas absent from the definitionof interstatewars,whichwere defined entirelybased on a territorial onception of the state. Over time, "the

emphasison the statememberas aterritorial ntitynecessitatedare-thinkingand ulti-

mateabandonment f themetropole[andperiphery]distinction" Sarkees2000, 127),leading the COW project to redefine its coding rules and reclassify manyextrasystemicwars as civil wars.

Identifyingextrasystemicwars s morecomplicatedas a resultof theway the COW

codingrulesevolved. Initially,codingrules forextrasystemicwarsweredifferent romrulesforintrastatewar:"whereasaninterstatewarwhichqualifiedon thebasis of bat-tle deathswouldbe included f thetotalfatalityfigurefor theprotagonistson each sidereached he 1000 mark, he member tself(including systemmemberallies, if any)hadto sustain 1000 battle fatalities in order for the extrasystemicwar to be included"

(Singerand Small 1972, 36), and"ifthewar lastedlongerthan ayear, ts battle deathshad to reachan annualaverageof 1,000"(SmallandSinger 1982, 56; see also Sarkees

2000, 129).

Eventually, hat criterionwas relaxed,andnonsystemmemberdeathswere takeninto consideration.Butcould we assumethat we wouldget anaccuratecount of colo-nized people's deaths in uprisingsagainst the metropolegiven the imperialpowers'contempt for human life in their colonies? If data measurement problems for

extrasystemicwars aresystematicallydifferent romdataproblems nother civil wars,then combining the two categories may bias analyses.25 The Mau-Mau(1952-1955)

rebellion in Kenya,forexample,was excludedfromCOWbecause it causedonly 591deathson the side of theUnitedKingdom(Singerand Small 1972,397). The war wasaddedtothe 1994and2000 updatesof COWas a resultof countingKenyandeaths,buthow manyothersuchwars failed even to appear ntheoriginal"excluded-wars"ist inthe first COW publications?One such example is the Rwandanrevolution(againstBelgium),which FearonandLaitin(2003) code as a war between 1956and 1961, butthis case does not appear n anyCOW list. A lingeringconcernwith extrastatewars,therefore, s if the originalcoding rules have hamperedour ability to compile a list

according to the currentdefinition without having to engage in painstaking and

resource-intensivehistoricalresearch.Another importantquestion is if extrasystemicwars should be coded as takingplace in themetropole(i.e., in theterritory f thesystemmember)or in theterritoryofthenonsystemmember.Givenourterritorial onceptionof the state(andof civil war),it seems clearthat thewar shouldbe coded in theterritoryof themetropole,butthis isnot establishedpracticeacrosstheliterature.Gleditschet al. (2001), forexample,code

extrasystemicwarsundercolonized states.Doyle andSambanis 2000) andLicklider

(1995) code someextrasystemicwarsas civil wars(wherethecountrywas under rust-

eeship,as inthecase of Namibia,or in thecase of the war nZimbabwe/Rhodesia rom

1972 to 1980, given Rhodesia'sindependentstatus).Other scholars have committedmore obvious errorsin coding, for example, Bangladesh'swar of independence in1971 as takingplace in Bangladeshrather hanPakistan Leitenberg2001), when it is

25. This may be an empirical question because governments everywhere have an incentive to

underreporthe numberof theirsubjectsthatthey kill.




clear thatBangladeshdidnot exist as anindependent tate until after he conclusionof

that war.In Sarkees and Singer (2001), the intention seems to be to code extrastate

wars as havingtakenplace in the territoryof the system member.

Thisbringsupanotherdifficulty.On onehand,extrastatewarsthatarecodedastak-ing place in colonized states will have to be dropped from quantitativeanalysesbecause data for the "usualsuspect" explanatoryvariables (e.g., democracy,grossdomesticproduct[GDP]) aretypicallynot availablefor dependentterritories.On the

otherhand,if the waris coded as takingplace in the territoryof the system member,thenexplanatoryvariableshave to be adjusted or the whole empire. Estimatingaver-

age GDPper capitaor the level of ethnic fractionalization orentireempireswould be

a very difficulttask and could only be done by takingshortcutsthatmight seriously

compromisethequalityof the data.Consideralso thataveragingGDPpercapitaover

entire empires will have the effect of turning highly developed countries, such asFranceorEngland, nto middle- to low-income countries.If GDPpercapita s usedas

a proxyfor statestrength, henaveragingGDP valuesfor empireswill havetheeffect

of underestimating tatecapacityin the metropoles (althoughit may accuratelycap-ture state capacity with respect to the rest of the empire). But distance from the

metropoleand militarytechnology are importantmissing variables here thatcould

explain the metropole'sreach.26

Empireswereuniquelyautocratic egimes, in which subjectslived underdifferent

forms of government,and one argument or settingextrastatewars apart s that the

legal structureof empires prevented he articulationof colonized peoples' demands(voice) and left themwith rebellion as theironly option. Might right-hand ide con-

trols forthe levelof democracyorautocracycapture hispressure o rebelincolonies?

Consider the example of France, which, accordingto Fearon and Laitin(2003), had

severalcivil warsfrom 1945 to 1960. France scores as a "deep"democracyover the

relevantperiod in the Polity IV database(Marshalland Jaggers 2000). However,a

standardnotionof democracyis that it provideschecks and balances to resolve dis-

putes peacefullyand that it affords the rightof political representationo individuals

and groups.But this notion does not apply to the peripheryof an empirewhere dis-

putes are resolved despotically, even when they are resolved by consensus in themetropole.Fearon and Laitinattemptto correct for this by using polity scores for

empiresthatareweighted by the shareof the colonized subjectsto the populationof

the metropole,therebyreducingthepolity scoreof countries,suchas Franceor Eng-land.But it is not clear to me that the combinationsof regimecharacteristics hat the

Polity IV databaseuses areeven applicableto cases of foreigndominationorempire."Wateringdown"France's, Britain's,or Belgium's democracy score in this way is

likely to make thesecountries seem like so-called "anocracies" i.e., regimes scoringin the middleof thePolity range).But one could arguethatas far as colonial subjects

were concerned,they were living in autocracies,whereas citizens in the metropolewere livingin ademocracy.Thus,consideringFranceorEnglandas anocracies s dou-

26. Moreover, he waves of decolonizationwars in the 1950s and 1960swere all related o systemicchanges thataffected entiregroupsof countries in similarways. This, at a minimum,calls for the use of

appropriate ontrols for systemic trends.




bly wrong, and the meaningof an anocracyis here inconsistentwith the theoretical

hypothesesthat scholarsusuallytest by using the anocracymeasure.27

Perhapsall theseproblemscan be addressedby adding appropriate ight-hand ide

controls to the regressionsby denoting,for example, if a countrypossessed colonies.Regionalinequality, orexample, maybe moresignificant n explaining politicalvio-

lence in a world full of empiresthan in a world composed mainlyof nation-states.28

Buteven tryingto measure nequality n theBritishempireor in theBelgian Congo is

likely to be systematicallymore difficultthanmeasuring t in theaveragenonimperialnation-state.If so, estimatesbased on the variableswith systematic coding errorfor

empiresarelikely to be biased.Thus, abandoning he distinctionbetween metropoleandperipherymayhavebeenill-advised. At itscore, theearlierCOWdistinctionhigh-

lighted thatsome peripherieswere constitutionallyexcluded from the political pro-

cess, whichpresented nsurmountable bstaclesto thepeacefulresolutionof politicaldisputeswith the metropole.

Despite these arguments, n many respects, extrastatewars are indeed similar to

civil wars:rebels aremostly locally recruited, hegroupshave a politicalagenda,and

thegoverningauthoritys involved nthefighting.Thus,analystscouldadd these wars

to their civil warlists, butdoing so would necessitateaddingappropriate ontrols to

thestatisticalmodel, as outlinedabove. Yet,only colonial wars shouldbe considered

for inclusion.29mperialwarsof annexationdo not meet theterritoriality ulebecause

the nonsystemmember'sterritorywas neverpartof the metropole'sterritorypriorto

the war. Warsin EastTimor, Tibet,or the WesternSaharacould be consideredcivilwars because(a) thefightingthatstartedduringa warof annexation ontinues afterthe

territoryis annexed by an imperial power, and/or (b) the annexation is de facto

acceptedby the international ommunity.;o

CODING CIVIL WAR

Drawingon theprecedinganalysis,I proposeanoperationaldefinitionof civil war

thatresolvessome of theproblemswe haveencountered o far.Partsof mycoding rulearenew,andpartsare basedon other,widely usedcoding rules. I providea list of civil

27. Onehypothesis(Hegreet al. 2001) is thatanocracies areproneto violent rebellion because theyareneitherautocratic noughtopreemptorcrushrebellionnordemocratic noughtoresolve conflictspeace-

fully. Many mperialpowerswere bothautocratic nough(intheircolonies) anddemocraticenough(in their

metropoles).28. A common notionof empireis that it uses the periphery o extractresourceswith no regard or

equality; hismayalso be true n some nation-states,although here s likely to be a systematicdifference n

thedegreeof regional nequalitybetweenempiresand nation-stateswiththe samelevel of democracyatthe

metropole.29. Warsof decolonizationmaybe combinedwith lists of secessionist warsbecausetheyarelikely to

sharesome of the same causallogic. However, he difficulties associatedwiththe measurement f key vari-ables in empires(see above)would still apply.

30. Suchcases in my data set include Morocco(WesternSahara), ndonesia EastTimor),andIsrael

(WestBank andGaza).Excluded s, forexample,the Malaysian Insurgency,because the level of violenceafter ndependencewas verylow,althoughdeathsduring he phaseof decolonizationwere high (see online

supplement). f there s an intervention yanother tate topreventannexationof theterritory,henthisshouldbe consideredan interstateconflict, even if local parties oin the fighting.




wars thatconformsto thecodingruleproposedhere and lateruse the newcivil warlist

in anempiricalanalysis.Extensivenotes that ustify thecoding of each case are avail-

able in a supplementposted online."3

An armedconflict shouldbe classified as a civil warif

(a)The wartakesplacewithintheterritory f a statethat s amemberof the international ys-tem32with a populationof 500,000 or greater.33

(b)Thepartiesarepoliticallyandmilitarilyorganized,andtheyhavepubliclystatedpoliticalobjectives.34

(c) Thegovernment through ts militaryor militias)must be a principalcombatant. f thereis nofunctioninggovernment, henthepartyrepresentinghegovernment nternationallyand/orclaiming the statedomesticallymustbe involved as a combatant.35

(d) The maininsurgentorganization(s)mustbe locally representedand must recruit ocally.Additionalexternal nvolvementand recruitmentneednotimply thatthe war s notintra-state.36 nsurgentgroups may operatefrom neighboringcountries,but they must alsohave someterritorial ontrol(bases)in the civil warcountryand/or herebels mustresidein the civil warcountry.37

(e) The startyear of the war is the firstyear that the conflict causes at least 500 to 1,000

deaths.38 f the conflict has not caused 500 deaths or more in the firstyear, the war is

31. Goto http://pantheon.yale.edu/-ns237/index/research.html#Data.otethat hiscodingrule smoredemanding hanmost of theothers.Sometimes,theinformation o code eachcase withcertainty s not

available,so I identifythose cases in the notes. I usuallyerr on the side of cautionandincludeambiguouscases, identifying hem nthedataset so thatresearchers an decide if theywantto includethemordrop hemfrom the analysis.

32. This includes states thatareoccupying foreign territories hat areclaiming independence e.g.,West Bank andGazain IsraelandWesternSahara n Morocco).A strictapplicationof thiscodingrulecould

drop hose cases in which the internationalommunity through heUnitedNations)rejects he state'sclaimsof sovereigntyon the occupied territories.

33. Wecouldincludecountriesafter heirpopulation eaches he500,000 markor,fromthe startof the

period, f populationexceeds the500,000 markatsome pointin thecountryseries. If a civil war occurs in a

countrywithpopulationbelow thethreshold,we could includeit andflag itas a marginal ase. Casesof civilwarclose tothe500,000 markareCyprus n 1963(578,000 population)andDjibouti n 1991(450,000 popu-lation).The percapitadeath measurewould allow us to relax the populationthreshold.

34. This shouldapplyto the majorityof theparties n the conflict. This criteriondistinguishes nsur-

gent groupsandpoliticalparties romcriminalgangsand riotousmobs.But thedistinctionbetween criminaland political violence may fade in some countries(e.g., Colombia after 1993). "Terrorist"rganizationswouldqualifyas insurgentgroupsaccording o thiscodingrule, f theycaused violence at therequiredevelsfor war(see othercriteria).Noncombatantpopulations hatare oftenvictimized in civil wars are notconsid-ereda"party"o thewar f theyarenotorganized n a militiaor othersuchform,abletoapplyviolence inpur-suit of theirpoliticalobjectives.

35. Extensive indirect support(monetary,organizational,military) by the governmentto militias

mightalso satisfythis criterion e.g., Kenyaduring he RiftValleyethnicclashes), althoughhereit becomesharder odistinguishcivil warfromcommunalviolence. Insomecases, wherethe state hascollapsed,itmaynot be possible to identify partiesrepresenting he state becauseall parties may be claimingthe state,andthese conflicts will also be hardto distinguishfrom intercommunalviolence (e.g., Somaliaafter 1991).

36. Intrastatewar can be taking place at the same time as interstatewar.37. Thisweeds outentirely nterstate onflicts withno localparticipation.The Bayof Pigs, for exam-

ple, would be excludedas a civil warbecausethe rebels did not have a base in Cubaprior o the invasion.

Somecases stretch he limits of this definitionalcriterion-for example,Rwanda nthe late 1990s,whenex-FAR(RwandanArmyForces)recruitswithbasesin the DemocraticRepublicof theCongo engaged nincur-sions and borderclashes against governmentarmyand civilians. If this is a civil war,then so is theconflict

between Lebanon-basedHezbollah and Israel(assumingthe other criteriaaremet)..38. This rulecan be relaxedto a rangeof 100 to 1,000becausefightingmightstart atein theyear(cf.

SenegalorPeru).Given the lack of high-qualitydatato accuratelycode civil waronset, if we do nothavea

goodestimateof deathsforthe firstyear,we can code the onsetatthe firstyearof reported arge-scalearmed




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. Effec-

tive resistance is measured by at least 100 deaths inflicted on the stronger party. A sub-

stantial 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 gov-

ernment wins the war, a period of peace longer than 6 months must persist before we codea new war (see also criterion k).

conflict, provided hatviolence continuesorescalates in thefollowing years.Note that n thedataset, I also

code the start/endmonth,where possible. Insome cases, my coding rulescan be used to identify the start

month(e.g., in cases where the warcauses 1,000deaths in the firstmonthof armedconflict). But in most

cases, the monthonly indicatesthestartof majorarmedconflictor thesigningof apeaceagreement,which

can give us a pointof referencefor the startor end of the war,respectively.39. This rulealso suggestswhen to code warterminationf the3-year averagedoes not addupto 500.

In such a case, we can code the end of the war at the last yearwith morethan 100deathsunless one of the

other rules applies (e.g., if there is a peace treatythatis followed by more than6 monthsof peace).40. This criterionmakescoding verydifficultbecause dataon deaths hroughouthe durationof a con-

flictarehard ofind.However, uchacodingrule s necessary opreventone fromcodingtoomanywarstartsin the same conflictorcoding anongoingcivil warforyearsafter theviolence hasended.Threeyearsis an

arbitraryutoffpointbut s consistentwith other hresholds ound inthe literature.Thedatanotes(see online

supplement)give severalexamplesof cases in which thecodingof war terminationhasbeen determinedbythis criterion.A more lenientversionwould be a 5-year thresholdwith fewer than500 deaths.

41. This criterionmust be proportionalo the war's intensityin the firstyearsof the war. If the war's

onsetis codedthefirstyearwithonly 100deaths asoftenhappens nlow-intensityconflicts),thenwe would

notbe able toobserveeffective resistance n the firstyearof thewarif we defined effectiveresistanceas 100

deathssufferedby the state.42. This criteriondistinguishescases in which insurgentviolence was limitedto the outbreakof the

warand, for the remainderof the conflict, the governmentengaged in one-sided violence. A hypothetical

exampleis a case in which insurgents nflicted 100 deathson thegovernmentduring he firstweek of fight-

ing,andthen thegovernmentdefeated heinsurgentsandengaged npogromsandpoliticideforseveralyearswith no or few deathson thegovernment's ide. If we cannotapplythis ruleconsistentlyto all cases (dueto

data imitations), henperiodsof politicideatthe startor endof the warshould be combinedwith warperi-ods. This implies that civil wars will often be observationallyequivalentto coups that are followed by

politicide or othersuch sequences of different orms of political violence.

43. Treatiesthat do not stop the fightingarenot considered(e.g., the IslamabadAccords of 1993 in

Afghanistan'swar;the December 1997agreementamongSomali clan leaders).If severalinsurgentgroupsareengagedin thewar, hemajorityof groupsmustsign.Thiscriterion s useful for thestudyof peacetransi-

tionsbutmaynotbe as importantf researchers re nterested nstudyingcivil warduration, orexample.44. Thus, in secessionist wars thatarewon by the rebels who establisha new state, if a war erupts

immediately n the new state,we would code a new war onset in the new state(anexample is Croatia rom

1992 to 1995),even if the violence is closely related o theprecedingwar.A continuationof theold conflict

between the old partiescould now count as an interstatewar,as in the case of EthiopiaandEritrea,which

foughta war between 1998 and2000 afterEritrea's uccessful secession fromEthiopia n 1993.

45. This criterionallows researcherso studythe stabilityof militaryvictories.Analysisof thestabil-

ityof civil waroutcomeswould be biasedif we codedan endtocivil warthroughmilitaryvictory only when

thevictorywasfollowedbyaprolongedperiodof peace.This wouldbias the results nfavorof findingapos-itive correlationbetweenmilitaryoutcomesandpeaceduration.Thiscriterion s importantoranalyzingwar

recurrencebut not necessarilywarprevalence.




(j)Acease-fire,ruce, rsimply nend ofighting analsomark heendof acivilwarftheyresult n at least2 yearsof peace.46 heperiodof peacemustbe longer hanwhat s

requiredn thecaseof apeaceagreementecausewedo nothaveclear ignals f thepar-ties' ntent onegotiate nagreementn the caseof a truce/cease-fire.47

(k)If newpartiesnter hewarovernew ssues,a newwaronset houldbecoded, ubjectothesameoperationalriteria.48fthesameparties eturno warover hesame ssues,we

generallyode hecontinuationf theoldwar,unlessanyof theabove riteriaorcodinga war's ndapply or theperiodbefore heresurgencef fighting.

Using thesecodingrules,I have coded 145 civil war onsets between 1945 and1999

(2.08%of 6,966 nonmissingobservations).Withoutcoding new war onsets in coun-

trieswithalreadyongoing civil wars,the numberof civil wars is 119 (1.93%of 6,153

nonmissingobservations).Outof thesecases, 20 are"ambiguous"-that is, they maynot meet one or more of the

codingrules.

DIFFERENCES IN THE CODING OF CIVIL WAR

AND THEIR SUBSTANTIVE IMPLICATIONS

Thediscussionthus farhasestablishedsome of thesources of disagreementamongcivil war ists. Disagreementsover thecodedyearof onset and terminationof civil war

may matterfor the inferencesdrawnwhen we analyze civil war onset, duration,or

recurrenceusing

differentdatasets. Theextentofdisagreement

overthecodedyear

of

civil war onset is apparent n Table la-d, which presentscorrelationsbetween war

startsduringtheperiodfrom 1960 to 1993-a periodcoveredby mostdatasets.49The

unitof analysisis thecountry-year.Thedependentvariable s civil waronset, abinaryvariable.All yearsof nowararecodedequalto 0. Therearetwo versionsof codingwar

starts: n version(a), I code a 1when a civil war startsanddropobservationsof ongo-

ingwar in thatcountryuntilthewarends.Thus,if anotherwar starts n thesamecoun-

trywhile anotherwar is ongoing, we would not considerit. In version(b), I code a I

whenevera warstarts,even if anotherwar is ongoing. Country-yearswithno new war

startsarecoded0,

andin thisway,

we endup

with morewarstarts.'5

46. Peaceimpliesnobattle-related eathsor,in a lenient versionof thiscriterion, ewer deaths han he

lowest thresholdof deaths used to code waronset, thatis, fewerthan 100 deathsperyear.47. Thesesituationsaredifferent rom hoseinwhich there s no violence as a resultof armiesstanding

down withouta cease-fireagreement,which would fall undercriterion(f).48. These incompatibilitiesmust be significantlydifferent,or the wars must be fought by different

groups n differentregionsof thecountry.Forexample,we wouldcode threepartiallyoverlappingwars in

Ethiopia Tigrean,Eritrean,Oromo)betweenthe 1970sand the 1990s.New issues aloneshouldnot be suffi-

cient to code a new warbecause thereis no "issue-based" lassificationin the definitionof civil war.We

couldapplysucha rule f we classifiedcivil wars ntocategories-for example,secessionist warsversusrev-

olutionsovercontrolof the state.In addition o havingnew issues, mostpartiesmust also be new beforewe

can code a new war onset.49. See theonlinesupplement or a summaryof the definitionsandoperational riteriausedby major

dataprojects.Theperiodwas selectedto includethe CollierandHoeffler(2001) dataset (1960-1999) in the

analysis, giventheprominenceof thatstudy n theliterature.Mostof the otherdatasets start n 1945.I com-

pareresultsfroma smallernumberof data sets coveringthe entirepost-1945 period.50. CollierandHoeffler(2001), Hegreet al. (2001), and Sambanis 2001) use version(a).Fearonand

Laitin(2003) use version(b).

(textcontinueson p. 835)



00

TABLE I

CorrelationsamongCivil WarLists, 1960-199

a. Version a) of WarOnset(3,198 Obse

Collier Fearon

and and

Hoeffler Licklider Gleditsch Laitin LeitenbCOW1994 COW2000 (2001) (1995) et al. (2001) (2003) (2000

warstlawarst2a warst3a warst4a warst5a

warst7a warst

warstla 1.00

warst2a 0.96 1.00

warst3a 0.82 0.83 1.00

warst4a 0.74 0.75 0.71 1.00

warst5a 0.42 0.46 0.52 0.57 1.00

warst7a 0.69 0.70 0.70 0.70 0.54 1.00

warst8a 0.60 0.66 0.56 0.66 0.46 0.59 1.00

warst9a 0.70 0.70 0.66 0.66 0.46 0.67 0.59

warstl0a 0.69 0.69 0.80 0.68 0.48 0.72 0.55

warstlla 0.76 0.76 0.75 0.79 0.53 0.77 0.61warstnsa 0.74 0.74 0.73 0.83 0.51 0.80 0.62



b. Version b) of WarOnset(3,503 Obs

Collier Fearon

and and

Hoeffler Licklider Gleditsch Laitin Leiten

COW1994 COW2000 (2001) (1995) et al. (2001) (2003) (2000

warstlb warst2b warst3b warst4b warst5b warst7b warst

warstlb 1.00

warst2b 0.92 1.00

warst3b 0.83 0.83 1.00

warst4b 0.57 0.55 0.62 1.00warst5b 0.37 0.45 0.46 0.40 1.00

warst7b 0.58 0.61 0.61 0.58 0.49 1.00

warst8b 0.51 0.55 0.49 0.50 0.37 0.56 1.00

warst9b 0.74 0.70 0.68 0.52 0.40 0.58 0.51

warstl0 0.69 0.67 0.80 0.60 0.41 0.61 0.46

warstllb 0.67 0.66 0.73 0.68 0.45 0.63 0.50

warstnsb 0.64 0.64 0.68 0.69 0.44 0.69 0.51

00



TABLE1 (continued)

c. CorrelationsamongListsof Civil WarPrevalence

Collier Fearonand and

Hoeffler Licklider Gleditsch Laitin LeitenbCOW1994 COW2000 (2001) (1995) et al. (2001) (2003) (2001)

atwarl atwar2 atwar3 atwar4 atwar5 atwar7 atwar

atwarl 1.00atwar2 0.94 1.00atwar3 0.88 0.90 1.00atwar4 0.75 0.76 0.82 1.00atwar5 0.62 0.69 0.70 0.68 1.00atwar7 0.67 0.70 0.73 0.75 0.65 1.00atwar8 0.63 0.67 0.66 0.70 0.59 0.70 1.00atwar9 0.62 0.66 0.64 0.66 0.53

0.70 0.62atwarl0 0.70 0.73 0.77 0.74 0.64 0.76 0.66

atwarll 0.69 0.73 0.75 0.78 0.66 0.78 0.67atwarns 0.69 0.74 0.77 0.79 0.66 0.83 0.69

d. WarOnsetsand Prevalence

Data Set Numberof WarStarts

Singerand Small (1994) 61Sarkees andSinger(2000) 73Collier and Hoeffler(2001) 70

Licklider(1995) 58Gleditsch et al. (2001) 83Fearon and Laitin(2003) 79

Regan(1996) 116

Doyle and Sambanis 2000) 100Sambanis current tudy) 102

NOTE:Years orwhichthedependentvariable s codedmissing(dueto lack of state ndependence rotherreason)civil wars in those years.Leitenberg 2000) is excluded from this table becausehis civil war list ends beforethee

0o

44.




As is evidentfrom Tablela,b, the correlationbetween mostpairsof civil war onset

lists is low.Amongthe lowestcorrelations 0.42 and0.46) are thosebetweenGleditsch

et al. (2001) and the two versionsof the Correlatesof Wardata set. These correlations

are even lower with version (b) of the dependentvariable (0.37 and 0.45, respec-tively).5'The correlationbetweenGleditsch et al. (2001) andmy data set is 0.44 [ver-sion (b)], andthecorrelationwith FearonandLaitin(2003) is 0.69 [version(b)]. The

most highly correlatedcivil war lists are the two versions of the COWdataset.52

The correlationbetween war prevalence in any two lists would be higher if one

comparedwars that startedwithin the same2- or 3-yearperiod(see Fearonand Laitin

2003 for such a comparison).But that would be a comparisonof war lists, not war

onsets.A differenceof 2 to 3 years n thecodingof waronset is significantbecauseval-

ues of right-handside variables, such as economic growth or political instability,

wouldbe immediately nfluencedby war,which wouldin turn nfluence theempiricalresults.In TableIc, I correlatecivil warprevalence combinedonset andduration) n

the different ists. If two lists includeddifferentwar startsbutthe waroverlapped or

most of its duration, henthe correlationbetween the two databaseswould be higherthan n Table 1a,b. Indeed,the correlationsbetweenpairsof lists are now higher(e.g.,between Gleditschet al. 2001 and COW2, it rises to 0.69). Thus, there is more dis-

agreementon the questionof war onset and termination han thereis over whether a

warhappenedat all.

However, there is still considerable disagreementabout which armed conflicts

shouldbe classified as civil wars.Manywars are coded inonly one out of a dozendatasets.Table1d lists thenumberof war startsandtotalcountry-yearsof warinthe differ-

ent lists from 1960 to 1993.5"The highest numberof war starts (116) is found in

Regan's (1996) list andthe lowest two in Licklider's(1995) list (58) and Singerand

Small's (1994) list (61).Now thatI have establishedthat there is substantialvariation n the coding of war

onsetandterminationn most datasets, it is appropriateo askif these differenceshave

substantive mplications.To answer thatquestionin a tractablemanner,I regressthe

samecivil war model on all versionsof the civil war onsetvariableand measurevaria-

tion in parameter stimates.Toisolatethe differences thatare due to thecoding of thedependentvariable,Iuse the same sources of datafor theright-hand ide variablesand

analyzethesame set of countriesoverthesameperiod(1960 to 1993,annualobserva-

tions).54Thus,all differences nparameter stimatesshouldbe due to differences nthe

51. HereIconsideronlywarsfromthe Gleditschet al. (2001, 2002) list. Intheregressionanalysis ater

on, I adda second measurethatincludedminor andintermediatearmedconflicts from that list.52. If we expandthe comparisonto the entireperiodfrom 1945 to 1999, the correlationsare even

smaller.My data set correlateswith others as follows: 0.62 with Sarkees and Singer (2000), 0.34 with

Gleditschet al. (2001), 0.63 with Fearon andLaitin(2003), and 0.67 withDoyle and Sambanis(2000).

53. The numberof observationsacrossmodelsdiffers nTable2 becauseIdropobservationsof ongo-ing war.Thus,any differencesaredirectlythe result of coding waronset andcontinuation.In Table4, thenumberof observations or the Collier-Hoefflervariable s slightlysmallerbecausethe laggedwar variable

is missing for the year 1960 since theirdataset starts n 1960, andwe cannotknow how they would havecodedtheprecedingyear.SingerandSmall's(1994) dataset has fewerobservationsbecauseit does not codeevents in 1993, andLeitenberg's 2001) dataset ends in 1990. Othermodels all have the same N.

54. Leitenberg's 2001) dataset is an exceptionbecause it ends with the end of the cold war.




coding rules."5 also restrictthe analysis to a smallernumberof lists thatcover theentireperiodfrom 1945 to 1999.

Themodel is very similar o thosedeveloped by FearonandLaitin(2003) andCol-

lier andHoeffler (2001). Because these studies areby now well known, I do not dis-cuss thetheorybehindthe selectionof variables.Briefly,civil war onset is expectedtobe less likely the higherthe level of development, proxied by realper capitaincome

(gdpll).56Anocracies(anoc211),states atthemid-range n thePolity IV series, shouldhaveahigherrisk of warastheyareneitheraseffectiveasautocracies nrepressionnoras good as democracies npeacefulconflictresolution(FearonandLaitin2003; Hegreet al. 2001). States withpolitical instability inst311)andregimetransition hould be at

higherrisk of war onset.57Ethnicheterogeneity(ef) should increase the risk of waronsetbypittinggroupswith differentpreferencesagainsteachother.58 ompeting he-

ories expect a differentrelationshipbetween this variable and civil war onset, andempiricalresultson the link between ethnicityand civil warare mixed. I control for

population(measuredby the natural og of population, 1popnsll), which has beenshown to be significantandpositivelycorrelatedwith war onset.59 also controlfor the

percentageof Muslims in thepopulation muslim)as a proxyof partof the "clash-of-civilization"hypothesis and as a measureof religious division." I control for eco-nomic growth, measuredas annualpercentage change in the level of realper capitaincome (groll), because Collier andHoeffler(2001) find this to be significantlyand

negativelycorrelated with war onset. I control for countries that are significantoil

exporters oil211).6'Such countriesarethought o be athigherrisk of war for a numberof reasons-the most commonly encounteredhypothesis is that oil corruptspoliticalinstitutionsor that t generates ncentives for secessionist war.I control forcountries'mountainous errain mtnll), following FearonandLaitin,who view terrainas partofthe technology of insurgency(mountainsprovidehideouts for rebels). I also controlfor a variablemeasuringtime at peace since the last war (pwt). All right-handsidevariablesare lagged 1 year.62

55. A risk s that hedegreeof multicollinearityn thedatamaybedifferentacross civil war ists, influ-

encingstandard rrorsdifferentially.But thiswould be a resultof differencesin thecodingof thedependentvariable,so it does not pose a problemfor the analysis.

56. I use data from Fearonand Laitin(2003) for this variable.Accordingto civil wartheories,higherstatecapacityshoulddiscouragerebellion orallow the state to repress t in its early stages. Higherdevelop-ment levels shoulddiscouragerebellionby raisingthe economic opportunitycosts of violence.

57. Iuse the "Polity2"seriesfrom thePolity IVdataproject,version2002, to construct heanocracyand instabilityvariables.

58. Ethnicheterogeneitywas 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 fromthe WorldBank and other sourcesto completethe populationseries. See online

supplement or more information.60. The source for this variable s Fearonand Laitin(2003).61. This series hasseveraldifferencesfrom herespectiveseries in Fearonand Laitin 2003), although

the same underlyingsources have been used (WorldBank data). I discuss all differences in the online

supplement.62. Fearonand Laitin(2003) do not lag all variables e.g., oil is not lagged)anduse the value for the

secondperiod nthecountryseriesto fill in missingvalues atthe startof thecountryseries forthose variablesthatarelagged.




This model capturesthe basic logic of prominent heoriesof civil war.63I did not

include FearonandLaitin's(2003) dummyvariable or"newstates"because this vari-

ableperfectlypredictsthe outcome when I lag all independentvariables.6Moreover,

it is notclearto mewhat thatvariablemeasures hat s notalreadycapturedby the con-trolsfor"instability,"anocracy,"nd "income.""New"statesaremorelikely to have a

civil war than "old"states,as Fearonand Laitinfound, butthey arealso more ethni-

cally diversethan "old"states.Controlling or "newstates"mayartificiallyreduce the

significance of ethnic fractionalization ef). I say "artificially"because the fact that

these statesare "new"maywell be causally linkedto theirhigh ethnicfragmentation.Most "newstates"are formercolonies withhigherethnic fractionalization han coun-

triesin whicha nation-buildingprocesshas reducedtheeffective level of ethnic frag-mentation.It might thereforebe the case that the "newness"of the stateexplains the

high degreeof ethnicfractionalization.Alternatively, ooking furtherupstream n thechainof causality,the "newness"of the statemay also be a consequenceof demands

for self-determination y ethnicallydistinctgroupsthata distantmetropolehad never

attemptedor succeeded to integrateinto a single nation in the predecessor state.65

Eitherway,if there s a causal connectionbetween new state andef, we cannot control

for both in the regression,andef is the moretheoretically nterestingconcept.We cannow check if the correlationsbetweenthesekey variablesand civil warare

influencedby thecoding of thedependentvariable.In whatfollows, I estimateprobitmodels corresponding o 12 differentcoding rulesandtwo versions of the war onset

variableandforoverallwarprevalence.I reviewtheseestimates,looking forchangesin significancelevels andverylargechanges incoefficient estimates(morethan1or2

standarddeviations), because such volatility would have importantpolicy implica-tionsinassessingthe link betweenchangesin theexplanatoryvariablesandchangesin

the probabilityof civil war onset. I presentthe resultsof these estimations in several

tables.66Table2 includesestimationresultsfor 12regressionsusing version(a) of the

dependentvariable,andTable 3 summarizesthe rangeof these parameter stimates.

Table4 includesestimationresultsfor 12regressionsusingversion(b) of thedepend-

63. This is, of course,only one of severalpossible specificationsof themodel. I do not includesomevariablesthat others have found significantbecause I find the theoretical ustificationfor includingthem

problematic r because othervariables ncluded n themodel are ikely tocapturemuchof what the excludedvariablesaremeasuring.Forexample,I do not includea controlfornoncontiguous erritorybecauseit is, byitself, nota goodcontrolof statestrengthwithoutalsocontrolling ormilitary echnologyandabilityto pro-ject militarypower.Itmayalso bethecasethatstronger tatesonlyhavethecapacity o maintainnoncontigu-ous territories.Althoughthis certainlydoes not apply to each case, an equalityof means test reveals thatcountrieswithnoncontiguous erritorieshavesignificantly higher average ncome ($5,855) than countrieswith only contiguousterritories $3,223). The t statisticfor this test is -17.78.

64. Forexample,inmydataset, thereareonly five warstarts hatoccur n the secondyearof the coun-

tryseriesand for which income is nonmissing,butall five aredroppeddue to missingvalues in the income

growthvariable.65. A

logit regressionof "new state"on "ethnic ractionalization"ef) yields a

statisticallysignificantandlargecoefficient (2.69) foref(z value= 5.52) with 316 observations I restricted he sampleto the firsttwo observations n each countryseries).

66. I do not report ixed-effects resultsbecause fixed-effects estimationis verysensitive to measure-menterror hencewe canexpectthecodingruledifferences o influencetheresults).Ipresent hese results ntheonlinesupplement.Thereare mportant ifferencesacrosswar ists,evenamongvariables hatarerobustin ordinary ogit models.

(textcontinues on p. 840)



TABLE2

ProbitModels of Civil WarOnset, 1960-199

Collier Gleditsch Gleditsch Fearon

and et al et al. and

Hoeffler Licklider (2001) (2001) Laitin LeiteCOW1994 COW2000 (2001) (1995) (Wars) (All) (2003) (20

Variable warst1 warst2 warst3 warst4 warst5 warst6 warst7 w

GDP -0.109 -0.114 -0.097 -0.075 -0.039 -0.047 -0.093 -0.

(0.032) (0.040) (0.033) (0.030) (0.020) (0.018) (0.032) (0.GDP growth -0.384 -0.237 0.493 0.564 -1.218 -0.560 -0.225 0.

(1.015) (0.898) (0.810) (0.760) (0.798) (0.542) (0.787) (1.

Instability 0.339 0.321 0.131 0.238 0.263 0.228 0.239 0.

(0.147) (0.134) (0.153) (0.154) (0.126) (0.104) (0.143) (0.

Anocracy 0.212 0.233 0.199 0.234 0.217 0.287 0.251 0.(0.141) (0.121) (0.133) (0.147) (0.130) (0.089) (0.136) (0.

Oil exporter 0.274 0.222 0.180 0.188 0.204 0.170 -0.060 0.

(0.186) (0.157) (0.119) (0.149) (0.145) (0.150) (0.142) (0.Ethnic frac-

tionalization 0.118 0.144 0.175 0.209 0.317 0.524 -0.006 0.

(0.225) (0.206) (0.200) (0.217) (0.258) (0.180) (0.210) (0.

Population

(log) 0.102 0.090 0.104 0.113 0.111 0.090 0.109 0.

(0.032) (0.030) (0.032) (0.037) (0.033) (0.032) (0.033) (0.Terrain 0.005 0.005 0.006 0.003 0.004 0.002 0.004 0.

(0.003) (0.003) (0.003) (0.002) (0.002) (0.002) (0.003) (0.

00



PercentageMuslim -0.001 0.001 0.002 0.003 0.002 0.003 0.003 0.

(0.002) (0.002) (0.001) (0.001) (0.002) (0.001) (0.001) (0.

Peaceduration -0.003 -0.004 0.000 -0.006 -0.012 0.000 -0.002 -0.

(0.005) (0.006) (0.006) (0.006) (0.005) (0.004) (0.005) (0

Constant -3.844 -3.626 -4.014 -4.147 -3.987 -3.668 -3.903 -4.

(0.563) (0.527) (0.582) (0.718) (0.587) (0.506) (0.574) (0

Observations 3,770 3,861 3,815 3,765 3,926 3,530 3,611 3,372

Loglikelihood -229.21 -263.55 -254.81 -224.91 -290.74 -435.29 -250.53 -238.

WaldX2(10) 67.38 75.69 72.15 73.59 102.21 82.62 89.31 48.

Pseudo-R2 0.0933 0.0995 0.0876 0.0944 0.0877 0.0719 0.0934 0

NOTE:Coefficients standardrrors)arepresented.Ongoingwarsdropped fterstart.Boldindicatessignificanceatdomesticproduct.

00




TABLE 3

Summaryof ParameterEstimatesfromTable2

Standard

Model 1 Variables Observations Mean Deviation Minimum Maximum

GDP coefficient 12 -0.084 0.024 -0.114 -0.039

GDP SE 12 0.029 0.006 0.018 0.040

Growthcoefficient 12 -0.031 0.550 -1.218 0.816

GrowthSE 12 0.776 0.152 0.542 1.026

Instabilitycoefficient 12 0.273 0.066 0.131 0.382

InstabilitySE 12 0.136 0.016 0.104 0.154

Anocracycoefficient 12 0.217 0.048 0.082 0.287

AnocracySE 12 0.127 0.014 0.089 0.147

Oil coefficient 12 0.177 0.104 -0.060 0.371

Oil SE 12 0.146 0.018 0.119 0.186

Ethnic fractionalization oefficient 12 0.255 0.139 -0.006 0.524

EthnicfractionalizationSE 12 0.211 0.019 0.180 0.258

Population og coefficient 12 0.101 0.012 0.076 0.116

Population og SE 12 0.033 0.002 0.030 0.037

Terrain oefficient 12 0.003 0.002 0.001 0.006

TerrainSE 12 0.003 0.000 0.002 0.003

Muslim coefficient 12 0.001 0.001 -0.001 0.003

Muslim SE 12 0.001 0.000 0.001 0.002

Peace durationcoefficient 12 -0.004 0.003 -0.012 0.000

Peace durationSE 12 0.005 0.001 0.004 0.006Constantcoefficient 12 -3.820 0.224 -4.147 -3.370

ConstantSE 12 0.582 0.053 0.506 0.718

NOTE:Meancoefficientsandstandard rrors representedoreach variable.GDP= grossdomesticproduct.

ent variable.Table4 models allow us to gauge the effect of warin thepreviousperiod(warll) on the probability of a new war onset in the same country.67Table 5

summarizes he rangeof parameter stimatesin Table4.

Oneof the most robustvariables s incomepercapita.Gdpll

isalways highly sig-nificant,althoughits coefficient as reported n Table3 varies considerably,ranging

from-0.039 to-0.114. In Table4, the resultsaresimilar,with aslightly largerrangeof

variation.Wecansee inTables3 and 5 thatthe coefficient movesby slightlymorethan2 standard eviations,dependingon thecodingof thedependentvariable.But therela-

tionshipbetween income andcivil war is extremelyrobust.

Growth(groll) is neversignificantand switches sign in about half theregressionsinTables2 and4.6 Othernonsignificantvariablesdo notswitchsigns as frequentlyas

growth, and the switching sign of growth may suggest a particularly strong

67. InFearonandLaitin's 2003) dataset, 14 warsstartwhile anotherwar s ongoingin the samecoun-tryfrom 1945 to 1999.Inmydataset, thatnumber s 29 and 17 for theperiodfrom 1960to 1993.The warllvariablecapturessome of thetimedependenceof peaceand warso it replacesthepeacetimevariable pwt)thatI use in Table2 models.

68. Using a 3-yeargrowthratedid notchangethis. By lagging growth,we lose the first two observa-tionsin thecountryseries because we mustlaggdpen by I yearto calculate thegrowthrate gdpgro),whichcannotbe calculated or the firstyearof the series.Icoded one versionof thegrowthvariable,grollm, where

(textcontinues on p. 843)



TABLE4



and et al. et al. and

Hoeffler Licklider (2001) (2001) Laitin LeitenbergCOW1994 COW2000 (2001) (1995) (Wars) (All) (2003) (2001)

Variable warstIb warst2b warst3b warst4b warst5b warst6b warst7b warst8b

GDP -0.126 -0.134 -0.092 -0.077 -0.056 -0.054 -0.098 -0.088

(0.035) (0.039) (0.029) (0.028) (0.023) (0.016) (0.029) (0.030)

GDP growth -0.295 -0.072 0.195 0.584 -1.087 -0.895 -0.351 0.431

(0.914) (0.791) (0.716) (0.738) (0.804) (0.560) (0.774) (0.896)

Instability 0.316 0.348 0.130 0.217 0.284 0.134 0.167 0.393

(0.161) (0.145) (0.145) (0.152) (0.119) (0.094) (0.131) (0.143)

Anocracy 0.314 0.289 0.226 0.215 0.204 0.247 0.239 0.170(0.139) (0.127) (0.121) (0.143) (0.132) (0.082) (0.124) (0.123)

Oil exporter 0.358 0.291 0.194 0.222 0.240 0.302 0.111 0.373

(0.180) (0.155) (0.130) (0.156) (0.141) (0.116) (0.140) (0.153)

Ethnicfrac-

tionalization 0.095 0.157 0.295 0.260 0.391 0.591 0.084 0.075

(0.212) (0.199) (0.203) (0.222) (0.277) (0.169) (0.188) (0.228)

Population log) 0.090 0.083 0.094 0.129 0.143 0.133 0.131 0.101

(0.031) (0.029) (0.032) (0.045) (0.035) (0.033) (0.031) (0.028)

Terrain 0.005 0.006 0.006 0.002 0.003 0.002 0.004 0.001

(0.003) (0.003) (0.003) (0.002) (0.002) (0.002) (0.002) (0.003)

Percentage -0.001 0.000 0.003 0.003 0.002 0.002 0.003 -0.001

Muslim (0.002) (0.002) (0.001) (0.001) (0.002) (0.001) (0.001) (0.001)

Warat (t - 1) -0.351 -0.351 -0.447 -0.677 -0.005 -0.103 -0.274 -0.371

(0.192) (0.176) (0.200) (0.236) (0.169) (0.107) (0.124) (0.215)

40



TABLE4 (continued)



Hoeffler Licklider (2001) (2001) Laitin Leitenberg

COW1994

COW2000

(2001)(1995) (Wars) (All) (2003)

(2001)Variable warst1b warst2b warst3b warst4b warst5b warst6b warst7b warst8b

Constant -3.669 -3.575 -3.942 -4.520 -4.733 -4.353 -4.289 -3.747

(0.554) (0.516) (0.581) (0.787) (0.605) (0.530) (0.545) (0.530)

Observations 4,045 4,179 4,099 4,178 4,179 4,179 4,179 3,782

Loglikelihood -248.10 -289.52 -266.58 -237.15 -334.42 -561.33 -303.22 -270.64

WaldX2(10) 63.13 75.12 79.28 74.27 102.19 101.28 75.07 40.53

Pseudo-R2 0.0936 0.1026 0.0863 0.0952 0.0908 0.0844 0.0838 0.0715

NOTE:Ongoingwarscoded 0 if no new war starts.Coefficients(standard rrors)arepresented.Bold indicatessignificanGDP = grossdomesticproduct.

ti




TABLE 5

Summaryof ParameterEstimatesfromTable4

Standard

Model2 Variables Observations Mean Deviation Minimum Maximum

GDP coefficient 12 -0.093 0.024 -0.134 -0.054

GDP SE 12 0.028 0.006 0.016 0.039

Growthcoefficient 12 -0.199 0.489 -1.087 0.584

GrowthSE 12 0.705 0.137 0.529 0.914

Instabilitycoefficient 12 0.249 0.085 0.130 0.393

InstabilitySE 12 0.131 0.020 0.094 0.161

Anocracycoefficient 12 0.243 0.043 0.170 0.314

AnocracySE 12 0.121 0.016 0.082 0.143

Oil coefficient 12 0.238 0.0800.111 0.373

Oil SE 12 0.140 0.021 0.106 0.180

Ethnic fractionalization oefficient 12 0.280 0.155 0.075 0.591

EthnicfractionalizationSE 12 0.206 0.027 0.169 0.277

Population og coefficient 12 0.104 0.026 0.050 0.143

Population og SE 12 0.033 0.005 0.028 0.045

Terrain oefficient 12 0.003 0.002 0.001 0.006

TerrainSE 12 0.002 0.000 0.002 0.003

Muslimcoefficient 12 0.001 0.001 -0.001 0.003

Muslim SE 12 0.001 0.000 0.001 0.002

Laggedwarcoefficient 12 -0.286 0.212 -0.677 0.101

Laggedwar SE 12 0.164 0.041 0.107 0.236Constantcoefficient 12 -3.915 0.493 -4.733 -2.911

ConstantSE 12 0.587 0.081 0.516 0.787

NOTE:Meancoefficientsandstandardrrors representedor each variable.GDP= grossdomesticproduct.

endogeneityproblemwith this variable.Growth s very sensitive to civil war,so get-

tingthedateof onsetof civil war"wrong" n some datasets should influencethe coef-

ficient sign of this variable.69A possible explanationfor the instabilityof growthis

that, n theimmediatepostwarperiod,growths

typicallyveryhigh,butthese are also

periodsof high risk of war recurrencedue to other influences.Thus, some periodsof

war recurrencemightalso be high-growthperiods,whereas the generaltrend should

be for growthto reducethe risk of civil war.

Political nstability s mostly significantandalways positive(see Table2), althoughintworegressions CollierandHoeffler2001; Licklider1995)it is nonsignificant,and

in another wo cases, it is significantonly at the .10 level. In Table4, it appears o be

laggingis delayedby one observation, o this restores he second lost observation n eachcountryseries. In

Tables 2 and4, this addsonly five observations, ncludingone war(Rwanda n 1963). Using this measureinsteadof groll makesanocracyslightlymoresignificantand terraina little moresignificant n two regres-sions in Table2 andefa littleless significant n Table4 (Rwandahasa low level of ef,at0.18). Theresultson

growthdo not change.69. Given my concern with endogeneity here, I droppedthe growth variableand reestimated all

regressions.See versions of Tables2, 3, 4, and 5 withoutgrowthin the online supplement.The results are

very similarto the ones presentedhere.




less robust,being significant n only half of the regressionsandnonsignificant n the

otherhalf.

Anocracybordersstatisticalsignificancein six regressions nTable2 andis clearly

significantonly in Gleditsch et al.'s (2001) list, which combines all armedconflicts.Similarly, tis significant nonly six outof the twelve regressions nTable4. Its coeffi-

cientrangesfroma low of 0.170 (Leitenberg2001 list) to a high of 0.314 (Singerand

Small 1994 list) in Table4, andthe rangewidens in Table2 froma low of 0.082 to a

high of 0.287.

Oil exports are basically nonsignificantin Table2, althoughthe picture is more

mixedin Table4, whereit is significant n fourregressionsand borderlinesignificantinanother hree.70Its coefficientvarieswildly from -0.060 to0.371 (Table3; althoughit is less unstable nTable5). This variable s not robust ochangesin thecoding of war

onset.Ethnicheterogeneity(ef) is positiveand almostalways nonsignificantwith a largestandard rror,except for regression6 (Gleditschet al. 2001--all armedconflict) in

Table2. Efis nearly significant(at the .10 level) in Regan's (1996) model (warst9),where again a numberof additional wars 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), althoughits coefficient has a wide

range,from0.075 to 0.591 (see Table5). Thatefis generally nonsignificant n Table2

andmuch moresignificant n Table4 mayhavesomethingto do withthoseextrawars

thatwe are able to include when we use version(b) of waronset.7'The coefficient of population(lpopnsll) is quite stable and always positive and

highlysignificant,except for model9 in Table4, where it is notsignificant.This inter-

esting resultseems tojustify my earlierclaim thatthe significanceof populationsize

maybe an artifactof thehigh absolutethresholdof deathsimposedto classify a civil

war. In model 9, Regan's lower death thresholdcapturesmore small wars in small

countries.

InbothTables2 and4, mountainous errain mtnll) is relativelystable andis only

significant n the Correlatesof WarandCollier andHoeffler(2001) data sets. Moun-

tainousterrain s a key variable n theoriesof civil warthatemphasizetheopportunitystructureorrebellion(e.g. FearonandLaitin2003; Collier and Hoeffler2001), but it

is generallynot statistically significant.The percentof Muslims in a country (muslim) is marginallysignificant in two

regressionsand in another wo at the .10 level (see Table2) and becomes somewhat

more significant in Table 4. There, we find it significant in the Fearon and Laitin

(2003) model(warst7b)with a positivesign, despitethe findingsof these authors hat

religiousdivision is nonsignificant n models of civil war onset. However,the period

analyzedseems to matter,as later(see Table6) we will find thatthe significance of

70. Iuse thevariableoil2, laggedonce (oil211).If I wereto use Fearonand Laitin's(2003) oilll series

(laggingitonce), it would be significant nmodel6 and,atthe . 10level, inmodels 8 and9 andnonsignificantin all othermodels, includingsome of those in which oil211 was significant(COW 1994 andmy data).

71. Countrieswith more than one warat the sametime have slightly higherethnic fractionalization.Thereare26 such cases (of chronologicallyoverlappingwars in my dataset), anda means test for ethnicfractionalization hows a statisticallysignificant I10-point ifference between themandthe othercases.



TABLE6


Gleditsch Gleditsch Gleditsch Fearon Fearon Fearon

et al. et al. et al. and and and

(2001) (2001) (2001) Laitin Laitin Laitin(Wars) (Wars) (Wars) (2003) (2003) (2003) E

Variable 1 2 3 4 5 6

GDP -0.065 -0.079 -0.078 -0.095 -0.104 -0.099

(0.022) (0.023) (0.023) (0.025) (0.025) (0.024)

GDP growth -0.962 - -1.350 -0.396 - -1.443

(0.581) - (0.539) (0.537) (0.617)

Instability 0.305 0.279 0.278 0.222 0.176 0.186

(0.095) (0.088) (0.089) (0.101) (0.096) (0.097)

Anocracy 0.157 0.244 0.214 0.265 0.332 0.296(0.108) (0.090) (0.095) (0.101) (0.093) (0.096)

Oil exporter 0.294 0.310 0.301 0.145 0.192 0.173

(0.127) (0.128) (0.135) (0.128) (0.119) (0.124)

Ethnicfractionalization 0.417 0.268 0.236 0.254 0.209 0.245

(0.217) (0.211) (0.208) (0.175) (0.166) (0.164)

Population log) 0.140 0.137 0.144 0.142 0.141 0.148

(0.028) (0.025) (0.027) (0.029) (0.029) (0.030)

Terrain 0.003 0.002 0.002 0.003 0.003 0.003

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)

PercentageMuslim 0.000 - - 0.002 - -

(0.002) - - (0.001) - -

Warat (t - 1) 0.010 0.016 -0.007 -0.309 -0.342 -0.364

(0.150) (0.143) (0.145) (0.097) (0.104) (0.102)

Constant -4.670 -4.500 -4.571 -4.581 -4.472 -4.568

(0.487) (0.447) (0.474) (0.504) (0.489) (0.514)

~tl



TABLE6 (continued)

Gleditsch Gleditsch Gleditsch Fearon Fearon Fearon

et al. et al. et al. and and and S

(2001) (2001) (2001) Laitin Laitin Laitin

(Wars) (Wars) (Wars) (2003) (2003) (2003) E

Variable 1 2 3 4 5 6

Observations 5,893 6,092 6,051 5,893 6,092 6,051 5,8

Log likelihood -445.83 -491.67 -479.81 -408.14 -456.90 -442.08 -47

WaldX2(d/) 112.68 112.73 111.37 84.87 70.12 77.65

Pseudo-R2 0.0910 0.0870 0.0943 0.0908 0.0897 0.0957

NOTE:Ongoingwarscoded0 if no newwarstarts.Coefficients standardrrors)arepresented.BoldindicatessignificanceGDP = gross domesticproduct.

o00




muslimdisappearswhen we analyze the entire 1945 to 1999 period. But these new

resultssuggest thatnewerwars (post-1960) may have some differentcharacteristics

fromolder wars (1945-1959) with respectto the role of religion.

Peace duration s significantonly in a single regression(Gleditschet al. 2001-warst5) nTable2. Weget a much more mixedpicture n Table4, where war nthepre-vious period (warll) is significantand negativein abouthalf of the regressionsand

nonsignificant n the otherhalf.72

Comparing hehigh-and ow-deaththresholds nthe tworegressions hatarebased

on Gleditschet al.'s (2001) data reveals importantsubstantivedifferences between

lowerlevel violence andcivil war.Inmodel 6 (all armedconflict), we findthatethnic

heterogeneity,percentageMuslim,and oil exporterstatus(in Table4) are all very sig-nificant,whereas they become nonsignificantwhen we restrict the analysis to civil

wars.Instability,by contrast, s significant only for civil warsanddoes not influencetherisk of lowerlevel violence significantly.Thiscomparison s particularlynforma-

tive because thesameresearch eam did thecodingof bothlists,andanydifferencesin

parameter stimates mustbe attributedo the death threshold.Also, these two regres-sions highlight the differences between the two versions of the dependentvariable.

Using version(a), instabilityenterssignificantly nbothregressions5 (civil war)and6

(all armedconflict), and oil exportsarenonsignificant.Using version(b), oil exportsbecome significantforregression6, and the coefficient for instabilitydropsalmostbyhalf (see Table4).

InTable6, Iran hesameregressions,restricting he numberof civil war iststo fourso thatIcouldanalyzetheentire 1945 to 1999periodand use the most recentand well-

documented data sets. Here, again, we observe that income and population are

robustlysignificant,andanocracybecomes more significant (althoughnot in regres-sion 1).Instability s significant n most lists, though marginallynonsignificant,usingFearonandLaitin's(2003) list in regressions5 and6, whereI use theircoding scheme

anddo not lag the firstobservation n each countryseries. Anocracy is more robustthanpreviously.Growth s nowsignificantandnegative nthreeof fourlists,butonly if

we do not lag the firstobservation,which results in artificialstartingvalues for each

countryseries.Adding growth oregressions3, 6, and 10 causesus to lose someobser-vations,which reduces the significanceof some variables(e.g., ef in regression 10).

Althoughthere is more agreementhere thanin the previoustables, we can also see

some importantdifferencesin the results on ethnicfractionalization now significantin two lists andborderlinesignificant n regression 1 with Gleditsch et al.'s 2001 list)andwith respectto oil exporterstatus,which is positive andsignificantin two out of

fourlists (Gleditschet al. 2001; Sambanis2004). Similarly,war nthepreviousperiodis significantandnegativeonly in two out of four lists (FearonandLaitin2003; Doyleand Sambanis 2000). Other variables-percentage Muslim and mountainous

terrain-are consistentlynonsignificant.

72. It would havebeeninteresting oexplorethe effects on waronset of interactionsbetweenthis vari-ableand otherright-hand ide variablesbut,becausethereareonly a few instancesof chronologicallyover-

lappingcivil wars in the samecountry, hese interaction ermsmighthavehad the effect of overfitting hemodel to a few observations.




Theresultspresented n Tables2, 4, and6 suggest wide-rangingagreementon the

robustnessof income andpopulationand cast doubton the robustnessof most other

variablesused in civil warmodels, especially when we consider the truncatedperiod

analyzedin Tables2 and 4. The most significantdifferences are with respect to theimpactof oil exports,ethnic heterogeneity,and warin the previousperiod. Overall,thereseems to be agreement hat the mountainous errainvariable,which is a signifi-cant variable n Collier andHoeffler's(2001) model, is not robust o alternativemea-

suresof civil war.Thereis also no evidence of a robustassociationbetween civil war

andthe percentageof Muslims in a country.GDP growthis generally nonsignificantand may well be endogenous to levels of violence. If we restrictour analysis to the

three most recent datasets-Gleditsch et al. (2001), Fearon and Laitin(2003), and

Sambanis (This Study)-then we find more agreementamong key variables. But

when we extendthe analysisback to 1945, we mustalso considerif therewas some-thingdifferentabouttheseearliercivil wars,particularlyhose thatstartedbefore 1945

(a few civil warsare left-censoredat 1945).Thesignificanceof ethnicheterogeneity ef) insomeof theregressions nTable6 is

worthexploringfurther,given the prominenceof thatvariable n the civil war litera-

ture.BothFearonandLaitin(2003) andCollier and Hoeffler(2001) makestrongstate-

mentsagainst hesignificanceof ethnicheterogeneityas a factor eadingto civil war.I

have shown that thereis a very strong relationshipbetweenethnicheterogeneityand

anaggregate ndicatorof armedconflict andmuchless so with civil war.Tothe extent

that violence escalates from minor to high levels, we should find ethnicfractionalization o be significantin a dynamic model of violence escalation.

But whenwe look atcivil warsalone, why mightefnot be significant n the Fearon

andLaitin 2003) dataset andanalysis, given some of the resultspresentedhere?Lag-

ging right-hand ide variablesmakesa difference,as we saw in Table 6. Fearonand

Laitin ncludein theiranalysis 10warsthatoccur on the firstyearof thecountryseries.

Using thisdata anddroppingthose warsby lagging all explanatoryvariables n their

replicationdataset bringsefvery close to statisticalsignificance (see online supple-ment).Thosewarsaredropped n mydataset (exceptwhereI notethatIreplacemiss-

ing observationsat the startof each countryseries with values for the second yearofthe country series, as Fearon and Laitin do). For some variables (e.g., GDP), this

approach o preserveobservationsmakes sense andapproximateswhat any imputa-tion programwould do. However, other variables, such as instability,economic

growth,andanocracy,areharder o "impute" or the startof thecountryseries, espe-

cially for new states.73Note also thatamong those droppedobservations,the mean

levelof ethnic fractionalizationorcases of waronset(usingmycivil war ist) is higherthanthatfor cases of no war.Thus,restoring hose observationsshouldnotreduce the

significanceof ef in my data.Indeed,I followed Fearonand Laitin'sapproach o pre-

serve those observationsandrecomputed he model in Table6 (regressions3, 6, 10).Ethnic fractionalizations nonsignificant n the Fearonand Laitinmodel,butit is still

73. In Fearonand Laitin's(2003) replicationdataset, forexample,outof a total 966 country-yearsof

political instability,only 4 were yearsof instability n a newly establishedstate,which soundsimplausiblealmostbydefinition.Infact,theinstabilityvariable,asdefinedby Fearonand Laitin agreater han2 changein the Polity scale) cannot be measured or a country's irstyearof independence.




significantusing my data(itis marginallyabove the .05 level), so differences ncodingrules influencethe resultson thatvariable.Thatsaid,efseems verysensitive tothe cod-

ing rules.Notice, forexample,thatbykeeping 158observations nregression3 (bynot

lagging the firstobservation)usingGleditsch et al.'s (2001) civil warlist, the coeffi-cient forefdrops from0.42 in regression1 to 0.24. This suggests thattheresultson efarefragile.But the questionalso hinges on whetherwe can use this arbitrary odingscheme to preservethose observations.

Another,perhapsmoresubstantive,reason fordroppingthose few cases of warin

the firstyearof the countryseries may be thatseveral of them are left-censored.For

example, the warsin Greece or the USSR all startedbefore 1945, even thoughsomedatasets list them as startingon orafter 1945.If violence wasongoingbeforethatdate,thenit would have affected theright-hand idevariables n 1945. Two of thosecases-

Greece and South Korea (also Philippines)-are particularly important for thenonsignificanceof ef in the FearonandLaitin(2003) model because they have verylow efscores. Insum,theremaybesomethingdifferentaboutthe warsthatstarteddur-

ing the few yearsafterWorldWarII.Most were communist nsurgencieswithno clear

"ethnic"dimension,andthe meanvalue of ef for countries at warfrom 1945 to 1949was abouthalf thatof countriesatwarin the 1950s, 1960s, or anyother decade up to

1999. Thus, losing those cases by lagging explanatoryvariablesexplains how efcomes close to significance using FearonandLaitin'smodel and data.

I nowturn o ananalysisof civil warprevalence.Prevalence s defined as the union

of onset andcontinuationof war,so thedependentvariable s coded 1forallperiodsofwar.Iassume thatcivil waris a first-orderMarkovprocess (i.e., there s no significanttimedependencebeyondthe firstperiod)andestimateadynamicprobitmodelbyadd-

ing to the previousmodel a set of interaction ermsbetween a variabledenotingthe

prevalenceof civil war in the previous periodand all right-handside variables.The

estimatedcoefficients of the interaction ermscan, after a minoradjustment,be inter-

pretedasestimatesof therelationshipbetween theindependentvariablesandwarcon-

tinuation,conditionalon the occurrenceof warin thepreviousperiod.Theadjustmentinvolves correctingthe coefficient and standarderrorsof the interactionterms: for

example,thecoefficientof gdpll withrespectto warcontinuation s the sum of the twocoefficients for gdpll andwlgdp: (-0.124 + 0.293). The standard rror s the squareroot of the variance of gdpll plus the variance of wlgdp plus two times their

covariance.The estimatesin Table 7 arealready adjusted n this way; thus,asterisks

nexttocoefficient estimatesfor theinteraction erms ndicatesignificancewithrespectto warcontinuation.Estimatesof the linearterms refer to civil waronset, as before.

Theoverallpicture s one of substantialdifferences across civil war lists. Studying

prevalencebringsus a stepcloser to analyzingwarduration,and it is not surprising,

giventhelargedifferences nwaronset andterminationhighlighted nTable 1a-d,that

we should see differencesin estimatesof civil warprevalence.GDP is againrobustandnegative n allregressions,althoughpopulationnowis not

alwayssignificantwithrespectto waronset. Mostvariables(exceptGDP) aresignifi-cantinonlyone or afew models,and nstability s the most oftensignificant in9 out of12 models). Growth is not significant.74

74. Using thevariablegrollm (insteadof groll) againaddsonly five observationsanddoes not affect

the results substantially.

(text continues on p. 853)



TABLE7

DynamicProbitModels of Civil WarPrevalence,1960-1




atwar atwar2 atwar3 atwar4 atwar5 atwar6 atwar7 atwar8

GDP -0.126 -0.135 -0.098 -0.103 -0.061 -0.051 -0.100 -0.087

(0.038) (0.042) (0.031) (0.034) (0.024) (0.017) (0.033) (0.031)GDP growth -0.425 -0.281 0.418 0.460 -1.187 -0.692 -0.307 0.720

(0.995) (0.899) (0.778) (0.733) (0.854) (0.557) (0.829) (0.985)

Instability 0.361 0.348 0.122 0.259 0.289 0.258 0.259 0.403

(0.149) (0.136) (0.157) (0.154) (0.130) (0.106) (0.141) (0.155)

Anocracy 0.217 0.241 0.194 0.226 0.218 0.307 0.262 0.197

(0.142) (0.124) (0.135) (0.150) (0.135) (0.092) (0.137) (0.131)Oil exporter 0.327 0.273 0.206 0.269 0.237 0.201 -0.051 0.421

(0.206) (0.178) (0.127) (0.172) (0.164) (0.156) (0.147) (0.187)Ethnic frac-

tionalization 0.033 0.062 0.130 0.077 0.257 0.483 -0.064 0.170

(0.226) (0.211) (0.196) (0.242) (0.261) (0.175) (0.206) (0.220)

Population log) 0.053 0.043 0.058 0.038 0.054 0.057 0.077 0.028

(0.033) (0.032) (0.029) (0.031) (0.034) (0.031) (0.032) (0.031)Terrain 0.004 0.005 0.006 0.003 0.003 0.002 0.004 0.002

(0.003) (0.003) (0.003) (0.002) (0.003) (0.002) (0.003) (0.002)Percentage -0.001 0.001 0.003 0.003 0.002 0.002 0.003 0.000

Muslim (0.002) (0.002) (0.002) (0.001) (0.002) (0.001) (0.001) (0.001)

00



War(t 1)*Muslim -0.005 -0.003 -0.001 -0.007 -0.002 0.000 -0.003 0.001

(0.004) (0.004) (0.003) (0.004) (0.003) (0.003) (0.003) (0.003)

War(t 1)*GDP 0.169 0.147 0.033 0.201 -0.197 0.029 0.082 -0.003

(0.139) (0.110) (0.099) (0.111) (0.066) (0.053) (0.068) (0.058)

War(t 1)*Growth -0.021 -0.369 0.198 1.395 -0.451 1.036 1.200 0.419

(0.745) (0.615) (0.653) (0.838) (0.828) (0.644) (1.148) (0.592)

War(t 1)*Instability -0.533 -0.371 -0.266 -0.164 -0.172 -0.152 -0.256 0.158

(0.241) (0.223) (0.211) (0.217) (0.190) (0.179) (0.191) (0.173)

War(t 1)

*Anocracy 0.077 0.144 0.339 0.115 0.336 0.019 0.112 0.405

(0.238) (0.215) (0.193) (0.194) (0.175) (0.189) (0.200) (0.154)War(t 1)

* Oil -1.001 -0.643 -0.508 -0.286 -0.312 -0.283 -0.088 -0.198

(0.555) (0.514) (0.421) (0.383) (0.436) (0.334) (0.354) (0.367)

War(t 1)*Ethnic 1.561 1.379 0.761 1.001 1.233 0.306 1.330 0.562

(0.717) (0.587) (0.541) (0.772) (0.484) (0.443) (0.553) (0.370)

War(t 1)*Population 0.220 0.200 0.253 0.209 0.221 0.244 0.257 0.195

(0.043) (0.043) (0.039) (0.045) (0.040) (0.036) (0.041) (0.030)

War(t 1)*Terrain 0.000 0.000 0.000 0.008 0.005 0.005 0.002 0.004

(0.006) (0.005) (0.004) (0.005) (0.004) (0.004) (0.005) (0.004)

C0



TABLE 7 (continued)




atwar atwar2 atwar3 atwar4 atwar5 atwar6 atwar7 atwar8

_cons -3.016 -2.842 -3.245 -2.898 -3.201 -3.095 -3.381 -2.683

(0.573) (0.548) (0.486) (0.551) (0.550) (0.512) (0.549) (0.525)Observations 4,045 4,179 4,099 4,178 4,179 4,179 4,179 3,782

WaldX2 1,050.19 1,169.87 1,270.07 952.18 988.45 1,037.77 1,240.66 1,701.62

Loglikelihood -337.62 -398.73 -375.13 -366.71 -431.94 -687.92 -371.13 -417.82

Pseudo-R2 0.6996 0.6834 0.7199 0.7465 0.6134 0.6510 0.7887 0.7015

NOTE:Coefficients standardrrors) represented.Bold indicatessignificanceat.05orhigher; talicsindicates ignificanction termshavebeen adjusted o refer to war continuationonly.GDP = gross domesticproduct.

00



Sambanis WHATSCIVILWAR? 853

Thepicturebecomes muchmoreunstablewhen we consider resultswithrespecttowar continuation n the shadedportionof Table7. Here we see that the interactionbetween incomeandlaggedwarprevalence s oftennonsignificant,butoccasionallyit

is significantwith switching signs: in Gleditsch et al.'s (2001) civil war list (atwar5),high incomesignificantlyreduces warcontinuation,whereasin the two models basedon Doyle and Sambanis's(2000) data(atwarlO,atwarl 1), higherincomeonce a civil

waris ongoing has the effect of increasingthe risk of the warcontinuing.In Regan's(1996) and Licklider's(1995) data,the samerelationship s borderlinesignificant(at

.10).Instabilitywhile waris ongoing significantlyreduces the riskof warcontinuation

in two models(Singerand Small 1994;Regan 1996)andis nonsignificant ntherest.75

Similarly,anocracy s significant none model andborderline nanother wo butother-

wise nonsignificant.Oilexportsaregenerallynonsignificant.There s wideagreementthatpopulouscountrieswill havelongerwarsandno evidence thatcountrieswithsig-nificant Muslimpopulationsor mountainous errainwill have longercivil wars.

Interestingresultsemerge again with respectto ethnic fractionalization ef). The

interaction ermwith efis significantandpositively correlatedwith warcontinuationin fourregressions e.g., inregression7, usingFearonand Laitin's 2003 civil warlist)and close to significantin one more.Thus, theresultswith respectto thisvariableare

genuinely divided: about half the models would tell us that a very diverse countrywould havelong wars,once a waractuallystarted,whereastheother halfwouldpoint

to no statisticallysignificantrelationshipbetweenethnicityand warduration.Thereismore supportfor the positive association between ethnic fractionalizationand war

continuation nTable8, whereI presentprevalenceresultsfromthefour civil war lists

thatspantheentireperiodfrom 1945 to 1999, restricting heanalysisto fourcivil warlists. Here,we see the same sortof substantialdisagreementwithrespectto the effectsof income on war continuation: n one model, the coefficient is significantandnega-tive;inanother, t is significantandpositive;and in the othertwo, it is nonsignificant.76The morewe pushthe data,the more the differences in coding ruleswill matter.

CONCLUSION

This articledoes threethings.First, tdemonstrates hatthereare substantialdiffer-ences acrosscivil warlists with respectto the coding of the onset andterminationofcivil war.Exploringthose differencesanalyticallyreveals some conceptualconfusion

75. I am not concernedhere with the theoreticalexplanationof these results. How instabilitycanreduce warduration s unclear. tmaybe thata regimetransition owardsignificantdemocratizationwhichwould be coded as instability) atisfiessomeof therebels'demandsand eadstowartermination.Thecodingof the

instabilityvariablemaybe affected

bythe occurrenceof war,butit is not

my purposehereto resolve

these questionsof endogeneity.76. The differencesactuallybecome greater f we do not lag the first observations n each country

series(giving6,051 observations).Withrespect oonset,growth s significantandnegative ntwooutof four

regressions, nstability s significant n one regression,anocracy s significant n all regressions,oil exportsaresignificant n one regression,andpopulation s significantin one regression.Withrespectto continua-tion, there is no significant change from the differences noted in Table 8 (see the results in the online

supplement).




TABLE8

DynamicProbitModels of Civil WarPrevalence,1945-1999

Gleditsch Fearon Doyle and Sambanis

et al. (2001) and Laitin Sambanis 2000, (This

(Wars) (2003) Expanded) Study)

GDP -0.073 -0.095 -0.102 -0.085

(0.025) (0.027) (0.026) (0.025)GDP growth -1.006 -0.204 -0.187 -0.113

(0.626) (0.473) (0.411) (0.392)

Instability 0.273 0.243 0.219 0.213

(0.108) (0.108) (0.103) (0.109)

Anocracy 0.172 0.291 0.274 0.294

(0.112) (0.114) (0.115) (0.099)Oil exporter 0.276 0.067 0.148 0.148

(0.146) (0.151) (0.147) (0.130)Ethnic fractionalization 0.220 0.115 0.295 0.221

(0.218) (0.192) (0.199) (0.200)

Population log) 0.055 0.078 0.034 0.043

(0.027) (0.027) (0.026) (0.024)Terrain 0.003 0.003 0.001 0.002

(0.002) (0.002) (0.003) (0.003)

PercentageMuslim 0.001 0.002 0.000 0.002

(0.002) (0.001) (0.001) (0.001)

War(t 1) *GDP -0.153 0.042 0.111 0.022(0.073) (0.048) (0.041) (0.026)

War(t 1) ' Growth 0.108 0.288 -0.399 0.580

(0.615) (0.791) (0.495) (0.661)

War(t 1) *Instability -0.108 -0.309 -0.169 -0.040

(0.184) (0.178) (0.153) (0.166)

War(t 1) * Anocracy 0.319 0.054 0.159 0.021

(0.196) (0.200) (0.146) (0.177)

War(t 1) ' Oil -0.103 -0.042 -0.228 -0.087

(0.323) (0.291) (0.302) (0.284)War(t 1) *Ethnic 1.004 1.052 0.700 0.639

(0.417) (0.349) (0.303) (0.353)War(t 1) e Population 0.211 0.265 0.197 0.226

(0.035) (0.032) (0.029) (0.032)

War(t 1) * Terrain 0.004 0.003 0.002 0.003

(0.005) (0.003) (0.003) (0.003)War(t 1) * Muslim -0.001 -0.002 0.000 0.000

(0.003) (0.002) (0.003) (0.002)

_cons -3.159 -3.501 -2.689 -2.880

(0.436) (0.448) (0.429) (0.412)Observations 5,893 5,893 5,893 5,893

WaldX2 791.89 1,736.07 1,898.86 2,022.78Log likelihood -596.74 -548.11 -694.25 -671.34Pseudo-R2 0.5797 0.7731 0.6963 0.7185

NOTE:Coefficients(standard rrors)arepresented.Bold indicatessignificanceat .05 orhigher; talics indi-catessignificanceat .10. Parameters or all interaction ermshave beenadjusted o refer o war continuation

only. GDP = gross domesticproduct.




aboutthemeaningof civil war and can serve as the foundation or atheoreticaldiscus-sion thatestablisheswhat civil war s and how itcanbe distinguished rom otherformsof politicalviolence. Second, it proposesa new coding ruleforcivil war thatattempts

to sidestepsomeof theproblems dentified nothercodingrules andoffersa new listofcivil wars that is based on this new coding rule. Third,it measuresthe substantive

implications of differences in coding rules by formally comparing the empiricalresultsthatwe get froma civil warmodel when we use 12differentrulesto code civilWar.

Thequantitativeiterature n civil warrevealsaremarkable egreeof disagreementon how tocode theonset and termination f wars,and the literatures fuzzy on how to

distinguishamongdifferent ormsof politicalviolence. Thisimplies the need fortheo-

rizingaboutcivil war andthenforpropermeasurementof theconcept.Giventhecod-

ing complexitiesI have identified,researchers hould conductrobustnesstests usingdifferentcivil warlists andjustify theircoding decisions as transparently s possible.Differencesinthecodingof civil warhave substantivemplications,butperhapsnotas

largeas one mighthaveexpected.The resultspresentedhere show that the estimatedcoefficients of most variablesvary widely as a result of changes in the coded onset ofcivil war.Ina few cases, thesign of the coefficient also changes, andsignificancelev-els also varyacross data sets. Several variablesare not robust o changesin thecodingof civil war onset, most strikingly in the case of lagged war prevalence, ethnicfractionalization ndoil exports,and,to a much lesserextent,anocracyandinstability.

Predictionsof whenandwhere civil warmightoccur arelikely to dependcriticallyon which coding rulewe use. Moreimportant, stimates of the substantiveeffects of

policy interventions oreducetheriskof civil warby manipulatinghe level of income,

politicalinstability,oranyof theother"manipulable" ariables n the model will also

varywidely, dependingon the coding rule for war onset andtermination.At the same time, some variables are remarkablyrobust to coding differences.

Incomelevel andpopulationsize areveryrobustandsignificantlyassociatedwithcivilwar.Thatsignificancelevels forthese variablesdo notchangemuchdespite largedif-ferences in coding rules is likely because these variableschange slowly over time. A

problemthatthis highlights is that we cannotrely on themto make accuratepredic-tions of the timing of war onset. Othervariables-especially mountainousterrain,Muslim population, and economic growth-are consistently nonsignificant (withminorexceptions).

The results from models of war prevalencesuggest thatpredictionsof civil wardurationwill be even less accuratethan predictionsof civil war onset. There was

greater nstabilityof empiricalresults ntheprevalencemodel, so analysesof civil wardurationwill be much more affectedby differences in the coding rules. Robustnesstests are thereforeessential when analyzingwardurationor termination.

Overall,civil war models seem to be good at identifyingcountrieswith long-termproclivities o civil war.Butif the modelsarefurtherdevelopedto be ableto predict he

timingof waronsetbetter, henwe are ikelytosee codingdifferences n thedependentvariableaffect theparameter stimatesmoresignificantly(thechangeswe found inthecoefficient sign of economic growth,which changes substantiallyover time, are anindicationof this). Thus, differencesin coding will begin to matter f the theoretical




models "catchup"to the databy includingmore time-sensitive variables.As the the-

orydevelopsmore,it seems likely thatanalystsof civil warswill haveto go backto the

datato develop a higherdegreeof consensus on the meaningof civil war before they

canproduceaccurateandcrediblepredictionsof where andwhen civil warswill occur.Thatsaid,Ihaveno clearansweron whether t is better ohave a single definitionor

codingrulefor civil war orif it is more beneficialto havedifferingdefinitions. On one

hand,some standardization f ourcodingruleswouldreflecttheoreticalconsensus on

whatcivil war s, andhomogeneityinmeasurementwould ruleout one possible source

of differencein empiricalresults. On the otherhand,thatwe have many coding rules

suggests that the concept of civil warmay mean differentthings to differentpeople,andcoding rules should reflect differencesin subjective understandingsof that con-

cept. Analyzing the differencesacross coding rules can be instructivebecause these

differences map out the space within which we might be able to find a sharedunderstanding f civil war.

A substantiveresult from this analysis, and to which I must return, s that ethnic

fractionalizationmaynot be as nonsignificantacorrelateof civil war as manyscholars

haveargued. tssignificancehinges perhaps oo muchonthecodingrules forcivil war,but ethnic fractionalization s clearly important n explaining a broad category of

armedconflict that ncludesminor nsurgency.Thisis animportant esult,consideringhowdifficult t was toclearly distinguishbetweencivil war and other forms andlevels

of violence. Ethnicfractionalizationwas sometimes significantlycorrelatedto both

waronset and warcontinuation,but these resultswere at times affected by the fewwarsthat aredroppedwhen we lag explanatoryvariablesby 1year.However,differ-

ences in coding rules seem to explain the nonsignificanceof thatvariablein Fearon

andLaitin's(2003) results.Inother runs(see supplement),religious fractionalization

andthesize of thelargestconfessionweresignificantand often dominated heeffect of

ethnicfractionalization.77hus,ethnoreligious dentitymayhave been writtenoff too

quickly as a correlateof large-scale armed conflict-even "civil war"-by manyscholars.

Several other substantiveresults are worth noting. Economic growth may be

endogenousto civil war,which impliestheneedfora differentestimationstrategy ormodels that includethat variable.The political variablesof anocracyand instability,which are central to some models of civil war,are sensitive to the coding rules but

muchless so if we examinethe entirepost-1945period.The significanceof thepopu-lationvariablemaybe tiedto thehigh thresholdof violence used to identifycivil war

anddistinguishit from minor violence. Mountainous errain-a key measureof the

technologyof insurgency n civil warin some models-is notrobust o changes in the

coding rules, and neither is the measure used to identify countries that are major

exportersof oil. Peacedurationor war in thepreviousyearis also sensitive tochanges

incoding.Moreover, hesignificanceof someof these variables s affectedby whetherobservationsof ongoing warare dropped(e.g., anocracyis much more consistently

significantin Table4 thanin Table2). Finally,the periodanalyzed influences some

77. Forananalysisof thesensitivityof empiricalresults n the civil warliterature o smallchangesinthe specificationof the model, see HegreandSambanis(2004).




results due to the small number of war starts: for example, instability is more

consistentlysignificantin Table 6 thanin Table4.

Despitethese differencesanddifficulties,theconclusion from thisstudyshould not

be thatcoding warsandanalyzingthemquantitativelys a futileexercise. Rather hanabandon hese efforts,I favorredoubling hemby improvingthe coding rules,apply-ing themtransparentlyo thedata,andstudyingtheimplicationsof differences across

codingrules.Thelegacyof theCorrelatesof Warproject s thatwe nowhaveatourdis-

posalreplicabledataon civil warthatwe cananalyzequantitativelyo pointto thethe-oreticalandempiricalcomplexities of defining andmeasuringcivil war.This article

suggests ways of buildingon thatlegacy.

REFERENCES

Collier,Paul,and AnkeHoeffler.2001. Greedandgrievance n civil war.PolicyResearchPaper2355, World

Bank.

Connor,Ken. 1998. Ghost orce: The secrethistoryof the SAS. London: Weidenfeldand Nicolson.

Doyle, Michael,and Nicholas Sambanis.2000. International eacebuilding:A theoreticalandquantitative

analysis.American Political Science Review94 (4): 779-801.

Fearon,JamesD. 2003. Ethnicandculturaldiversityby country. ournalofEconomicGrowth8:195-222.

Fearon,JamesD., and David D. Laitin.2003. Ethnicity, nsurgency,and civil war.AmericanPolitical Sci-

ence Review 97 (1): 75-90.

Ghobarah,H., PaulHuth,and BruceRussett.2003. Civil wars kill and maimpeople, long afterthe fighting

stops.American Political Science Review97 (2): 189-202.

Gleditsch,Nils Petter,HAvard trand,MikaelEriksson,Margareta ollenberg,and PeterWallensteen.2001.

Armed conflict 1945-99: A newdataset.Unpublishedpaper,PRIO, Oslo, Norway.Gleditsch,Nils Petter,PeterWallensteen,MikaelEriksson,Margareta ollenberg,andHavardStrand.2002.

Armed conflict 1946-2001:A new dataset.Journalof Peace Research 39 (5): 615-37.

Harff,Barbara. 003. No lessons learned romthe Holocaust?Assessingrisksof genocideandpoliticalmass

murder ince 1955.American PoliticalScience Review 97 (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.

Hegre,Havard,nd Nicholas Sambanis.2004.

Sensitivity analysisof

empiricalresults n the

quantitativeit-

eratureon civil war.Unpublishedmanuscript,PRIO andYaleUniversity.

Leitenberg,Milton. 2001. Deathsin wars andconflicts between 1945and2000. Paperprepared or theCon-

ference on Data Collection in ArmedConflict, June8-9, Uppsala,Sweden.

Licklider,Roy. 1995.Theconsequencesof negotiatedsettlements ncivil wars,1945-1993.AmericanPoliti-

cal Science Review 89 (3): 681-90.

Marshall,Monty,and KeithJaggers.2000. Polity IV project[Codebookand 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.

McAdam,Doug, SidneyTarrow, ndCharlesTilly. 2001. Dynamics of contention.Cambridge,UK:Cam-

bridgeUniversityPress.Price,D. L. 1975. Oman:Insurgencyanddevelopment.Conflictstudies. London:Institute orthe Studyof

Conflict.

Regan,Patrick.1996. Conditions orsuccessful thirdparty nterventions. ournalof ConflictResolution40

(1): 336-59.

Sambanis,Nicholas. 2001. Do ethnic and nonethniccivil wars have the same causes? A theoretical and

empirical inquiry Part1).Journalof ConflictResolution45 (3): 259-82.




. 2004. Expandingeconomic models of civil warusing case studies.Perspectiveson Politics 2 (2):259-79.

Sarkees,MeredithReid. 2000. The Correlatesof Wardataon war:An update o 1997.ConflictManagementand Peace Science 18 (1): 123-44.

Sarkees,MeredithReid,andJ. DavidSinger.2001. The Correlatesof Wardatasets:Thetotalityof war.Paperprepared or the 42nd Annual Convention of the InternationalStudies Association, February20-24,

Chicago.

Singer,J.David,and MelvinSmall. 1972. Thewages of war,1816-1965: A statistical handbook.New York:

JohnWiley.. 1994. Correlatesof Warproject:International nd civil wardata,1816-1992 [Computer ile, Study

#9905]. Ann Arbor,MI:InteruniversityConsortium or Political and Social Research [distributor].

Small, Melvin, and J. David Singer.1982. Resort to arms:Internationaland civil war,1816-1980. BeverlyHills, CA: Sage.

Strand, HAvard,Lars Wilhelmsen, and Nils Petter Gleditsch, in collaboration with Peter Wallensteen,

MargaretaSollenberg, Mikael Eriksson,HalvardBulhaug, and Jan Ketil Rod. 2003. Armedconflictdatasetcodebook,version1.2a. Accessed June21, 2004, fromhttp://www.prio.no/page/Project_detail//9244/44532.html?PHPSESSID=e92e0a3b72738fae3ff8585c2cd291b/.

Tilly,Charles. 1978. Frommodernization o revolution.New York:RandomHouse.

. 2003. Thepolitics of collective violence.Cambridge,UK: CambridgeUniversityPress.

Valentino,Benjamin,PaulHuth,andDylan Balch-Lindsay.2001. Draining he sea: Mass killing, genocide,and guerillawarfare.Unpublishedmanuscript,StanfordUniversity.

Walter,Barbara. 002. Committingopeace: Thesuccessfulsettlementof civil wars.Princeton,NJ: Prince-

ton UniversityPress.

Documents

What Is Civil War.pdf