Contextualizing-Riots:-Changing-Explanations-of-Political-Violence-in-India,- 1971>2000 JayKrehbielandSunitaParikh jkrehbiel@go.wustl.edusaparikh@wustl.edu DepartmentofPoliticalScience WashingtonUniversityinSt.Louis PRELIMINARYDRAFT.DONOTCITEORQUOTEWITHOUTPERMISSION. INTRODUCTION Contemporaryresearchacrossthesocialsciencesisweakatconstructingmodels andexplanatoryframeworksthatincorporatefactorsfromdifferentparadigms.the tendencytodescribetheoriesas"institutional"or"behavioral",forexample, suggeststhattherearenaturaloppositionsbetweentheseframeworksthatcannot beovercome,eventhoughmostscholarsagreethatthemostpowerfulexplanations ofsocialphenomenacombineattributesfromdiverseapproaches.explanationsof collectiveviolencehavebeenparticularlysusceptibletothiseithersorapproach,and fewstudieshaveintegratedindividualpreferencesandstrategieswithcontextual andstructuralfactors(forrecentexceptionstothesetrends,seeclaassen,christia). Similarly,studiesofprotestdivideinto"ethnic"and"nonSethnic"categories,and althoughscholarsinonestreammayciteresearchfromtheother,thetwocamps rarelyinformeachothertheoreticallyandempirically. Indiaprovidesaparticularlyfertilecasetounderstandtheintersectionof
interests,identity,andviolenceinpolitics,sinceethnicheterogeneityandthe practiceofdemocraticpoliticshavecombinedtomakecaste,religious,linguistic, andregionalattachmentshighlysalientintheindianpoliticalarena.itprovidesan equallyattractivecontextinwhichtodevelopatheoreticalframeworkthat incorporatesamoreheterogeneousrangeofidentitysbasedbehaviorinexplaining collectiveviolence. Inthispaper,whichispartofalargerstudy,wefocusoncollectiveviolence thattakestheformofriotevents.ourreasonsforchoosingthisfocusareboth practicalandsubstantive.whilewearesympathetictotilly'sargumentthatthe designationofanincidentasa"riot"byauthorities"embodiesapoliticaljudgment ratherthanananalyticaldistinction"(2003:18),thedatausedinthispaperare governedbyaconsistentdefinition:riotsascategorizedbytheindianpoliceusing thedefinitionfoundintheindianpenalcode.whilethisdefinition(andthedata thatresultfromit)hasitsownproblematicaspects,itprovidesimportant consistencyacrosstimeandspace. TheCase:CollectiveViolenceinIndia1970S2000 Indiaprovidesapromisingcontextinwhichtoexplorehowrepertoiresof identityandrepertoiresofstrategyinteracttoproduceviolentcollectiveaction. First,Indiansocietyisheterogeneousonseveralculturaldimensions,including religion,socialstratificationthroughcaste,andregionaldifferentiationthrough language,dress,andothersocialpractices.thesedimensionshavebeenhighly salient,althoughvariablyso,forcenturies,andtheycontinuetobeissuesaround whichgroupidentitiesaredefinedtoday.thereare,therefore,arangeofidentities 2
aroundwhichpeoplecanformattachmentsandwhichhavesalienceinsociety. Second,theIndianpoliticalsystemprovidesconsiderableopportunityfor participantstodrawnonrepertoiresofstrategytopracticecontentiouspolitics. Withtheexceptionofthesuspensionofdemocraticpoliticsfrom1975to1977, Indianpoliticshasbeencharacterizedbyregularandcompetitiveelectionsandthe steadyincorporationofnewgroupsintotheelectoralarena.evenintheheydayof Congressdominanceduringthe1950sand1960s,otherpartiessuccessfully contestedstateandnationalelections,andthecongresssplitwhichoccurredinthe late1960swasgalvanizedbythelossesthepartysufferedinthe1967elections. SincethedeclineoftheCongressParty,beginninginthemidS1980s,electoral competitionhasbecomeevenfiercerandlesspredictable.afterdominatingthe nationalpoliticalscenethroughmajoritypartygovernmentsforthirtyofitsfirst thirtysthreeyears,congresshasfailedtoformonesince1989.atthesametime,the wayinwhichelectoralcoalitionsareconstructedhaschanged,asthestrategiesof the1950s,whichemphasizedmobilizationthroughtraditionalhierarchies,gave waytodirectappealstomiddleandlowcastegroups,andelitedominanceofparty leadershippositionswereincreasinglychallengedbypoliticiansfromthesesocial groups. Third,theelectoralsalienceofidentityissueshasbeenamainstayofIndian politicssincethedevelopmentofindigenouspoliticalinstitutionsduringcolonial rule.intheearlystagesofmobilizationinthe1910sand1920s,thecombinationof adiverse,largelyilliterate,andpoliticallyunsophisticatedcitizenrywiththeneed foreachpoliticalpartiestocommandamassbaseledpartiestousesymbolsthat 3
hadsalienceforlargenumbersofsupporters.notsurprisingly,themostsuccessful tendedtobesymbolswithregional,religious,orcastesignificance.afterpartition andindependence,thedominantcongresspartyavoidedexplicitappealsofthis typeforavarietyofreasons,butsincethe1970stheyhavereemergedandbecome evenmorepowerfulastoolsforpoliticalmobilizationandcoalitionbuilding. Finally,thesocialandpoliticalcontextinIndiahasfacilitatedtheuseof collectiveviolenceinthesocialandpoliticalarenas.althoughthemotivationsof participantsareascribedtoarangeofidentitymotivations,theusefulnessof collectiveviolenceinachievingshortsandlongstermgoalshasremainedrelatively consistent.thosewhohavebenefitedincludedsocialandpoliticalelites,political parties,andcivilsocietyorganizations.therecontinuestobedebateoverthe motivationsofparticipants(brass,wilkinson,varshney,shani);whethercollective violenceisprimarilyidentitysbasedorinstrumental(wilkinson,chandra,chandra andwilkinson,brass);andwhetherviolenceisincreasingordecreasing(wilkinson, Brass,Gilley).However,thereisgeneralagreementthatcollectiveviolence,andriots inparticular,areaconstantinindianlife. TRENDSINEXPLANATIONSOFINDIA SVIOLENCE Inthepasttwodecades,scholarsandotherobservershavefocusedoncommunal violence.theemergenceofthehindunationalistbharatiyajanataparty(bjp)asthe majornationalchallengertocongress,thewidespreadcommunalviolencethat followedthedemolitionofthebabrimasjidin1992,andtheriseofantismuslim rhetoricinpublicdiscourse,haveallfueledconcernsaboutcommunalconflict. 4
Moreover,thecontinuedstrengthoftheBJPinGujaratandtheantiSMuslimcarnage thatfollowedthegodhrariotsin2002hasdemonstratedthateconomic developmentandincreasedprosperitydoesnotnecessarilydampencommunal tensions. ManyoftheprevailingexplanationsoftheincreaseinHindunationalist feelingandthesuccessofthebjpstressnonsaffectivefactorsratherantismuslim sentiments,increasedreligiositybyhindus,orhighscastebacklashagainstthe ambitionsoflowscastegroups.theincreaseincasteandcommunalviolencehas beenattributedtoelitepoliticalstrategies,muchasothertypesofethnicconflict (e.g.,separatisminthenortheaststatesandpunjab,assameseviolenceagainst Bangladeshis)areexplainedintermsofpoliticallystrategicchoices(Weiner1978, 1989,Ludden1996,Varshney1993,Jaffrelot1993).MicroSstudiesofviolenceand politicalstrategiesprovidesupportforthesehypotheses.yagnikandbhatt's excellentanalysisofthe1981riotsinahmedabad,gujarathighlightstheroleof politicalparties,hiredthugs,andpoliticalelitesindeepeningandcontinuingriots thatbeganforonesetofreasonsandweresustainedwithothers(1984).case studiesofthe1992carnageinsurat,gujaratafterthedemolitionofthemosquein AyodhyailluminatetheroleoftheBJPlocalgovernment(cites),andthecomplicity ofthebjp'sstateandlocalunitsintheaftermathofgodhrain2004hasbeenstarkly demonstrated(wilkinsonandhaid2009). Theseargumentsarecompellingandmakeintuitivesense.Thereis widespreadagreementthataspoliticalpartieshavehadtocompeteforvotes,they haveusedcastesandcommunitysbasedpoliciessuchasreservationstoattract 5
support(basuandkohli1998,parikh1997,wilkinson2004,brass2003,chandra 2004,,Jaffrelot1993).Partiesregularlynominatemembersoflocallydominant casteandreligiouscommunitiesforparliamentaryseats;eventhebjpputsup MuslimandScheduledCastecandidatesinconstituenciesdominatedbythose groups(boseandsingh1985).thecombinationofincreasinglyfierceparty competitionandanethnicallydiverse,politicallyawareelectoratemeansthat culturalandpoliticalfactorsareoftensothoroughlyintertwinedthatitisdifficultto distinguishtheseparateeffectsofeach. Nevertheless,thetendencyofsomanywriterstoattributeIndia's contemporaryethnicconflictsentirelyorprimarilytopoliticalmachinationsseems tobetoosweeping.itisdifficulttoimaginethatevenverycleverpolitical manipulatorscouldwhipupcommunalfrenzyonpreviouslyuntilledground.and theemphasisontheroleofthebjpanditsalliedorganizationsinfomentingviolence sincethemids1980sunderplaystwoequallyimportantfactors:first,communal violencehasalonghistoryinindia,onethatprecedesthedominanceofthebjp;and second,communalviolenceisfarfromtheonlytypeofviolencethatindians experience.theacceptanceofproshindu,antismuslimrhetoricinthepoliticalarena bymajorpoliticalpartiesisamarkeddeparturefromthepoliticaldiscourseofthe eraofcongressdominance.buthasthischangeinrhetoricchangedthenatureof collectiveviolence,orhasitrefocusedattentiontocommunalviolencebecause politicalpartiesappearmorewillingtotolerateit? Iftheansweristhelatter,thatcommunalviolenceisnowmoreacceptablein thepoliticalarena,thenthesocialandpoliticalramificationsofthischangeare 6
importanttounderstand,butnotattheexpenseofstudyingothertypesofidentitys framedviolenceorviolencethatisnotclearlyrelatedtoidentityissues.athorough analysisofthefactorsthatshapecollectiveviolenceneedstobeginbyconsidering allpossiblefactors,whichincludecommunalmotivationsbutarenotlimitedto those. Ifwelookbackatstudiesofcollectiveviolenceoverthepastfewdecades,we candiscerncertaintrendsintheliterature.whileriotsofalltypes(i.e.,communal, caste,andnonsidentitybased)havebeenobservedthroughoutthisperiod,thefocus ofattentionhasshiftedovertime,especiallywhenexplainingviolencethatfitsthe definitionofcontentiouspolitics,orpoliticsbyothermeans(tilly). Inthelate1960sand1970s,economicgrievanceswerefrequentlycitedas thefactorsbehindsocialunrest,withcommunalandcasteriotsseenasless efficaciouspolitically(cites).fromthelate1970suntilthe1992supremecourt decisioninindra&sawney&v.&union&of&india,casteviolencebecamethefocusof attentioninpoliticalconflictstudies.andfrom1992untilthemids2000s,communal conflicttookoverasthemostcommonexplanationforpoliticalviolence.atpresent, theincreaseinregionalandeconomicconflicthasledtorenewedattention,butit hasnotdisplacedcommunalconflictfrompreeminence. Theseshiftsinemphasisreflectchangesinthepoliticalefficacyofdifferent motivationsforcollectiveconflict.buttowhatextentdotheyreflectempirical changes?inordertoanswerthisquestion,weanalyzequantitativedataonalltypes ofriotsinindia,from1971to2000.thisperiodcomprisestheendofthecongress PartyasitexistedunderNehru(andthenbrieflySastri)andtheemergenceofdirect 7
massmobilizationofcastes,especiallylowcastesanddalits.itfeaturesthehighs watereraofindiragandhi'sgovernmentfollowedbytheendofcongressmajority partygovernment,theriseofthebjp,andtheemergenceandconsolidationofthe coalitiongovernmentregimeinindianelectoralpolitics.thusthedataincorporate variationsinpoliticalandinstitutionalconditionsaswellasvariationsin explanationsforcollectiveconflict. THEORYANDHYPOTHESES Inpreviousandcurrentresearch,Parikhhassetforthatheoryoftheindividual propensitytoriot(cameronandparikh1999,parikhn.d.).atthemostbasiclevel, riotsarecreatedbyindividualparticipantsactingcollectively.theirdecisionsto jointheriotarebasedonfactorsspecifictoeachindividual,butthesefactorsare rootedinandtriggeredbythesocialcontextinwhichtheyoccur.theinfluenceof contextcanbeassimpleastheabsenceofthepolice,whichdirectlylowersthecosts ofparticipation,oritcanbeascomplexasacombinationofsocialattitudes,political receptivity,andhistoricalexperiencewithcollectiveviolence.butwhethertheevent ismonocausalorcomplex,riotsaretheproductoftherelationshipbetween individualdecisionstojoinandthesocialcontextinwhichtheriotsoccur. Riotsmayhaveagendasandconsequencesthatarenotexplicitlypolitical. Butifarioteventhasapoliticalagenda,thentherelationshipbetweenindividual motivationsandsocialcontextacquiresathirdcomponent:collectiveviolencemust beseenasacceptablewithinthepoliticalarenaforanygivenincidenttoachievea politicalgoal.thesefactorscanbesummarizedasfollows: 8
Foranytypeofriottooccur,theremustbeasufficientnumberofindividuals whoarewillingtoengageinviolentbehavior.thiswillingnesscanstem fromavarietyofpreferences,motivations,andinterests.riotsthatmeetthis basicthresholdarefrequentlysmallandlocal,withnosignificancebeyond theirimmediatetimeandplace. Forariottohaveaneffectbeyonditsimmediateimpact,thesocialcontextin whichittakesplacemustbereceptivetotheuseofviolenceasameanstoa goal.ataminimumtheremustexistasocialcontextthatisneutraltotheuse ofviolence,andmoreoftenthereexistsareceptiveorapprovingcontext. Riotsthatmeetthisthresholdcanbesmall,buttheyaremorelikelytobe large.theymayberestrictedtoalocalityorextendbeyondtheinitial territory,dependingonthegoal. Forariottohaveapoliticaloutcome,theremustexistacontextofpolitical efficacy,anormativeviewwithinsocietythatcollectiveviolenceisatypeof "politicsbyothermeans."riotsthatmeetthisthresholdareunlikelytobe small.dependingonthepoliticaloutcomessought,theymayremainlocalor extendbeyondtheterritory. Therearethreetypesofparticipants:violenceSseeking,affective,and instrumental.wedescribetheseinturn. Almosteveryviolentcollectiveeventhasagroupofviolence3seeking& participantswhoappearandtakepartforthepurepleasureofengaginginviolent behavior.theseactorsareoftendescribedpejorativelyasthugs,asbosinbritish parlance,orbyothertermsindicatingantissocial,aberranttendencies.whilethese 9
typesofactorsweremorefrequentlytakennoteofinearlystudiesofmobbehavior (Shoemaker2004,LeBon),theywerepaidlessattentionintheoriesthat emphasizedrationalandpoliticallystrategicactors.nevertheless,acomprehensive modelofriotsshouldincludethem,notonlyforthesakeofcompleteness,but becausetheycanhaveanimpactonwhetherariotoccurs. Thesecondtypeofplayeristheaffective&participant,whojoinstheriotto expresshisfeelingsaboutthereasonfortheevent.thesefeelingsdonotnecessarily havetohaveapoliticalcomponent.theaffectiveparticipantmaychooseto demonstrateonaholydaytoemphasizereligiouscommitment;hemayprotest lowerscastemembers'useofacommunitydrinkingswaterwell;orhemaysimplybe joiningotherswhoareprotestingagainstmembersofarivalsportsteam,inanact ofsolidaritythatgiveshimasenseofprideandenjoyment.thistypeofbenefitis differentfromtheviolencesseekerbecausethemereactofparticipationisnot sufficientsstheremustbeaperceivedsubstantiveissuessbutaffectiveparticipants arenotnecessarilytryingtochangeastatusquothroughpressureonpolitical agents. Thethirdtypeofplayeristhepolicy3oriented&participant,whojoinstheriot inordertobringaboutapoliticaloutcome.thisoutcomecanbeasspecificasa pricesubsidyoranelectionresult,orabroaderrangeofpolicies,suchasthe implementationofsharialawincertainissueareas.policysorientedparticipants onlyjoinrioteventswhentheythinktheyhaveareasonablechanceofbringing aboutthisfavoredoutcomeasaresultoftheevent.participationintheeventisa meanstoanend,notanendinitself(remembertheparticipantsinthismodelare 10
allpuretypes).theseparticipantsmostcloselyresemblethestrategicactorsofthe politicalopportunitymodelandtheinstrumentalistethnicityliterature.forallthree typesofparticipants,therearecoststoprotesting,whichrangefrompersonalinjury andarresttotheopportunitycostsofrioting.forallparticipants,theoverallcosts decreaseasthenumberofparticipantsincreases.forviolencesseeking participants,benefitsaccrueassoonasthedecisiontoparticipatehasbeen acteduponandhejoinstheriot.eveniftheriotisbriefandrelativelyunsuccessful, somebenefitsfromparticipatingwillbeobtained. Foraffectiveparticipants,benefitstoparticipationaccrueaslongasthe thresholdconditionofariotwiththedesiredissuesaliencehasbeenmet.issue saliencecanbeachievedinavarietyofways,includingacoordinatingeventthat signalsaparticularissueortheinterpretationoftheriotasissuesorientedbynons participants. ForpolicySorientedparticipants,benefitsoccurasaconsequenceoftheriot butnotnecessarilyduringtheeventitself.theseparticipantshaveapayoffthatis conditionedontheprobabilityoftheriotbringingaboutpolicyoutcomes.ifthe outcomesareanonsdivisiblegood,suchasageneralpolicychangeoranelection result,bothparticipantsandnonsparticipantswhoarepolicysorientedwillreceive thebenefitsoftheriotevent.thistypeofpayoffhastwoconsequencesfor participation:(1)participationintheriotdoesnotguaranteebenefits;and(2)there areincentivestofreesrideontheparticipationofothers. Theutilityfunctionsoftwoofthethreetypesofparticipantsallowusto deriveseveralhypotheses(thepresenceorabsenceofviolencesseekingparticipants 11
isdifficulttopredictwithoutindividualslevelpsychologicaldata),especiallyina contextsuchasindia,whereeconomic,ethnic,andpoliticalfactorsarethoughttobe connectedtoriotevents.wecannotdirectlyconfirmindividualmotivationswith observationaldata,butthepresenceofidentitycharacteristicscanhelpusinfer whetherthereisasufficientpotentialpopulationtojoinriotevents.wederive severalhypothesesfromtheindividualsleveltheory,theexistingliterature,andthe ourpreviousdiscussionabouttheextenttowhichscholarlyandpublicnarratives aboutriotschangegearsacrossthetimeperiod. First and most basically, affective attachments and policy changes matter in explaining conflict; if they do not, then our assumptions are implausible. Hypothesis 1: Identity/attachment factors are positively correlated with conflict. In the Indian context, these salient identities include caste and religion. Hypothesis 2: Economic factors are positively correlated with conflict. In our discussion of economic conditions as creating salient identity attachments, these factors include migration patterns, levels of industrialization, and the distribution of landless to landed agricultural workers. Hypothesis 3: Political issues are positively correlated with conflict. Given our theory of the role of policy-oriented participants in riot events and the existing literature on the relationship between electoral politics and violent collective action, we should observe that political events, particularly elections, are correlated with heightened levels of violence. We note, however, that scholars have argued for a nuanced view of how competitive elections may suppress communal violence. 12
RIOTSINFIVEINDIANSTATES:THEDATA Wegathereddataaccordingtothe1971boundariesforfiveIndianstates:Uttar Pradesh,MadyaPradesh,Bihar,Gujarat,andKarnataka(threenewstateswere createdin2000,butthishasnoeffectonouranalysis).thesestatesconstitute nearlyhalfofindia'stotalpopulation.thedataforeachstatewerecompiledfrom threesources:the&crime&in&india&reports,publishedbythenationalcrimerecords Bureau;Statistical&Reports&of&General&ElectionsfromtheElectionCommissionof India;andtheCensusofIndia'sdecennialreports.Rawfigureswerecollectedatthe districtlevelforarangeofvariables,includingnumberofriots,totalnumberof electors,totalvotes,topthreevotesgettingparties,anddemographicmeasuressuch aslevelsofliteracy,agriculturalworkersandthelike. OurdatacompriseneitherthefulluniversenorarandomsampleofIndian administrativedistricts.entering,coding,andcleaningdataforalldistrictswas beyondourcapacity,andarandomsamplewouldhavebeenrepresentativeforallof Indiabutnotnecessarilyforparticularstates.Thereisreason,however,toexpect thatoursampleisafairlyrepresentativeone.first,thefivestateswhosedistricts levelcharacteristicsweanalyzearemajorindianstatesfromdifferentregionsofthe countryandvaryontheindependentanddependentvariablesofinterest. Second,theincidenceofriotsbasedonoursampleiscloselyrelatedtothe incidenceofriotsforallofindia.theyearlymeannumberofriotsoccurringwithin thedistrictsincludedinoursampleandtheyearlymeannumberofriotsoccurring throughoutallofindiaexhibitsimilarpatternsovertime.inaddition,thesimilarity intheannualmeannumberofpercapitariotsbasedonthedistrictsinoursample 13
andthatbasedonindiaasawholeisevenmorestriking.indeed,thebivariate correlationbetweentheseyearlypercapitameansisroughly.80. Weapproachourmeasureofriotsdifferentlythanmostotherquantitative studiesofindianviolence.previousstudieshavetendedtodefineariotasanevent inwhichatleastonefatalityoccurs(e.g.,varshneyandwilkinson1996,wilkinson 2004,White1993,Moaddel1994,Guptaetal.1993,Thompson1989,Midlarsky 1988),whereasourmeasureofriotsisinclusiveofarangeoflethalandnonSlethal eventsthatinvolvecollectiveviolence.inthisstudyariotisgiventheindianpenal Code(IPC)definition:aneventinvolvingagroupoffiveormoreindividualswho areillegallyassembledandwhouseviolenceinpursuitofacommongoal.(ipc Section146). Thisdefinitionisthebasisforcategorizingriotsbylawenforcement authorities.thisstrategyallowsustospeakmoreconfidentlyaboutcausalfactors forriotsgenerallyandnotaboutthoseforcommunalriotsonly.thismoreinclusive definitionmeansthatourmeasureofriotsisnotconfinedtocatastrophicevents withwithdeathsandisassomethingexperiencedonamuchwiderscalethanother measuresallow.itshouldbenoted,however,thatjustasriotsmeasuredintermsof deathsarelikelytoundercountsmaller,andoftenmorerural,riots,ourdataare likelytoundercountlarge,lethalriots.whilelargeriotssuchasthosethattook placeindecember1992willberepresentedbyanaggregatecountofrioteventsat thetime,theremaybeanoverrepresentationofsmall,fleetingriotevents. Ourmeasurediffersinasecondandcriticalwayfrompreviousstudiesby relyingondistrictlevelstatisticsforentirestatesratherthanselectcity,state,or 14
countryleveldata.thismeasureisdesirablebecauseitallowsustotestfor variationinriotswithinparticularstatesaswellasbetweenstateswithinindiaasa whole.thisflexibilityisimportantgiventhesizeanddiversityofmanyindian states:uttarpradesh,forexample,hada1991populationofnearly140millionand includeswithinitspres2001boundariesthegangeticplainsandthehimalayas. Moreover,bycapturingdifferenceswithinstateswecanavoidgeneralizationsbased onstatesleveldata;forinstance,gujaratisregularlydescribedasaprosperousand literatestate.buttherearedistrictsingujaratwhoseliteracyratesarelowerthan somedistrictsinthestateofbihar,whichisgenerallyconsideredtobebackward andpoor.thediversityissignificantwiththenumberofdistrictsineachstate varyingbetween17and54foratotalof151districtsacrossthefivestates. 1 DATAANDVARIABLES Inouranalysis,weemployanegativebinomialestimationprocedure.Two featuresofthedatamadethisapproachthemostappropriate.first,thedataare measuredacrosstime(from1971 2000)andacrossspace(with151districts),and weexpectedahighdegreeofcorrelationacrosstimeand/orspace,whichmakes morestandardolsregressioninappropriate.second,ourdependentvariable "riot,"whichcorrespondstothedefinitionabove,isacountofthenumberofriotsin eachadministrativedistrictforeveryyearbetween1971and2000.asastraight countvariablethatisstronglyskewedtotheright,theappropriatestatistical 1 The districts have been standardized to the 1971 boundaries in order to allow crosssectional time series regressions in our analyses. In less than five cases, this has involved about a 5% error margin when reaggregating the figures. 15
techniqueisbinomialregression. TheindependentvariablescanbegroupedintosocialScultural,political,and economicfactorsasspecifiedinthehypotheses.thesocialsculturalvariablesare intendedtocaptureculturallysalient,identitysbasedinfluencesandaremeasured bythepercentageofadistrict'spopulationthatismuslimandmembersofa SCHEDULEDCASTE.Thehypothesisisthatbothofthesefactorswillcontributeto increasedlikelihoodofriotsoccurringbecausethemoreindividualstherearewith ethnicattachments,themorelikelytherearetobesufficientpeoplewith preferencesthatarehighenoughtoprovideincentivestoriot. 2 Inaddition,we estimatemodelsthatsubstitutetheratioofhindustomuslimsinadistrict (HINDU/MUSLIMRATIO)toexplorethepossibilitythatthepropensitytoriotis contingentonthepoportionofhindustomuslims. Weincludeeconomicvariables,whicharetakenfromtheCensusofIndia's decennialreports.thevaluesareinterpolatedfortheinterveningyears. INDUSTRIALWORKERisthepercentageofadultsemployedintheorganized nonsagriculturalsector.thiscategoryincludesindustrialworkers,artisans,and professionalworkers.weusethisvariableasaroughproxyforindustrialization anddevelopment.whilewewouldliketoincludeliteracyasavariableforsimilar reasons,itistoohighlycorrelatedwithothervariablesofinterest. AGCULTistheratioofthenumberofindividualswhoarelandlesslaborersto thenumberwhoarecultivatorsineachdistrict.thisvariableisameasureofland 2 The basis for this hypothesis are derived from Dumont 1980, Beteilled 1983, Ludden 1996, and Srinivas 1962, each of whom argues that individuals who are Muslim or scheduled caste have especially marked preferences for affective attachment. 16
inequalitythatrisesindistrictswhererelativelyfewindividualshaveproperty rightsoverthelandanddipsindistrictswheremostindividualshavesuchproperty rights.weincludeitbecausevariousstudiesofdomesticpoliticalviolencehave emphasizedtheeffectsofinequalityintheownershipofland(wangetal.1993, MullerandSeligson1987,Midlarsky1989,Brockett1992).Thevariablealso capturesanindiasspecificaspectofsocioseconomicrelationsbecauselandless laborersaremorelikelytobelowcaste,whilecultivatorsaremorelikelytobelong tothehighandmiddlecastes(withtheexceptionofkarnataka,wherelowcastes arefrequentlycultivatorsandoccupyadominantsocioseconomicposition).inthe absenceofreliablecastedata,thisvariableallowsustohavearoughproxyforrural casterelationshipsandtheircontributiontoriotevents. DENSITYisthepopulationdensityofeachdistrictforeveryyear.Itisthe ratioofadistrict'spopulationtoitstotalgeographicalarea(measuredinsquare kilometers).weincludethisvariablebecausetherehasbeenconsiderableinterest expressedinwhetherindianriotsaredisproportionatelylikelytooccurineither urbanorruralareas.communalriotsinparticularareconsideredtobeprimarily urbanphenomena(varshneyandwilkinson1995,engineer1984,aiyar1967).if thevarshneyandwilkinson(1995)andwilkinson(2004)resultsholdforthis dataset,weexpecthigherlevelsofdensitytocontributetohigherratesofriots. Finally,MIGRATIONisthepercentageofthemalepopulationofadistrictfor eachyearthatwasbornoutsidethatdistrict,afactorthatisconsideredtobea reasonablygoodmeasureofmigrationforemployment(sinha1988).theexisting literatureindicatesthatincreasedinsmigrationshouldincreasetheprospectofriots. 17
Qualitativestudieshavepointedtotheroleofmigrantsinfomentingriotsand migrantneighborhoodsarefrequentlyriotepicenters(times&of&india1992),and interviewswithvictimsofriotviolencehavereiteratedtheroleofrecentmigrants askeyparticipantsinthatviolence(parikhinterviews:surat,gujarat,1995). Weincludeanumberofpoliticalvariables.NATIONALELECTIONisa dichotomousvariablethatindicatestheyearsinwhichnationalparliamentary electionswereheld.theelectionyears,forwhichanationalelectionoccurredare codedas'1'andallothersare'0,'are1971,1977,1980,1984,1989,1991,1996, 1998and1999.STATEELECTIONisasimilardichotomousvariablethatindicates yearsinwhichstateelectionswereheldandthoseyearsvarybystate.turnoutis thenumberofvotersdividedbythetotalnumberofelectorsforeachparliamentary election.marginisthepercentageofvotescastforthewinningcandidateminus thepercentageofvotescastforthesecondsplacecandidateinparliamentary elections.incumbentpartyisadichotomousvariablethatindicateswhetherthe partyinpowerwasreelectedintheyearofanationalelection.weexpectthis variabletobecorrelatedwithriots,butwedonothaveapredicteddirection. RESULTSANDDISCUSSION WepresenttheresultsofouranalysisinTable1.Thefirstmodelpresentsthefull 1971S2000dataset,withandwithoutalaggeddependentvariable.Whilethereis debateoverthedesirabilityofincludinganldvinpaneldataanalysis,inthiscase thereisatheoreticalreasontoincludeit:argumentsintheliteraturepointtothe importanceofriotentrepreneursor riottenders, inbrass sphrase(brass2004). 18
Moreover,decisionstojoinrioteventscanbeshapedbyinformationaboutprior events. Inthisspecification,thevariablesforScheduledCasteispositiveand significant,aspredicted,bothwithandwithouttheldv.thepoliticalvariablesthat arepositiveandreachsignificancearebjp,turnout,andnationalelection,butthey arenotsignificantwhentheldvisincluded.hindu/muslimratioisnegativeand significant,whichindicatesthatthegreaterthenumberofmuslimsinadistrict relativetohindus,thelowerthelikelihoodofriots.thisresultisconsistentwith Wilkinson sresults,whichfindthatcompetitionformuslimvotesreduces communalriots,ifweassumethatlargernumbersofmuslimsbecomeattractiveas voteblocs.densityisnegativeandsignificant,whichrunscountertovarshneyand Wilkinson(1995)aswellasWilkinson(2004),whichfindthatcommunalriotsare morelikelytooccurinurbansettings.butthisdivergencecouldbedueeithertothe data slikelyoverinclusionofsmallriotsoritsinclusionofallriots,notjustthose thatarecommunalinorigin.marginandstateelectionfailtoreachsignificance. ThenextthreemodelspresenttheresultsofdecadeSspecificpanelanalyses, withandwithouttheldv.themoststrikingaspectsoftheseresultsarethe differencesacrossthepanels.ourassessmentthatthenarrativeofexplainingriots changesacrossdecadesoffersapotentialreason,butexaminationoftheresults indicatesthatallthreedonotconformtothethreenarrativesweoutlined. The1970sconforminparttothepoliticalandscholarlynarratives:they weredescribedaseconomicinnature(i.e.,breadriots),andtheresultsreflectthis. Agcultispositiveandsignificant,asisDensity,andaccountsofriotsemphasizethat 19
economicriotsoccurredinbothurbanandruralsettings.ahigheragcultratioalso reflectsalargerlowscastepopulation,whichcouldindicateeithereconomicor affectivedissatisfaction.similarly,scheduledcasteispositiveandsignificant,which clearlyrepresentsanaffectiveidentitycategorybutcannotberuledoutasbeing motivatedbyeconomicissues.thesignificanceanddirectionofthesetwovariables, AgcultandScheduledCaste,lendcredencetoourtheorythateconomicconditions canhaveaffectivevalence,butwecannotbepositiveusingonlythistypeof observationaldata. Severalofthepoliticalvariablesreachsignificance,buttheyarenotallinthe samedirection.incumbencyandnationalelectionarenegativeandsignificant, indicatingthatdistrictsinwhichincumbentswereretainedandyearsinwhich nationalelectionsoccurredwerelessriotous.butturnoutispositiveand significant,indicatingthatdistrictswithmoreengagedvotershadhigherlevelsof riotevents.noothervariablesreachsignificance.ineverycase,variablesare consistentlysignificant(orlacksignificance)bothwithandwithouttheldv. The1980sresultsareweakeranddonotsupporttheprevailingnarrative. Despitetheemphasisoncasteconflictinthisdecade,noneoftheaffectivevariables thatincorporatecastereachsignificance.theonlyvariablesthatreachsignificance aremargin,whichisnegative(i.e.,closercontestsarepositivelycorrelatedwith riots),andstateelection,whichispositive. Finally,the1990sresultsrunsomewhatcountertotheprevailingnarrative thatcommunalriotsdominatedtheindianlandscape.moststrikingly,thebjp variableisnegativeandsignificant,indicatingthatthegreaterthebjpvote,the 20
lowertheincidenceofriots.thismayreflectthebjp sdesireinthe1990stowin electionsbyappealingbeyondtheirbase,butitdoesnotsupporttheconventional wisdomthatbjpstrongholdsweremorelikelytoberiotsprone.inaddition, Migration,whichhasalsobeenlinkedwithincreasedriotpropensityinpoliticaland scholarlydiscussions,isnegativeandsignificant,i.e.,increasedmigrationis correlatedwithlowernumbersofriots.thenationalelectionsvariableisnegative andsignificant,i.e.,yearsinwhichnationalelectionstakeplace(therewerethreein the1990s)arecorrelatedwithlowerriottotals.turnoutisnegativebutsignificant onlywithouttheinclusionoftheldv.finally,thehindusmuslimratioispositiveand significant,indicatingthatthegreatertheproportionofhindusovermuslims,the greaterthenumberofriots. CONCLUSION Theseresultsaretentativeandpreliminary.Weintendtoextendouranalysisto includemultislevelmodelsthatcangiveusmoredetailedresultsaboutstateslevel differences,andwewillcontinuetoperformrobustnesschecks.buttherearesome intriguingfindings.theconsistentlynegativeresultsforhypothesizedandwidely acceptedrelationshipsbetweenbjppoliticalsuccessandincreasedviolence suggeststhatatleastsomeoftheconventionalwisdommightneedtobe reconsidered.thestrengthofcasteandeconomicvariables,particularlythe relationshipsbetweenscheduledcastes,landlesslaborers,andincreasedriots,point tothesalienceofeconomicconditionsandcasteidentity.andfinally,whilethe politicalandscholarlynarrativesseemtobeonpointduringthe1970s,theyareless 21
reflectiveofonthegroundconditionsinthefollowingtwodecades,atleastforthese states.giventhatthesampleincludesmanyofthecommunalandpolitical flashpointsofthelast30years,however,thisdisconnectshouldgiveuspause. 22
Table 8: Neg. Binomial Regressions with District Fixed E ects Full Full Model 1970 s 1970 s 1980 s 1980 s 1990 s 1990 s Model w/ LDV w/ LDV w/ LDV w/ LDV (Intercept) 5.63 4.81 8.95 10.03 7.75 7.39 8.36 7.08 (0.20) (0.19) (5.27) (5.40) (0.54) (0.54) (0.60) (0.58) Margin 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) BJP 0.13 0.02 0.21 0.43 0.18 0.19 0.28 0.26 (0.06) (0.06) (0.39) (0.39) (0.17) (0.17) (0.09) (0.08) Agcult 0.55 0.45 2.30 2.19 0.00 0.04 0.23 0.08 (0.09) (0.08) (0.67) (0.70) (0.33) (0.33) (0.13) (0.12) Incumbent Party 0.07 0.06 0.15 0.16 0.01 0.01 0.03 0.03 (0.02) (0.02) (0.05) (0.05) (0.03) (0.03) (0.02) (0.02) Turnout 0.26 0.19 1.00 0.87 0.10 0.07 0.33 0.15 (0.11) (0.10) (0.36) (0.37) (0.32) (0.32) (0.14) (0.13) Density 0.00 0.00 0.01 0.01 0.00 0.00 0.00 0.00 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) State Election 0.02 0.01 0.01 0.00 0.06 0.06 0.01 0.00 (0.02) (0.02) (0.03) (0.03) (0.02) (0.02) (0.02) (0.02) Nat l Election 0.05 0.02 0.21 0.21 0.00 0.00 0.05 0.04 (0.02) (0.02) (0.03) (0.05) (0.02) (0.02) (0.02) (0.02) Hindu/Muslim Ratio 0.00 0.00 0.10 0.13 0.00 0.00 0.01 0.01 (0.00) (0.00) (0.08) (0.08) (0.00) (0.00) (0.00) (0.00) % Indus. Worker 0.09 0.13 31.85 44.50 2.61 2.49 0.13 0.02 (0.21) (0.19) (28.05) (28.74) (2.03) (2.00) (0.20) (0.19) % Sched. Caste 7.92 4.40 30.48 33.75 0.08 0.11 4.35 4.74 (0.71) (0.68) (12.35) (12.63) (2.74) (2.70) (3.31) (3.16) % Migrants 1.67 1.46 99.70 129.08 0.82 1.22 2.93 2.77 (0.88) (0.84) (72.16) (73.88) (3.12) (3.10) (1.13) (1.09) (0.11) (0.13) (0.27) (0.31) (0.30) (0.31) (0.42) (0.47) LDV 0.00 0.00 0.00 0.00 (0.00) (0.00) (0.00) (0.00) Standard errors in parentheses Models do not include literacy variable. indicates significance at p<0.05 9