英文【世界銀行】沖突環(huán)境中的災(zāi)害影響

PublicDisclosureAuthorizedPublicDisclosureAuthorized

PolicyResearchWorkingPaper10995

ImpactsofDisastersinConflictSettingsEvidencefromMozambiqueandNigeria

KarimaBenBih

ChloeDesjonqueres

BramkaJarafino

ElodieBlanc

SoleneMasson

WORLDBANKGROUP

Urban,DisasterRiskManagement,ResilienceandLandGlobalDepartmentDecember2024

PolicyResearchWorkingPaper10995

Abstract

Thispaperestimatesthedifferentiatedeconomicimpactofnaturalhazard-relateddisasters(thespecificdisastersandclimateshocksstudiedherebeingfloods)whentheyoccurinconflictversusnon-conflictaffectedareas.Existinglit-eratureshowsthatdisastersandclimateshockscancausesignificantdistresstocountriesandpeopleonaninstitu-tionalandhouseholdlevel.However,assumptionsaremadethattheirimpacttendstobelargerinconflict-affectedareas,withlittleevidenceavailableonthedifferentiatedextentofthesedamages.Thispaperinvestigateswhether,andtowhatextent,thepresenceofconflictshasamplifiedtheimpactsoffloodsoneconomicactivityandpeople,andhamperedrecovery.Thepaperappliesa“top-down”approachtoesti-matingthedifferentialimpactsofdisastersandclimate

shocksbetweenconflictandnon-conflictaffectedareasusingsatellite-derivedimageryofnightlightradianceasaproxyforeconomicactivity,alongwithgeospatialfoot-printsoffloods.Theanalysisconsiderstwocasestudies:the2019tropicalcyclonesIdaiandKennethandsubsequentfloodsinMozambique,andtheJuly2022floodsinNige-ria.Usingdifference-in-differenceestimations,theanalysisfindsthattherearesignificantdifferencesindisasterandclimateshockimpactsandrecoverybetweenconflictandnon-conflictaffectedareas.Particularly,thereisagreaterdeclineineconomicactivityandalongerrecoverytimeinconflictaffectedareas,asproxiedbythegreaterchangeintheintensityofnightlightradiance.

ThispaperisaproductoftheUrban,DisasterRiskManagement,ResilienceandLandGlobalDepartment.ItispartofalargereffortbytheWorldBanktoprovideopenaccesstoitsresearchandmakeacontributiontodevelopmentpolicydiscussionsaroundtheworld.PolicyResearchWorkingPapersarealsopostedontheWebat

/

prwp.Theauthorsmaybecontactedatkbenbih@.

ThePolicyResearchWorkingPaperSeriesdisseminatesthefindingsofworkinprogresstoencouragetheexchangeofideasaboutdevelopmentissues.Anobjectiveoftheseriesistogetthefindingsoutquickly,evenifthepresentationsarelessthanfullypolished.Thepaperscarrythenamesoftheauthorsandshouldbecitedaccordingly.Thefindings,interpretations,andconclusionsexpressedinthispaperareentirelythoseoftheauthors.TheydonotnecessarilyrepresenttheviewsoftheInternationalBankforReconstructionandDevelopment/WorldBankanditsaffiliatedorganizations,orthoseoftheExecutiveDirectorsoftheWorldBankorthegovernmentstheyrepresent.

ProducedbytheResearchSupportTeam

ImpactsofDisastersinConflictSettings:EvidencefromMozambiqueandNigeria

Novembre20,2024

KarimaBenBih,WorldBank

ChloeDesjonqueres,WorldBankBramkaJarafino,WorldBank

ElodieBlanc,MotuEconomicandPublicPolicyResearchCenterSoleneMasson,WorldBank

Keywords:EconomicImpactsofDisastersinCon?ict,Climateshocks;EarthObservations;NPP-VIIRS;Floods;Nigeria;Mozambique;Con?ictsin?uenceonDisasterImpactsandRecovery;GDP.

JELClassi?cation:D74;O23;O47;O57;Q34;Q54.

TheauthorsaregratefultoStephaneHallegatte,OscarIshizawa,andJunRentschlerfortheirthoughtfulcomments,suggestions,andguidance.

2

Introduction

Theaimofthisstudyistoexaminethedifferentialimpactofdisastersandclimateshocksonpopulationsincon?ict-affectedregions,speci?callyinvestigatingtherepercussionsof?oodingincon?ictversusnon-con?ictareas.Usingremotesensingtechnology,weattempttoovercomethechallengeofdatascarcityincon?ict-affectedcountries,allowingustoaccountforshort-termimpactsofrecentdisasterandclimateshockevents.Despitetheinherentlimitationsofusingnightlightintensityasaneconomicactivityindicator,itprovidesanempiricalfoundationfortheanalysisandenoughobservationsforanex-postquasi-experimentalimpactevaluation.Weemployadifference-in-differenceeconometricapproach,usingsatelliteimageryofnightlightradiancealongsidegeospatialdataon?oodandcon?ictevents.ThismethodologicalframeworkisappliedtoassesstheaftermathoftheMarch-April2019FloodsinMozambiquefollowingCyclonesKennethandIdai,aswellasthe2022?oodsspanningJulytoOctoberinNigeria.

Resultsshowsigni?cantdisparitiesintheeffectsofdisastersandclimateshocksbetweencon?ict-affectedandnon-con?ict-affectedareas.Speci?cally,weobserveamorepronounceddeclineineconomicactivitiesincon?ict-affectedregions.

Thepaperisstructuredasfollows.The?rstsectionoutlinesthecontextof?oodandcon?icts.Itpaysattentiontotheinterconnectednessofcon?ictanddisastersandclimateshocks,outliningthemethodologyandempiricalstrategyderivedtoestimatesuchex-postimpact.Inthesecondsection,wepresenttheresultsandsupportingdataderivedfromthestudy,includingthecasestudiesonMozambiqueandNigeria.Finally,wediscusslimitationsaswellasbroaderimplicationsbeforeconcluding.

Context:Floodimpactandcon?ictaffectedpopulation(Literature)

1.Impactof?ood

Quantitativeeconomicanalyseshavefrequentlyusednightlightradianceasproxyforeconomicactivity(Chen&Nordhaus,2011;Hendersonetal.,2012).Thesehavealsobeenusedtoestimatetheimpactsofweathervariabilityanddisastersandclimateshocks(Bertinelli&Strobl,2013;Elliottetal.,2015;Felbermayretal.,2022;Heger&Neumayer,2019;MirandaMonteroetal.,2017)and,morespeci?cally,?oods(Kocornik-Minaetal.,2020).Mostanalysesusingnightlightdatausuallydemonstrateanegativeimpactofdisasterandclimateshocksonnightlightsbutwitheffectsresorbingwithintheyearfollowingtheevent(Bertinelli&Strobl,2013;Elliottetal.,2015;Gillespieetal.,2014).Schippers&Botzen(2023)?ndthatforaseveredisastersuchasHurricaneKatrina,theeffectcanbelongerlasting.

However,thereisadebateabouttheaccuracyofnightlightsasaproxyforeconomicactivity.Criticsarguethatnightlightintensitymaynotcaptureeconomicactivityaccuratelyinallcontexts,suchashighlyruralareas,wherechangesinlightingefficiencycouldaffecttheamountoflightobservedwithoutnecessarilyre?ectingchangesineconomicactivity.Possiblyotherculturalandsocialfactorsorgovernmentpoliciesonlightingcouldalsoin?uencetheamountofnightlightobserved.

3

Despitetheseconcerns,nightlightshaveseveraladvantagesasadatasource.Theyaregloballyavailable,providingcoverageeveninregionswhereeconomicdatamightbescarceorunreliable.Nightlightsalsohaveastandardspatialresolutionandtimeintervals,whichallowsforconsistentcomparisonsovertimeandacrossdifferentgeographicareas.Whenprocessedandinterpretedcorrectly,takingintoaccountthepotentiallimitationsandbiases,nightlightdatacanindeedserveasausefulproxyfortheintensityofeconomicactivity(Gibsonetal.,2021).

2.Relationshipbetweendisasterandcon?ict-affectedpopulation

Explicitstudiesoftherelationshipbetweendisasterandclimateriskandcon?icthavegainedtractionoverthepastdecade(Siddiqi,2018),speci?callyfocusingonco-locationandcausationdebatesassociatedwithclimate-relatedhazards,violentandarmedcon?ict,andinsecurity(Gemenneetal.,2014;Gleditsch,2012).Often,previousstudieshavefocusedontheimpactsofdisastersoncon?icts–whethertheyexacerbateexistingcon?icts,ignitenewones,orinsomecaseshaltongoingcon?icts(Nel&Regharts,2008;Schleussneretal.,2016;Slettebak,2012;Ghimireetal.,2015;Nardullietal.,2015).Duetosuchuncertainimpactsofdisastersanddisasterrecoveryeffortoncon?icts,otherstudiesexplorehowdisasterriskreductionandrecoverymeasuresshouldbedonedifferentlyincon?ictcontexts(Brzoska,2018;Petersetal.,2019;WorldBank,2016).

Despitethegrowingbodyofliteraturerelatedtotheintricaciesofdisastersandcon?icts,lessattentionhasbeengiventounderstandingandquantifyingthein?uencesofcon?ictsondisasterimpacts–theadditionaleconomicimpactsofdisastersshouldtheytakeplaceincon?ictareasanditseffectoncon?ict-affectedpopulation–aswellasthecausalpathwaysandmechanismsbehindsuchadditionalimpacts.Theabsenceofcomprehensiveeconomicdataandgroundtruthdatatovalidatedisasterimpacts,coupledwiththecomplexityofde?ningcon?ict-affectedpopulationsareamongthescienti?cchallengesprohibitinganalyzingthein?uenceofcon?ictsondisasterimpacts.Thispaperseekstoaddressthisgapandsupportfurtherquantitativeanalysesontheadditionalimpactonhouseholds’welfareandnations’economicgrowthincountriesexperiencingthesecompoundedcrises.

EmpiricalstrategyData

Inthisstudy,weusepixel-levelgeospatialdata,includingnightlights,?oodfootprints,populationdensity,andadministrativeboundaries,toeconometricallyanalyzethespeci?ceffectsof?oodeventsinMozambiqueaswellasNigeria'scon?ictandnon-con?ictaffectedregions.

Nightlightsdata

Furthermore,weutilizecompositeimagesofnighttimeradiancedatacapturedbytheVisibleInfraredImagingRadiometerSuite(VIIRS)sensoraboardtheNASA-NOAASuomisatellite.Thesemonthlycompositesareavailablesince2012ataresolutionof15arcsecondsby15arcseconds(approximately463metersattheequator).VIIRSDayNightBands(DNB)dataexcludegridcellsaffectedbylightning,straylight,lunarillumination,andcloudcover(Elvidgeetal.,2017).WefavorVIIRSdataovertraditionallyuseddatafromtheDefenseMeteorologicalSatelliteProgram(DMSP)

4

duetoseverallimitationsidenti?edinthelatter,includingblurring,lackofcalibration,top-coding,andpoorsuitabilityasaGDPproxyinruralareas(Gibsonetal.,2021).

ToaddresschallengesassociatedwithusingVIIRSnightlightsdataasaproxyforeconomicactivity(Skou?asetal.,2021),weapply?lterstoremovepixelswithextremevalues(i.e.,werestrictthesampletovaluescomprisedbetweenthe1stand99thpercentiles)andaccountforthenumberofobservationsavailableperpixel.

1

Wecalculatetheaveragenightlightradiancemonthlyspanningfrom1to6monthsbeforeandaftertheoccurrenceofthe2019?oodsinMozambiqueandthe2022?oodsinNigeria.Twovariablesarecomputed:"avg_rad,"representingthenightlightradianceatthe?oodedpixellevel,and"avg_radBuff05"whichaveragesforeach?oodedpixelthenightlightradianceofthepixelitselfandadjacentpixelswithina0.5-kilometerbuffer.

2

Thelattervariableispreferredtoensuremaximumobservationavailabilityandtocapturetheimpactonindirectlyaffectedgridcells.

3

Wealsoextractedtheassociatedvariables"cf_cvg"and"avg_cvgBuff05",whichindicatethenumberofcloud-freeobservationsinthemonthusedtocalculateaveragenightlightradiance.

4

Flooddata

FloodeventsaredeterminedbasedonthemethodologyoutlinedbyDeVriesetal.(2020).WeuseS1GroundRangeDetectedscenesfromtheSyntheticApertureRadarsensorsonboardtheSentinel-1satellite,partoftheEuropeanSpaceAgency'sCopernicusprogram(ESA,2023).ThesescenesprovidedataonZ-scoresderivedfromSARbackscattertimeseriesofsinglebandco-polarizationverticaltransmitverticalreceive(VV)anddualcross-polarizationverticaltransmithorizontalreceive(VH).SinceOctober2014,thisdatahasbeenavailableevery6daysata10-meterresolution.

Floodsarede?nedastheunexpectedpresenceofwaterobservedinanygivenpixel.Todistinguish?oodsfrompermanentorseasonallyoccurringsurfacewater,weutilizethehistoricalLandsat-derivedmonthlywaterprobabilitiesdatasetproducedbytheEuropeanCommission’sJointResearchCentre(Pekeletal.,2016).Floodcon?denceiscategorizedashighifbothVVandVHZ-scoresfallbelowtheidenti?edthresholds,andasmediumifonlyoneofthesepolarizationsisbelowthethresholds.Weclassify?oodsinareasnotdesignatedaspermanentopenwater(withaprobabilityofwatergreaterthan95%)orwithahistoricalinundationprobabilitylessthanorequalto25%.Foreachcasestudy,wepreselectahistoricalreferenceperiodbasedonexistingknowledgeofpast?oodingeventsintherespectivearea.

Con?ictdata

Con?ictareasareidenti?edutilizinggeocodeddatasourcedfromtheArmedCon?ictLocation&EventDataProject(ACLED)database(ACLED,2023),coveringtheperiodfromJanuary2012toDecember2023forNigeriaandfromJanuary2016toDecember2023inMozambique.Forthepurposeofthisstudy,con?ict,asde?nedbytheWBG(2024)is,“astateofacuteinsecurityresultingfromtheuseoflethalforcebyagroup—encompassingstateforces,organizednon-stateentities,orotherirregularbodies—drivenbyapoliticalpurposeormotivation.Suchforcemaymanifest

1Pixelswithnocloud-freeobservationsareexcluded.

2Apixelisaround100m2,wetestedwithoutbuffer,500mand1kmandchosea500mbuffertointroducemorevariationofnightlightintensitywithinfloodedpixels.

3Atimeseriesdepictingbothvariablesisprovided

inFigure16

intheappendicesforNigeria.

4Thecorrespondingtimeseriesforispresented

inFigure17for

Nigeria.

5

bilaterally—involvingengagementsamongmultipleorganized,armedfactions,occasionallyleadingtocollateralcivilianharm—orunilaterally,whereinagrouptargetsciviliansdeliberately.”Furthermore,forthemostprecisedepictionofareasseverelyaffectedbycon?ict,fatalitiesstemmingfromprotests,riots,andstrategicdevelopment(asperACLEDdata)havebeenexcluded,maintainingconsistencywiththeWBGClassi?cationofFragilityandCon?ictSituation’s(FCS)objectivesandthescopeofthisstudy.Ouranalysisfocusesoncon?ictrecordscategorizedas‘Battles’,‘Explosions/Remoteviolence’,and‘Violenceagainstcivilians’.Thesetypesofcon?ictsareselectedduetotheirviolentnature.

Settlementdata

Todeterminetheurbanizationlevel,weusetheGlobalHumanSettlementLayer(GHSL)whichcombinesgriddedpopulationdataestimatedbyCIESINGPWv4.11GHS-POPR2023andbuilt-upsurfaceinformationfromLandsatandSentinel-2dataGHS-BUILT-SR2023(Schiavinaetal.,2023).

5

Thesettlementdataareavailableatthe1kmresolution.Weconsiderthedatafortheyear2020,whichistheclosestavailabletothetimeperiodofinterestforbothcountries.IncaseofNigeria,wede?ned‘urban’areasascellsde?nedashigh-densitycluster,

6

‘suburban’asmoderate-densitycluster,

7

‘rural’asruralandlow-densityclusters

8

and‘Uninhabited’asverylowdensityruralandwatercoveredareas

(Figure1)

.

9

5InGoogleEarthEngine,thisImagecollectionisaccessiblethrough

/earth

-engine/datasets/catalog/JRC_GHSL_P2023A_GHS_SMOD.

6The‘urban’categoryincludestheclasses30:“UrbanCentregridcell”,23:“DenseUrbanClustergridcell”.

7The‘suburban’categoryincludestheclasses22:“Semi-denseUrbanClustergridcell”and21:“Suburbanorperi-urbangridcell”.

8The‘rural’categoryincludestheclasses13:“Ruralclustergridcell”and12:“LowDensityRuralgridcell”.

9The‘uninhabited’categoryincludestheclasses11:“Verylowdensityruralgridcell”and10:“Watergridcell”.

6

Figure1.SettlementcategoriesinNigeria

Populationdata

Tomaintainconsistencywithsettlementdata,populationdensityestimatesatthegridcelllevelaresourcedfromtheHigh-ResolutionSettlementLayer(HRSL)dataset(FacebookConnectivityLabandCenterforInternationalEarthScienceInformationNetwork-CIESIN-ColumbiaUniversity.,2016).

Thesedataareavailableataresolutionof1arc-second(approximately30meters)fortheyear2020.Additionally,alternativepopulationdataareextractedfromtheWorldPopdatabase(Linardetal.,2012;WorldP,2024),availableataresolutionof100metersfortheyear2020.

Table1

belowdescribesthevariablesusedintheanalysisof?oodandcon?ictimpactsoneconomicactivity,asmeasuredbynightlightchanges.The‘lat’(latitude)and‘lon’(longitude)variablesallowforlocationmappingandsituatingtheanalysisspatially.The'months_EE'variableaidsinunderstandingthetemporaleffectsof?oodsbyindicatingmonthsaftertheeventandnegativevaluesindicatingthemonthspreceding.

7

The'?ood'variableiscrucialforassessingtheimpactof?oodswithvaryingdegreesofdatareliability.'PopDens'providesinsightsintohowpopulationdensitymightin?uenceorbein?uencedbyspeci?c?oodevents.Thevariables'cf_cvgBuff05'and'avg_radBuff05'describedabove,measureeconomicactivitythroughcloudobservationsandnightlightradiance,respectively.Additionally,theyareaveragedoveradjacentpixelstoprovidecontextforeachlocation.

‘Treated'and'Treated_after'distinguishareasaffectedbycon?ictbeforeandafterthe?oodineachcountry:March-April2019aremonthswherethe?oodoccurredinMozambiquewhileJuly2022wasconsideredasthe?oodingmonthinthisanalysisforNigeria.Thisisessentialfordeterminingthecausalinferenceofcon?ictimpact.'Settlement'and'Urban_Suburban'categorizeurbanizationlevelstounderstandhowdifferenttypesofareasareaffectedbyandrespondtoboth?oodsandcon?ictevents.Lastly,'Fatalities'providesadirectmeasureofthehumancostofcon?icts.

Thesevariablescollectivelyenableacomprehensiveanalysisoftheeffectsof?oodsandcon?ictswhentheyco-occurinthesamelocation.Theyareusedtoanalyzetheimpactsof?oodsandcon?ictonspeci?caspectsofeconomicactivity,aspotentiallyinferredbynightlightchanges.

Table1

describesthesevariables,theirunitsofmeasurements,andhowtheywillbeusedintheanalysis.

Table1.Descriptionofvariables

Variables

Description

Unit

lat

latitude

Decimalcoordinates

lon

longitude

Decimalcoordinates

months_EE

Monthsincetheevent

Months(positiveifafterevent,negativeifbeforeevent)

flood

Floodvariable

=1or2ifmediumreliability,=3ifhighreliability

PopDens

HRSLpopulationdensity

Person/km2

cf_cvgBuff05

Totalnumberofcloud-free

observationsthatwentintoeach

pixel(averagedovertheadjacent

pixels)

avg_radBuff05

Averagenightlightradiancevalues(averagedovertheadjacentpixels)

nanoWatts/sr/cm2

Treated

Dummyvariablerepresenting

conflict-affectedarea

=1ifsubjecttoaconflictwithinthebufferareabeforethefloods,=0otherwise

Treated_after

Dummyvariablerepresenting

conflict-affectedarea

=1ifsubjecttoaconflictwithinthebufferareaafterthefloods,=0otherwise

settlement

DegreeofUrbanization

=11ifuninhabited,=12ifrural,=21ifsuburbanand=23ifurban

Urban_Suburban

Urbanizationdummyvariable

=1ifurbanorsuburban,=0otherwise

Fatalities

Totalnumberoffatalitiesassociatedwithconflictswithinthebufferarea

Overallempiricalstrategy

Todifferentiatetheimpactof?oodsoneconomicactivitypriortothe?oodbetweencon?ict-affected(treatmentgroup),andnon-con?ictaffected(controlgroup)areas,we?rstrestrictthesampleto?ood-impactedpixels.Wethenapplythedifference-in-differencesregressionmethod,aquasi-experimentaltechniquecommonlyusedforex-postimpactevaluations.Theunderlyingconcept

8

involvescomparingtwogroupsovertime.Duetotheirdistinctcharacteristics,weexpectdifferencesinoutcomesbetweenthegroups.However,theevolutionofthesedifferingoutcomesovertime,whileholdinggroupcharacteristicsconstant,shouldfollowasimilartrend(i.e.,thecommontrendassumption)untilanexogenousshockoccurs.Thepresenceofthisparalleltrendiscrucialforestablishingcausalevidenceofimpact.Thedifference-in-differencesresearchdesignisparticularlysuitablefor‘event’studiesandthequanti?cationoftheimpactofunexpectedshocksoneconomicoutcomes.Thismethodhasbeenextensivelyemployedinthereviewedliterature(Card&Krueger,2000;Galianietal.,2005).Inourcasestudies,weareusingthecanonicaldifferenceindifference,whichmeanstwogroupsandtwotimeperiods(beforeandafter).

Thedifference-in-differenceregressionisspeci?edasfollows:

yi,t=Treatediβ1+postperiodtβ2+Treatmenti,tβ3+covariatesi,tβ4+εit(1)

whereyi,tistheaveragelogofnightlightsdataforeach?oodedpixeliattimet.Theuseofremotesensingdataallowsustoexploreimmediatetoshorttermimpactofthe?oodoverourtwogroups.Treatediisadummyvariableequalto1for?oodedpixelilocatedwithinthecon?ictbufferzonebeforethe?oodevent,andto0for?oodedpixelilocatedinanon-con?ictaffectedareabeforethe?oodevent.postperiodtisadummyvariablethatrepresentstheperiodaftertheexogeneousshock.

10

Treatmentitisthetreatmentvariable,i.e.,thevariableofinterestinadifference-in-differencespeci?cationwhichaccountsfortheinteractionofthetreatedandPostperiodvariable;covariatesitisthesetofadditionalexplanatoryvariablesuspectedtoimpactthelevelofnightlightsradiance;andεitisanindependentandidenticallydistributederrorterm,clusteredattheadministrativelevel2,toavoidspatialautocorrelation.

Oneofthechallengesinouranalysisisthede?nitionofcon?ict-affectedareas.Con?icteventsinMozambiqueareclusteredgeographicallyassuchthatitisreadilyascertainablewhatregionsaremostimpactedbytheseevents,andthusde?nedascon?ict-affectedareasforthepurposeofthisstudy.

Duetothecomplexityandwidegeographicalspanofviolentandnon-violentcon?icteventsinNigeria,con?ict-affectedareasinNigeriaarenotde?nedbasedonnumberofeventsalonebutarebasedontheWBGFCScon?ictclassi?cation.Thisclassi?cationusespubliclyavailabledatatoannuallyassesscountries,pinpointingthosemostaffectedbyfragilityandcon?ict.ThismethoddifferentiatesbetweenterritoriesexperiencingFragilityand/orCon?ictsituations.Alignedwiththisde?nition,thestudyemploysthefollowingcon?ictindicatorsidenti?edbytheFCSindextodelineatecon?ict-affectedareasinNigeriaattheLocalGovernmentArea(LGA)scale:

10Thisvariabletakesavalueof1ift=August2020toestimatetheeffect1monthafterthedisaster,t=November2020fortheeffect3monthsafterthedisaster,etc.and0otherwise.

9

(1)Forongoingcon?ictaccordingtoACLED,(a)anabsolutenumberofcon?ictdeathsabove250,and(b)above2deathsper100,000population.

(2)ForrapidlydecliningsecuritysituationsaccordingtoACLED,(a)anabsolutenumberofcon?ictdeathsabove250,(b)between1and2deathsper100,000population,and(c)thenumberofcasualtieshasmorethandoubledinthepastyear.

Differenceindifferencemethodologyreliesondifferentassumptionswherethecommontrendisthemostimportantone.Tovalidatethisassumptionofcommontrendsbeforethe?ood,indicatingthatthedependentvariableforbothgroupswouldhavecontinuedmovingsimilarlyintheabsenceoftheextremeevent,weconductatestbycomparingchangesinthedependentvariableforthetreatmentandcontrolgroupsovermultipleperiodsprecedingthe?oods(i.e.estimatethedifference-in-differencebetweent-2andt-1,thet-3andt-2,etc.).Thisanalysishelpsascertainwhethertheeconomictrajectoriesofthetwogroupswereindeedparallelbeforetheoccurrenceofthe?oodevents.Theregressionisspeci?edas:

yi,t=Treatediβ1+postperiodtβ2+Treatmenti,tβ3+covariatesi,tβ4+εit(2)

Variablesarethesameasspeci?cation(1)buttheperiodofinterestisnotthesame.Weprovidestatisticaltestsaswellasthecommontrendvisuallyrepresentedforeachofourcasestudy.

Casestudyselection

Thestudy’sfocusondisasterimpactsincon?ictvsnoncon?ictaffectedareaslimitsthepoolofcountrycasestudies.Thereareseveralaspectsthatdeterminethechoiceofthecasestudycountries.First,theselectedcountriesneedtohavegeographicallylocalizedcon?icts,allowingforacontrolledcomparisonwherecon?ictistheprimarydifferingfactorofdisasterimpacts.Second,thereisarequirementthattheselectedcountrieswereaffectedbyarapidonsetdisasterinrecentyears,toincreasethepossibilityofavailabledatainassessingthedisaster’simpacts.Ifmultiplecountriesareselected,therapid-onsetdisastereventshadtohappenwithinthesametimeframe.Thiscriterionensuresthatthecasestudiesprovideafocusedexaminationoftheimpactofdisastersoneconomicactivitiesincon?ictsettingvsnoncon?ictsettings,withouttheconfoundingeffectsofdifferentcountryconditionsortimelinesofdisasterevents.Third,thedisasterfootprintsshouldcoverasubstantialgeographicalextentoftheselectedcountries’areaasopposedtolocalizeddisasters.Thiscriterionistoensurethattherearebothcon?ict-andnon-con?ict-affectedareashitbythedisaster.Giventhecriteriaabove,weselectedthe2019TropicalCyclonesIdaiandKennethinMozambiqueandtheJuly2022?oodsinNigeriaascasestudies.Furthermore,thetwocountrieshavecomparablecontextsintermsofcon?ictcharacteristicswhicharecrucialforisolatingthevariableofcon?ictincomparativeanalysis.

Mozambiquecasestudy:2019TropicalCyclonesIdaiandKenneth

The?rstcasestudyfocusesonthe?oodsinMozambiqueafterTCIdaiandKenneth.In2019,MozambiquewashitbytwoTropicalCyclones(TC),Idai(March4-15)andKenneth(April25-28),bothofwhichhavebeenquali?edasamongthestrongestTCsonrecordintheSouthernHemisphere(Charruaetal.,2021).Thenortheasternregionofthecountryischaracterizedbyawidespreadlong-

10

termhumanitariansituationduetotheongoingcon?ict,datingbackto2017.Asdepictedin

Figure

2,

TCIdai?rstmadelandfallonMarch4,2019,untilMarch9,beforechangingd