EectsEvidencefromSierraLeoneintheTimeofEbola OrianaBandieraNiklasBuehrenMarkusGoldsteinImranRasulAndreaSmurra y July2020 Abstract Schoolclosuresareacommonshortrunpolicyresponsetoviralepidemi ID: 953364
Download Pdf The PPT/PDF document "DoSchoolClosuresDuringanEpidemichavePers..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
DoSchoolClosuresDuringanEpidemichavePersistent E¤ects?EvidencefromSierraLeoneintheTimeofEbola ¤ OrianaBandiera,NiklasBuehren,MarkusGoldstein,ImranRasul,AndreaSmurra y July2020 Abstract Schoolclosuresareacommonshortrunpolicyresponsetoviralepidemics.Westudy Leone,acontextwherewomenfrequentlyexperiencesexualviolenceandfacemultipleeco- nomicdisadvantages.Wedosobyevaluatinganinterventiontargetingyoungwomenthat wasimplementedduringthe2014/15EbolaepidemicinSierraLeone.Thisprovidedthema protectivespacewheretheycan ndsupport,receiveinformationonhealth/reproductiveis- 4 700 younggirlsandwomenaged 12 to 25 trackedfromMay2014ontheeveoftheEbolacrisis,to thepost-epidemicperiodin2016.Incontrolvillages,schoolclosuresledyounggirlstospend signi cantlymoretimewithmen,teenpregnanciesrosesharply,andschoolenrolmentamong younggirlsdroppedby 17 e¤ectsonenrolmentarehalvedintreatedvillagesbecausetheinterventionbreaksthiscausal chain:itenablesgirlstoallocatetimeawayfrommen,reducesout-of-wedlockpregnancies by 7 pp,andsoincreasesre-enrolmentratespost-epidemic.Alongtermfollowupin2019/20 girls,timetheyspendwithmen,andqualityofpartnersmatchedwith.Ouranalysishas importantimplicationsforschoolclosuresinresponsetothecurrentCOVID-19pandemic incontextswhereyoungwomenfacesexualviolence,highlightingtheprotectiveandlasting rolesafespacescanprovideinsuchtimes. JELClassi cation:I25,J13,J24. ¤ Fernandezforexcellentresearchassistance.Wehavebene tedfromcommentsfromChrisBlattman,AureoDe Paula,EstherDu o,JamesFenske,EricaField,ElizabethFoster,SebastianGaliani,RachelGlennerster,Francesco Giovannoni,JessicaGoldberg,ScottMacMillan,BerkOzler,JamesRobinson,SanchariRoy,MatthiasSutter, UNICEFandtheWorldBankGroup sUmbrellaFacilityforGenderEqualityfor nancialsupport.Theviews presentedaretheauthors anddonotrepresentthoseoftheWorldBankoritsmembercountriesorDfID.This isanoutputoftheAfricaGenderInnovationLab.HumansubjectsapprovalwasobtainedfromtheIRBatIPA y Bandiera:LSE,o.bandiera@lse.ac.uk;Buehren:WorldBank,nbuehren@worldbank.org;Goldstein:World Bank,mgoldstein@worldbank.org;Rasul:UCL,i.rasul@ucl.ac.uk;Smurra:UCL,a.smurr
a.11@ucl.ac.uk. 1Introduction Low-incomecountriesaresusceptibletovariouskindsofaggregateshock,includingcommod- ityprice uctuations,con ict,climatechangeandviralepidemics.AsthecurrentCOVID-19 pandemichasstarklyillustrated,viralepidemicscancausesocietiestohavetorapidlyfacesi- multaneoushealthandeconomicchallenges[Rasul2020].Acommonpolicyresponseistoenforce socialdistancingmeasures,throughtravelrestrictionsandschoolclosures.Westudywhethertem- poraryschoolclosuresduringthe2014-16EbolaepidemicinSierraLeonehadpersistentimpacts ontheeconomiclivesofyoungwomen. WedosobyoverlayingtheEbolashockwitharandomizedcontroltrialevaluationofan interventiontargetedtoyoungwomen.Theintervention,providesyoungwomenasafespace(a club)atwhichtheycansocialize,andreceivelifeskillsandvocationaltraining.Bycombiningthe two,weprovidenovelinsightsonthemicroeconomicmechanismsthroughwhichtheseverityof theEbolaepidemic,andtemporaryschoolclosures,impactedtheeconomiclivesofyoungwomen inalow-incomeandfragilestate. SierraLeoneisasettinginwhichwomenfacearangeofsocialandeconomicdisadvantages. AsPanelAinFigureA1shows,ontheeveoftheoutbreak,SierraLeonerankedneartheglobal bottomoftheUNDPGenderInequalityIndex. 1 RelativetotheSub-SaharanAfricaaverage, ithashighratesofadolescentfertility(PanelB)andthehighestrateofmaternalmortalityin anycountryforwhichdataexists(PanelC).Thisispartlydrivenbytheextremelylowlevels ofpublichealthcareprovision(PanelD):pre-epidemictherewere 0 2 doctorsand 3 nursesper 10 000 people(thecorresponding guresformostOECDcountriesare 30 +doctorsand 100 + nurses),inacountrywithanestimated 1 4 millionwomenofchild-bearingageand 1 1 million under- vechildren.AccordingtotheWHO,teenpregnancyisoneoftheleadingcausesofdeath formothersinSierraLeoneandillegalabortionsarecommon.Itisalsoasettingwherethereisa highprevalenceofsexualexploitationandviolencetowardsyoungwomen.Forexample,overhalf ofwomeninSierraLeone( 56 %)reporthavingsu¤eredsomeformofgenderbasedviolenceduring theirlifetime[AmnestyInternational2015]. Itwasinthecontextofpre-crisisSierraLeonethatourdatacollectionexercisewasoriginally planned.Thiswasforarandomizedcontroltrialevalu
ationoftheEmpowermentandLivelihood forAdolescents(ELA)intervention,deliveredbytheNGOBRAC,andintendedtobuildonour earlierworkshowingpositivee¤ectsofthesameinterventioninUganda[Bandiera etal. 2020]. Theinterventionprovidesgirlsasafespace(clubs)wheretheycanmeetandsocialize,receive 1 Thisindexaggregatesinformationonmaternalmortalityrates,adolescentfertilityrates,educationbygender, femaleheldparliamentaryseats,andgenderinequalityinlabormarketparticipation. 1 lifeskillstoimprovetheirreproductiveknowledgeandhealth,andvocationaltrainingtoimprove theirlabormarketprospects.Fieldworkforourbaselinesurveywascompletedaweekpriorto the rstcasesofEbolabeingreportedinMay2014. The2014-16Ebolaoutbreakledtothe longest,largest,deadliest,and...mostcomplex[Ebola outbreak]inhistory [UNDG2015].Theoutbreaka¤ectedSierraLeone,GuineaandLiberia, infecting 28 652 individuals,with 11 352 deaths[CDCPestimate,April2016].Thereweremore casesanddeathsinthisoutbreakthanallearlieroutbreakscombined.SierraLeonewasthemost a¤ectedcountry,hostinghalfofallcases.Rapidcontagionforcedthegovernmenttoimplement familiarpoliciestoenablesocialdistancing:villagelock-downsandtravelbans,andallprimary andsecondaryschoolswereclosedthroughthe2014-15academicyear. Inthiscontext,schoolclosurescouldhaveespeciallyacuteconsequencesforyoungwomen. Withouttheprotectionoftimeinschool,youngwomenmighthavebecomemoreexposedtoearly pregnancyandsexualabuse.Furthermore,justbeforeschoolswereduetore-openinApril2015, theMinistryofEducationsurprisinglyannouncedthecontinuationofapre-Ebolapolicy;that visiblypregnantgirls wouldbeunabletore-enrol.ThesepoliciescreateapreciselinkwhyELA clubsmatterduringandaftertheepidemic.Mostdirectly,ifELAclubsreducethelikelihood ofpregnancy becausetheyo¤eralternativestospendingtimewithmenduringthecrisiswhen schoolsareclosed younggirlbene ciariesaremorelikelytobeabletore-enrolwhenschools re-openedpost-epidemic. Thesefactorscombinetolinkshortrunschoolclosureswithlongrunhumancapitalaccu- mulationamongyounggirls.Westudy theoreticallyandempirically thechainofoutcomes thatformthislink:timespentwithmenspen
tengaginginsexualactivities,teenpregnancy,and re-enrolmentbackintoschoolpost-epidemic. Wedrawtogetherthefeaturesofourcontextandinterventiontodevelopasimplemodelof youngwomen stimeallocationbetweenschool,ELAclubsandsocializing(includingtimewith men).Thismakesprecisehowschoolclosuresduringanepidemiccanhavepersistentimpactson schoolingpost-epidemic.Intuitively,theprovisionofsafespacesduringanepidemicreducesthe likelihoodofbecomingpregnant,andthusincreasesthe intertemporal likelihoodofreenrollinginto schoolpost-epidemic.However,post-epidemic,safespacessubstituteforother contemporaneous timeuses,includingschoolingandsoreducelongrunenrolmentallelseequal.Thetreatmente¤ect ofprovidingELAclubsonpost-epidemicschoolenrolmentcanbepositiveornegative,depending onwhichofthesee¤ectsprevail.Themodelpinsdownthatthepregnancyriskyounggirlsface duringtheepidemicdeterminesheterogeneityinthesignoftreatmente¤ectsonschooling.This mapsdirectlyintoourresearchdesign. 2 Ourevaluationsamplecomprises 200 villagesinfourdistrictsofSierraLeone.Theintervention wasrandomlyassignedto 150 villageswiththeother 50 heldascontrolvillages.Thefactthatour evaluationwasunderwayatthetimeoftheoutbreakwasentirelycoincidental:theELAprogram isnotintendedasaresponsetothecrisis.Weexploitthetimingofeventsandrandomizedroll outofELAclubstodocument:(i)howtheseverityoftheEbolashockcorrelatestochangesin theeconomiclivesofthe 4 700 trackedgirlsandyoungwomenaged 12 to 25 ;(ii)whetherthe availabilityofELAclubsmitigatedanyoftheseimpacts. Withtheonsetofthecrisisandall eldworksuspended,weimplementedtwovillage-level phonesurveys:(i)amonitoringsurveytoELAclubmentorsintreatmentvillagesinJune/July 2015toprovideinformationonclubfunctioning;(ii)avillageleaderssurveyadministeredbetween JuneandOctober2015,providinginformationonthelocalizedhealthimpactsofEbola,andpolicy responses(suchasthefunctioningofschoolsandhealthfacilities,andotherreliefe¤orts).After eldworkrestrictionswerelifted,ourendlinesurveywas eldedbetweenFebruaryandMay2016. Thisiswellafterschoolshadreopenedandmeasuresoutcomespost-epidemic.Finally,weengaged inonefurtherroundofdatacollectionbetweenJune2
019andJanuary2020,tostudylongterm outcomesandpersistentimpactsofthecrisis,ELAclubsandtheirinteraction. Ourmonitoringdatacon rmstherewasanextensiverolloutoftheELAprogramdespitethe circumstances: 70 %ofclubsopenedontime(bySeptember2014).Therewasalsohighdemand toparticipate: 71 %ofsurveyrespondentsintreatedvillageseverparticipatedinanELAclub meetingoractivity(versus 4 %incontrolvillages). FollowingthesimpleintuitionofthemodelthattheimpactELAclubshaveonschoolingvaries withpregnancyrisk,weusea 2 £ 2 researchdesign,whereonedimensionistherandomassignment ofvillagestoELA,andtheseconddimensionexploitsvariationintheriseinpregnancyriskthat girlsandyoungwomenfacespeci callyduringtheepidemic.Weconstructavillage-speci cindex ofpregnancyriskusingdatacollectedfromvillageleaderssurveyduringthecrisis.Theindex combinesinformationonwhetherthenearestsecondaryschoolre-openedontime(someschool re-openingsweredelayedfromApril2015,butallschoolswouldhavere-openedbythetimeof endlinesurveyinFebruary2016),anddisruptionstoservicesprovidedbylocalprimaryhealth units(PHU).Thesecomponentsrelatedirectlytoincreasedpregnancyriskforyounggirlsand womenbecause:(i)prolongedschoolclosuresincreasetheperiodoverwhichtheyareatriskto unwantedsexualabuseordemandsfrommen;(ii)PHUclosuresreduceaccesstocontraceptives. WemakeprecisetheidentifyingassumptionsforthedesigntoprovidecausalimpactsofELA clubs,andtheirinteractionwithgirls exposuretogreaterpregnancyrisk.Weprovidesupportive evidencefortheseassumptions. 3 Thatyoungwomenfacearangeofdisadvantagesinthissettingisstarklyquanti edinour baselinedata:respondentsareonaverage 18 yearsold,yet 60 %areinrelationships, 28 %are married,andnearlyhalfhavechildren.Whiletheirageatmarriageis 16 ,theaverageageat marriageoftheirhusbandsisalmostdouble,at 31 .Forthoseinrelationships, 46 %reportbeing subjecttointimatepartnerviolence.Thesetraitsofearlymarriage,childbearingandexposureto violenceallhaveclearlongtermconsequencesonwomen sabilitytoacquirehumancapitaland lead nanciallyindependentlives. 2 Weconductouranalysisseparatelyforyoungergirls(thoseaged 12 - 17 atbaseline),andolder girls( 18 - 25
atbaseline).Theyoungercohortareourprimaryfocusofattentionbecauseforthem thechiefconcernisthatschoolclosuresinresponsetoEbolameantthatwithouttheprotection oftimeinschool,theyfacedahigherriskofbecomingpregnantduringtheepidemic.Thiswas compoundedwiththepolicyofnotallowing visiblypregnantgirls tore-enrolinschoolsoncethey reopened.Thesefactorscombinetolinkshortrunschoolclosureswithlongrunhumancapital accumulationamongyounggirls,andprovideaclearchannelthroughwhichthesafespacethat ELAclubsprovide,canhelpmitigatethesee¤ects.Henceforyoungergirls,wepresentevidence alongthecausalchainofoutcomesofthislink:timespentwithmenspentengaginginsexual activities,teenpregnancy,andre-enrolmentbackintoschoolpost-epidemic. Ofcourse,oldergirlscanalsobene tfromtheavailabilityofELAclubs.Hencewealsotrace outtreatmente¤ectsforthemrelatedtotimespentwithmen,riskybehaviorsandchildbearing. Ourmainresultsareasfollows. First,incontrolvillagesabsentELAclubs,thepost-epidemicimpactsonyoungergirlsof higherpregnancyriskduringthecrisisarethat:(i)theyspendmoretimeexposedtomenfor sexualrelations:post-epidemicthetimespentwithmenincreasesby 1 27 hrs/wk(a 50 %increase overthebaselinemean);(ii)ratesofteenpregnancyriseby 10 5 pp,mostlydrivenbyarisein out-of-wedlockteenpregnancies;(iii)timespentinformallearningactivitiesfallsby 12 2 hrs/wk ( 25 %ofthebaselinemean),withthistimebeingreallocatedtowardswork/incomegenerationand householdchores;(iv)thisalltranslatesintofarlowerschoolenrolmentratespost-epidemic,by 17 pp.Toreiterate,thefallinenrolmentismeasuredwellafterthecountryisdeclaredEbolafree andschoolshaveactuallyreopened. TheavailabilityofELAclubsbreaksthiscausalchainintreatedvillages.Morespeci cally, inhighpregnancyriskvillagesthetreatmente¤ectofELAclubsonyounggirlsarethat:(i) 2 Itisnowwellrecognizedthateconomicdevelopmentandwomen sempowermentarecloselylinked[Doepke etal. 2012,Du o2012,Jayachandran2015],andthatcoredimensionsofdisadvantagestemfromwomenhaving limitedagencyovertheirbodies,facingbarrierstoinvestingintheirhumancapital,andhavingpoorlabormarket prospects[FieldandAmbrus2008,Dupas2011,Jensen2012]. 4 theyreducetheamou
ntoftimeyounggirlsspendwithmenby 1 86 hrs/wk ELAclubsprovide aclearalternativetospendingtimewithmenwithwhomtheyaresexuallyactive;(ii)ratesof out-of-wedlockteenpregnancyfallby 7 2 pp;(iii)timespentengagedinanyformallearningrises by 9 84 hrs/wk( 81 %ofthedirecte¤ectofhigherpregnancyrisk) thistimecomesfromgirlsbeing lessexposedtospendtimeonhouseholdchores;(iv)thisalltranslatesintoatreatmente¤ecton schoolingpost-epidemicof 8 7 pp(closelymatchingthefallinpregnancyrates),whileELAclubs alsoincreasethelikelihoodthatgirlscancombineschoolandworkby 9 7 pp. Third,theavailabilityofELAclubsalsohaveimpactsonoldergirls(thoseaged 18 - 25 at baseline).Intreatedvillages,oldergirlsreportincreasesinunwantedsex(by 5 4 ppor 38 %ofthe baselinemean),andreportengaginginmoretransactionalsex(alsobyexactly 5 4 pp, 115 %of thebaselinemean).AsinDupasandRobinson[2012]andlaterstudies,theuseoftransactional sexisoneformofincomegenerationavailabletowomeninatimeofsevereaggregatecrisiswhen conventionaleconomicopportunitiesarescarce.Reassuringly,ELAclubspreventthesechanges translatingintohigherfertilitybecauseoldergirlstakeonthelifeskillsprovidedbyELAand increasetheiruseoffemalecontrolledcontraceptives. Fourth,ourlongtermfollowup conductedfouryearsaftertheendoftheEbolaepidemic in2019/20 establishesthatforyounggirls,thereishysteresisinoutcomesofbeingexposedto higherpregnancyriskduringtheepidemic,butalsoofhavingaccesstoELAclubs,asthecountry entersapathtorecovery.Inparticular,amongthoseaged 12 - 17 atbaselinein2014,we ndthat inthelongtermfollowup,thoseinhigherpregnancyriskcontrolvillagesarealmost 15 ppmore likelytohavebecomepregnantsincebaseline,spendalmost 7 hrs/wklessonlearningactivities, andhaveenrolmentratesthatare 11 pplower. However,theimpactsoftheavailabilityofELAclubsareequallypersistent:thetreatment e¤ectinhighpregnancyriskvillagesisthatby2019/20,youngergirlshave 13 pplowerpregnancy rates, 14 pphigherenrolmentrates,andstillspendinglesstimewithmen(by 1 hr/wk).Tobeclear, thiscohortofgirlsareaged 17 to 22 inthislongtermfollowup,sothesewomenarestudying atthehighesttiersoftheformaleducationsystem(evenwithstandingtheyearof
losteducation thoughschoolclosures). Fortheoldercohortofgirls,thoseaged 18 to 25 atbaselinein2014,we ndthoseinhigher pregnancyriskcontrolvillagesare 11 ppmorelikelytohavebecomepregnantsincebaseline.For highpregnancyriskvillages,we ndlongrunpersistenceinthetreatmente¤ectofengagingin transactionalsex butthisdoesnotleadtoincreasedpregnancies,becauseoflonglastingchanges infemale-controlledcontraceptiveuse. Weshowthesecore ndingstoberobusttoadjustingp-valuesforrandomizationinference 5 ormultiplehypothesistesting,alternativeconstructionsofthepregnancyriskindex,andsocial desirabilitybiases,followingtheapproachsetoutinDhar etal. [2020]. Finally,ourlongtermfollowupgaveustheopportunitytointerviewmenforthe rsttime inourstudy.Weselectedpartnersoftrackedrespondents,andshedlightonwhetherthecharac- teristicsofpartnersrelatestotheavailabilityofELAclubstogirlsandyoungwomenduringthe epidemic.We ndpositivetreatmente¤ectsofELAclubsonpartnertraits:forexampletheyare bettereducatedandmoreaversetogenderbasedviolence.Whileonlysuggestive,theseresults pointtoanotherpromisingmarkeroflongrunwelfareimprovementsforgirlsandyoungwomen thathadELAprotectivespacesavailabletothemduringtheEbolaepidemic. Ouranalysisbreaksnewgroundinthreeliteratures:ontheeconomicsofepidemics,onhouse- holdresponsestoaggregateshocks,andonthelinkbetweeneconomicshocksandgenderinequality. PriortoCOVID-19,anascentliteraturehadbeguntostudyindividualbehaviorduringepi- demics. 3 WorkontheEbolaepidemicinWestAfricahasfocusedonmeasuringrealtimeimpacts ofthecrisisonhouseholdsand rms[Thomas etal. 2014,Bowles2016,Glennerster etal. 2016, Casey etal. 2017],orexploitingquasi-experimentalvariationinthegeographicincidenceofEbola tounderstandgovernmentresponses[Fluckiger etal. 2019,Ma¢oli2020].Incontrast,ourwork tracksasampleofyounggirlsandwomensoverthepre-crisis,crisisandpost-crisisperiods,to understandhowtheepidemicimpactedtheirhumancapitalaccumulation,thatiskeyforlifetime welfare.Methodologically,thepapermostcomplementarytooursisChristensen etal. [2020]: theyalsooverlayapre-plannedRCTinSierraLeonewiththeepidemicshock.Theydocument howinterventio
nsimplementedpre-epidemictoimproveaccountabilityoflocalhealthfacilities, laterledtothosefacilitiesfunctioningbetterthroughtheepidemic,intermsofhigherreported EbolacasesandlowerEboladeaths. Onhouseholdresponsestoshocks,avastliteraturestudies exante and expost mechanisms tomitigateidiosyncraticrisks.Asmallerliteratureexaminesresponsestoaggregateshocks,given greateridenti cationchallenges.Inthepresenceofaggregateshocks,copingstrategiesfordealing withidiosyncraticrisk(suchastemporarymigration),oftenbreakdown.Householdsmightthen beforcedtoengageinbehaviorsthatdonotmaximizetheirlongrunwelfareorpermanent income,suchaspullingchildrenoutofschool[JacobyandSkou as1997,FerreiraandSchady 2009],marryingo¤daughterstoobtainbrideprices[Corno etal. 2019],orwomenengagingin transactionalsex[DupasandRobinson2012].Wedocumenthowthelattercopingstrategyis relevantforoldergirlsandwomeninthecontextoftheEbolaepidemic. Mostimportantly,hard-earnedgainsinwomen sempowermentcanbequicklyerasedbyag- 3 ThisincludesAdda[2016]on u;AgüeroandBeleche[2017]onH1N1;Bennett etal. [2015]onSARS;and LautharteJuniorandRasul[2020]andRangel etal. [2020]onZika. 6 gregateeconomicshocks,anditisintimesofgreatestcrises,thatgenderdi¤erentialsinoutcomes aremostlikelytoopenup[Du o2012] anissueattheforeofpolicydiscussionsinallcountries inthecurrentpandemic.Thereisanemergingexperimentalliteratureevaluatinginterventions designedtoempowerwomeninperiodsofstability[Baird etal. 2011,Du o etal. 2015,Buchmann etal. 2018,Ashraf etal. 2020,Bandiera etal. 2020,Dhar etal. 2020,Edmonds etal. 2020]. Our ndingsaddtothisbodyofworkintwoimportantways.First,wedocumentpreciselyhow gendergapscanopenupintimesofhealthshockssuchasepidemics.Wedosobydemonstrating howschoolclosures,acommonpolicyresponsetoaidsocialdistancinginviralepidemics,have persistentdetrimentalimpactsonthelivesofyounggirlsandwomen. 4 Second,weshowthat simplepolicyinterventionscanhelpo¤setthesee¤ects.Thisinsighthasdirectrelevanceformuch oftheongoingdiscussionofwhethertheoptimalpolicyresponsestoCOVID-19di¤erinlow-and high-incomecountriesgivendi¤eringtrade-o¤sandstatecapaciti
es,aswellasissuesthatmany countriesaregrapplingwithintermsofhowthepandemicandpolicyresponsestoitarehaving hugelydi¤erentialimpactsacrossgenerations. Thepaperisorganizedasfollows.Section2providesbackgroundontheEbolaepidemic, policyresponsesit,andtheELAintervention.Section3describesourdataandpresentsmoti- vatingdescriptives.Section4developsasimplemodelofyoungwomen stimeallocation.This makesprecisehowschoolclosuresduringanepidemiccanhavepersistentimpactsonschooling post-epidemic,andhowtheseimpactsareheterogenouswithrespecttopregnancyriskduringthe epidemic.Section5mapsthemodeltodata,describesourresearchdesign,itsidentifyingas- sumptionsandpresentsevidencetosupportthem.Section6showsourcoreresultsontheimpact ofpregnancyriskduringtheepidemicontheeconomiclivesofyounggirls,andthemitigating e¤ectsofELAclubs.Section7extendsour ndingstothoseforoldergirlsandyoungwomen,and documentsthelongtermimpactsonbothcohorts.Section8discussespolicyimplicationsand futureresearch.AdditionalresultsandrobustnesschecksareintheAppendix. 2ContextandIntervention 2.1TheEbolaEpidemic EbolaVirusDisease(EVDorEbola)isanacutehemorrhagicfeverthatcanbefatalifuntreated. Ebola rstappearedinsimultaneousoutbreaksinSouthSudanandtheDemocraticRepublic 4 Toourknowledge,ArchibongandAnnan[2020]istheonlyotherstudythatdocumentshowgendergapsopen upinsuchtimes,inthecontextofthe1986meningitisepidemicinNiger.Theyshowasigni cantreductionin yearsofeducationforschool-agedgirlsrelativetoboysfollowingtheepidemic,drivenbyhouseholdsrespondingto theshockbymarryingo¤daughtersinordertoclaimbrideprices. 7 ofCongoin 1976 .FigureA2chartsthehistoryofEbolaoutbreaksinSub-SaharanAfrica,with fatalityratesvaryingbetween 25 and 90 %.Thevirusistransmittedfromwildanimalsandspreads throughhuman-to-humantransmissionviadirectcontactwiththeblood,secretions,organsor otherbodily uidsofinfectedpeople,andwithsurfacescontaminatedwiththese uids(suchas beddingorclothing).Transmissioncanalsooccurinburialceremoniesinvolvingcontactwiththe deceasedbody.Individualsremaininfectiousaslongastheirbloodcontainsthevirus.Hence socialdistancingmeasuresareakeypolicyinstrumentusedt
otackleEbolaoutbreaks. 5 SierraLeonewasthecountrymosta¤ectedbythe2014-16epidemic,hostingabouthalfof allcases.Thevirusisthoughttohavebeenbroughtintothecountrybyanindividualentering fromGuineaaroundMay2014.ByOctober2014,ithadspreadtoall 14 districtsinthecountry, withrapidcontagioncausedbyhighratesofgeographicmobilityandtheuseoftraditionalburial practices.Figure1chartsthetimelineoftheepidemicfromMay2014,showingthenumberof weeklycases(con rmedandprobable).Thepeak owofweeklycasesoccurredinDecember 2014,butitwasonlyinJuly2015thattheepidemicstartedtoslowdown.SierraLeonewas declaredEbolafreeinNovember2016, 42 daysafterthelastpatientwasdischarged.TheWHO estimatestherewere 14 124 casesinthecountry(includingsuspected,probableandcon rmed cases),resultingin 3956 deaths.Hencethe 28 %fatalityrateislowerthaninsomeearlieroutbreaks, butthescaleandspreadoftheoutbreakinSierraLeonewasunprecedented. 6 2.1.1PolicyResponses TheSierraLeoneongovernmentusethreepoliciestocombatrapidcontagion,allofwhicharewell familiarasresponsesusedtoalsocombatCOVID-19:(i)healthworkersweremobilizedtorecord door-to-doorcasesandtrackcontagion,andsomehealthfacilitiesweretransformedintoEbola holdingcenters;(ii)socialdistancingmeasureswereused,includingvillagelock-downsandtravel bans;(iii)primaryandsecondaryschoolswereclosedthroughthe2014-15academicyear.The lowerpartofFigure1showsthetimelineofpolicesenacted. 7 TheepidemichadsevereconsequencesforhealthcareprovisioninSierraLeone,thatasPanel DinFigureA1shows,wasalreadyweakpre-epidemic.Ebolaimpactedthehealthsystemin twoways:(i)thehumancapitalofhealthworkers;(ii)publictrustinusinghealthfacilities 5 Ongoingvaccinetrialsreportedencouragingresults[Huttner etal. 2018].InDecember2019,asaresultof ayearlongclinicaltrialintheDRC,thetrial scosponsorsattheWHOandNIHannouncedthattwoofthe treatmentsappeartodramaticallyboostsurvivalrates(https://www.wired.com/story/ebola-is-now-curable-heres- how-the-new-treatments-work/).AnthonyFauciwasaleading gureinthesee¤orts. 6 Forthe2014-16outbreak,theWHOestimatesLiberiahad 10675 casesand 4809 deaths(a 45 %fatalityrate), andGuineahad 3811 caseswith 2543 deaths(a 6
7 %fatalityrate). 7 Othersocialdistancingmeasureswereenactedtoclosebarsandrestaurants.Whetherthiswasenforcedoutside ofbigcitiesisquestionable,anddoesnotappeartohaveimpactedparticipationinELAclubs. 8 [Christensen etal. 2020].Onthe rstdimension,healthworkerswereunder-equippedandunder- preparedforEbola.Theirinabilitytorapidlyimplementinfectionpreventionandcontrolmeasures leftthemexposedtoinfectionduringroutinecontactandenabledfurthertransmissiontoother healthworkers[Evans etal. 2015]. 8 Ontheseconddimension,healthfacilitiesbecameassociated withEbolaassomeweretransformedintoholdingcenters.Visitstohealthcentreswerethought tobeamongthelargestcausesofEbolatransmission.Combinedwithalackofacureandhuge uncertainty,con denceinthehealthsystemwasundermined,leadingfamiliestokeepsickmembers athome,thusfurtherspreadingthevirus. Thecollapseofthehealthcaresystemmeantaccesstostandardmedicalservices,suchasante- natalandmaternalcare,wasseverelyhamperedduringtheoutbreak.Acombinationofcapacity constraintsandfearofhospitalsledtoconsiderablyfewerwomenaccessingantenatalcareorgiving birthinhealthfacilitiesduringthecrisis[UNICEF2014]. Theeconomicconsequencesoftheaggregateshockwerealsosevere.Inayear,GDPgrowth plummetedfrom +8 9 %to ¡ 2 0 %:borderclosuresshutdowninternationaltrade(predominantly inagriculture),internaltravelbansresultedinthebreakdownofdomestictrade,andallperiodic marketswereforcedtoclose.Theself-employmentsector,accountingfor 91 %ofthelaborforce, shedaround 170 000 jobs(withrevenuesforsurvivingenterprisesfalling 40 %),andafurther 9 000 jobswerelostinwageemployment[Thomas etal. 2014,2015,Evans etal. 2015,Himelein etal. 2015,Casey etal. 2016]. 9 The nalpolicyresponserelatedtotheeducationsystem.SchoolswereclosedinMay2014and re-openedinApril2015astheepidemicbegantoslowdown(theschoolyearrunsfromSeptember toJuly).Schoolclosuresmighthavehadparticularlyacuteimpactsonyoungwomen.Thelossof oneyearofhumancapitalaccumulation ifpermanent wouldbenon-trivialgiventheirlower levelsofhumancapitaltobeginwith.Moreover,withouttheprotectionoftimeinschool,young womenmighthavebecomemorevulnerabletosexualabuse,thatwouldfurthe
rlimittheirability toaccumulatehumancapitalinfuture[AmnestyInternational2015]. Thesegender-speci cconsequencesofschoolclosureswerecompoundedbyanotherpolicy responseduringtheepidemic.Justbeforeschoolswereduetore-openinApril2015,theMinistry ofEducation,ScienceandTechnologyannouncedthecontinuationofapre-Ebolapolicy,that visiblypregnantgirls wouldbeunabletore-enrol.Giventhedi¢cultyofcorrectlyidentifying 8 Evans etal. [2015]documenthowEboladeathsweredisproportionatelyconcentratedamonghealthpersonnel. Forexample,byMay2015,while 06 %ofthepopulationhaddiedfromEbola, 6 85 %ofhealthcareworkershad diedfromEbola.Inabsoluteterms,thiscorrespondedto 79 doctor,nurse,andmidwifedeaths.Bytheendof Novembertherehadbeenafurther 179 con rmedEbolacasesamonghealthworkers. 9 Thebestdataonimpactsonfoodpricescomesfromtrackersurveysconductedat 200 marketsduringthecrisis [Glennerster etal. 2016].Theydocumentrelativelymodestpriceincreases(e.g.forrice),butmorepronounced impactsonpricedispersion,withincreasesanddecreasesbacktonormalcyduringthecrisis. 9 apregnantgirl,schoolprincipalsgaineddiscretioninexactlyhowtheyenforcedthisban.In short,thispolicyincreasedthelongruncostofschoolclosuresforyoungwomen,thatwithoutthe protectionoftimeinschoolfacedahigherriskofbecomingpregnantduringtheepidemic. 10 Thesefactorscombinetolinkshortrunschoolclosureswithlongrunhumancapitalaccu- mulationamongyounggirls.Westudythechainofoutcomesthatformthislink:timespent withmenspentengaginginsexualactivities,teenpregnancy,andre-enrolmentbackintoschool post-epidemic. 2.2TheELAIntervention Theempowermentandlivelihoodforadolescents(ELA)interventionaimstokickstartyoung women ssocioeconomicempowermentthroughtheprovisionoflifeskills,vocationalskillsandmi- cro nance.TheELAprogramwasdesignedandimplementedbytheNGOBRACinBangladesh, wherefemaledisempowermentisalsoamajorconcern.Since1993,BRAChasestablished 9 000 ELAclubsworldwide,reachingoveramillionyoungwomen.Theprogramhasprovedtobe scalableandcost-e¤ectiveacrosscountries.Inearlierwork,weevaluatedtheprograminUganda [Bandiera etal. 2020],andbasedontheencouragingfour-yearimpactsdocumented,wedesigned af
ollow-upevaluationwithBRACinanotherSubSaharancontext. TheELAinterventiono¤ersamultifacetedapproachtosimultaneouslytacklemultipledis- advantagesyoungwomenface,relatedtohavingagencyovertheirbodiesandbarrierstoaccu- mulatinghumancapital.AllprogramactivitiesaredeliveredoutofELAclubs,a xed(rented) locationineachvillage,withnoattendancefee.Thisisaphysicalspacejointly owned byclub members.ELAclubso¤eraspacewhereyoungwomencansafelygatherandsocialize,awayfrom men.Clubswereoriginallydesignedtobeopen vedaysaweekduringafter-schoolhours.During thecrisis,ELAclubscanalsoserveasapartialsubstituteforschools:theyprovideasafespace wheregirlscanmeet,andtheyalsoprovideinformation,onreproductivehealthforexample,that arenotsuppliedbytheeducationsystem. ELAclubscanalsocrowdouttimespentatinformalinstitutionsofsecretsocietiesthatexist formenandwomeninSierraLeone[MacCormack1979,Bledsoe1990,MCormack-Hale2018].The primaryroleofthesewomen ssocieties(knownas Bondo intheNorthand Sande intheSouth) istoinitiategirlsintoadulthoodthroughvariousrituals,thathavehistoricallyincludedfemale genitalmutilation. 11 Thesesocietiescreatedistinctionsbetweenwomenwhohaveexperienced 10 InMay2015,itwasannouncedthatanalternative bridging educationsystemwouldbeestablishedtoallow pregnantgirlstocontinueschooling,butindi¤erentpremisesortimestotheirpeers.Othertemporarymeasures, suchascommunitylearningcentresandhome-basedapproaches,werealsoimplemented.Atbest,thisbridging systemvariedine¤ectiveness,anddidnothingtohelppregnantgirls ndanalternativewaytotakenationalexams. 11 SierraLeonehassomeofthehighestlevelsofFGM,indicatingthepervasivenessofsecretsocieties.The2013 10 thesecretsofchildbirthandthosewhohavenot,andsocietyleaderstypicallyfurtherattemptto separategirlsundergoinginitiationfromtheiruninitiatedpeers.Secretsocietiesalsoinstillnotions ofmoralityandnormsoversexualbehavior,theypromotewomen ssocial/politicalinterests,and expresssolidarityamongwomenvis-à-vissecretsocietiesformen. AllinterventioncomponentsaredeliveredfromELAclubs.Anoldergirlfromeachvillage isselectedandtrainedtobeaclubmentor.Hermainresponsibili
tyistomanagetheclub activitiesandfacilitatethelifeskillstrainingcourses.TableA1detailsthecurriculumforthe 10 lifeskillsmodules,thatcoversreproductivehealth,menstruation,pregnancy,STDs,HIV,family planning,rape,legalknowledgeonbrideprice,childmarriageandviolenceagainstwomen.As sexeducationisnotobligatoryinschools,thelifeskillsprovidedthroughELAgiveyoungwomen accesstoinformationtheymightnototherwisehavereceived.Girlsaged 17 andaboveareeligible forthevocationalskillscomponent,deliveredbyBRACprofessionals.Finally,thoseaged 18 or olderwereeligibleforthemicro nancecomponent. 12 2.2.1ImplementationandParticipation WeuseourELAclubmentorsurveyconductedinJune/July2015,toprovideevidenceontheroll outofELAclubsduringtheepidemic.PanelAofFigureA3providestimeseriesevidenceonELA clubopenings:(i) 70 %openedontime(bySeptember2014)andbyJanuary2015allhadopened; (ii)themajoritywerecontinuouslyopenthroughtheepidemic.PanelBshows:(i)themajority ofclubsprovidedlifeskillstraining;(ii)vocationaltrainingtooko¤aftertravelquarantineswere liftedinJanuary2015(thesetrainingsaredeliveredbyprofessionals,notclubmentors).Panel Cshowsthemedianclubhas 30 members,implyingmembershipratesofaround 31 %.PanelD showstheratioofclubattendance,basedonanunannouncedspotcheckinMay2015,toclub membership.Onanygivenday,ELAclubscanhavemanynon-memberspresent. ThereishighdemandforELAclubs.PanelAofTableA2showsthat 31 %ofeligiblesin treatedvillagesareregisteredclubmembers.PanelBreportsstatisticsfromtheendlinesurvey. ThereiswidespreadknowledgeofELAclubs: 89 %ofgirlsintreatedvillageshaveheardofthem, ashave 27 %ofgirlsincontrolvillages.Participationratesaremorethandoublemembership DHSreports 90 %ofwomenaged15to49havebeencircumcised.Duringtheoutbreakthegovernmentintroduced amoratoriumonFGM,butthisisnotthoughttohavebeenenforced. 12 Skillsprovidedincludetailoring,soapmaking,hairdressing,andtiedying.Clubsprovidediversi edcourses ratherthantrainingallparticipantsinoneactivity.Allcoursesinvolveda nancialliteracymodule,andupon completion,participantsreceivedbasicbusinessinputs,e.g.sewingmachineswereprovidedtothosecompleting tailoringcourses.Eachcoursewaso¤ereddailyfors
ixhoursperday,withcoursesvaryinginlengthdepending onthehumancapitalinvestmentrequired.Loanswererenewableonrepayment.Onthemicro nancecomponent, loansizeswereupto$ 100 ,repayableoverayear,withaweeklyrepaymentscheduleanda 30 %interestrate.The rstloancyclestartedinApril2015. 11 rates: 71 %ofgirlsintreatedvillageshave ever participatedinanELAclubmeeting/activity. Conditionaloneverparticipating:(i) 82 %ofyoungwomenhaveparticipatedinlifeskills( 77 % reportattendingatleastonceaweekandthemajoritycanrecallatleastfourtopicscovered); (ii) 25 %havereceived nancialliteracytraining;(iii) 34 %haveparticipatedinvocationalskills training(inT2/T3treatmentarms);(iv) 13 %reporthavingreceivedamicro nanceloaninT3. Thesepatternsofawarenessandparticipationareverysimilarforyoungandoldagecohorts (althoughtheoldercohortismorelikelytohavereceivedmicro nance).Hencethedemandfor ELAclubscomesbothfromgirlsenrolledfulltimeinschoolpre-crisis,aswellasoldergirlswho werepredominantlyengagedinincomegenerationpre-crisis. 3DataandDescriptives Ourevaluationcoversfourdistricts:PortLoko,Kambia,MoyambaandPujehun,where 20 %of thepopulationresidedpre-crisis.Figure2showsourtimelineofdatacollection,andhowthis relatestothetimingofthecrisisandELAclubactivities.InOctober2013we rstconducted acensusinthe 200 evaluationsamplevillages,covering 94 338 individualsin 17 233 households. Thiscensuswasusedtodrawasampleofwomenaged 12 to 25 andthuseligibleforELA. OurbaselinewasconductedbetweenFebruaryandMay2014,thusendingjustasthe rst casesofEbolawerebeingreportedinSierraLeone.Thesurveycovered 5 775 youngwomen, correspondingto 27 %ofalleligiblesinthe 200 samplevillages,andrecordedinformationontheir timeuse,educationandskills,labormarketactivities,andpregnancies/riskybehaviors. GiventhatELAclubsprovideanalternativeactivityforyoungwomentoengagein,timeuse dataplaysanimportantroleinouranalysis.Toelicitreliableinformation,ratherthanjustasking abouthours,weaskedrespondentstouseaphysicalrepresentationoftime(beans)toshowhow theydividedtheirtimeoverthepastweek.Wedidsoforbroadactivities(education,socializing, incomegeneration,householdchores),and
alsospeci callyfortimedevotedtosubcategoriesof socializing(sexualrelationshipswithmen,withfriends,socialactivities,alone). 13 Thebaseline andendlinesurveystookplaceduringtheschoolyearsorespondentscouldfeasiblyhavebeen attendingschoolintheweekofthesurvey.Thecredibilityofthetimeusedataisunderpinnedby thefactthat:(i)thenumberofbeansrecordedacrosscategoriessummedupto 24 for 90 %( 99 %) 13 Thequestionwordingforthebroadercategoriesis, "NowIwouldlikeyoutodoasimpleexercise.Hereon thesecardsaresomewaysyoucanspendyourtimeinatypicalweek.Hereare25beans.Pleasedividethesebeans betweenthecardsaccordingtohowmuchtimeyouspendineachactivity. Fortimeuserelatedtosocializing,the questionwordingis, "Herearethe25beansagain.Hereonthesecardsaresomewaysyoucanspendyourfree (leisure)time.Pleasedividethesebeansbetweenthecardsaccordingtohowmuchtimeyouspendineachactivity. Ifthereareanyotheractivitiesnotlistedonthesecards,youcanwritethemontheseblankcards. 12 ofrespondentsatbaseline(endline);(ii) 87 %ofrespondentsreportsleeping 5 to 10 hoursper night;(iii)theaveragenumberofhoursspentperweekatELAclubsisthree,whichisrealistic. Weconverttimeusemeasuresintohoursperweek. Withtheonsetofthecrisisandall eldworksuspended,weimplementedtwovillage-levelphone surveys.The rstwasamonitoringsurveytoELAclubmentorsintreatmentvillagesconducted inJune/July2015,providinginformationonclubfunctioning.Thesecondwasavillageleaders surveyadministeredbetweenJuneandOctober2015,providinginformationonvillageimpactsof Ebola(intermsofthenumberofhouseholdsquarantined,Ebolarelatedcasesanddeaths),and policyresponses(suchasthefunctioningofschoolsandhealthfacilities,andotherreliefe¤orts). 14 After eldworkrestrictionswerelifted,theendlinesurveywas eldedinpersonbetweenFeb- ruaryandMay2016(solikethebaseline,takingplaceduringtheschoolyear).AsFigure1shows, thisisaroundsixmonthsafterthein owofnewreportedcasesofEboladeclinedtonearzero, andwellafterschoolshadreopened.ItwasstillbeforeSierraLeonewasdeclaredtobeEbolafree (November2016).Theendlinecoveredthesametopicsasthebaselinesurveywithanadditional modulerelatedtothecrisisandexperiencesduringit.
Theendlinesurveymeasuresoutcomespost-epidemic.Wedonotclaimthe onset ofany changesinbehavioralwaysstartedduringthecrisis.Rather,theresultsshouldbeinterpretedas capturing persistent impactsintothepost-crisisperiodoftheepidemic,andofELAclubs. Onattrition, 83 %ofrespondentsweretrackedfrombaselinetoendline( 4 790 ).Amongthose tracked, 81 %( 3 865 )residedinthesamevillage,whileothersweretypicallytrackedtoanearby village.Hencealthoughgeographicmobilityishigh,itdoesnotleadtosevereattrition.Appendix TableA3presentscorrelatesofattritionandshowstreatmentassignmentandtheintensityof pregnancyriskduringtheepidemicdoesnotpredictattrition,nordoestheirinteraction,andnor istheredi¤erentialattritiononobservableswithtreatmentorrisk. Finally,weengagedina nalroundofdatacollectionbetweenJune2019andJanuary2020,to studylongtermoutcomes.Amonggirlstrackedtoendline,wehadfundingtosurvey 71 %ofthem inthislongtermfollowup.Wemanagedtotrack 84 %ofthisintendedsample,corresponding to 2852 respondents.Themainpurposewasbothtoexaminethepersistenceofoutcomespost- epidemic,butalsotocollectnewdatafromtheirpartners.The nalColumnofTableA3showthat treatmentassignmentandtheintensityofpregnancyriskduringtheepidemicdoesnotpredict attritionbetweentheendlineandlongtermfollowup,andnordoestheirinteraction. 14 Thevillageleadersurveycollectsdatacodedfromfocusgroupdiscussions.Prominentmembersofthesocioe- conomicandadministrativelifeofthecommunityattendedthesemeetings,withtheaveragefocusgroupinvolving 11 participants(theminimum(maximum)was 5 ( 18 )). 85 %ofthesemeetingswereattendedbyaChief(eithera Paramount,Section,RegentorVillageChief).Villageelders,women sandyouthleaders,imams,pastors,head teachers,nursesandELAclubmentorswerealsoinvited. 13 3.1RandomizationandBaselineBalance TheELAprogramwasrandomlyassignedto 150 villages,with 50 remainingascontrols.Districts weretherandomizationstrata.Theoriginalevaluationdesignhadthreetreatmentarms:int1, ELAclubswouldonlyprovidelifeskillstraining;T2wouldbeasT1butclubswouldalsoprovide vocationaltraining,andT3wouldbeasT2butclubswouldalsoprovidemicro nance.This designwasmeanttounpacktheimpactsfoundinourworkonELA
inUganda[Bandiera etal. 2020].Giventheepidemicanddelayedrolloutoftheseskillscomponents,wepooltreatmentarms throughout.CommontoallisthatELAclubsprovideasafespaceforyounggirlsandwomen. 3.1.1Villages PanelAofTable1showsthattreatmentandcontrolvillagesarewellbalanced.Hence,therewas high delitywithrandomizationprotocolsevenwithanunfoldingepidemic.Theremainderofthe Tableshowsdatafromthevillageleadersurvey,conductedduringthecrisis.PanelBhighlights thestigmaassociatedwithpregnantgirls:thereisnearuniversalagreementamongelderswith thestatementthat girlswhoarevisiblypregnanthaveabadin uenceontheirnon-pregnant peers. Incontrolvillages,only 12 %ofeldersagreewithstatementsthatpregnantgirlsshouldbe allowedtocontinuetheireducation,ortotakeformalexams.PanelCdetailsthatfewvillageswere quarantined( 6 %),butnearlyallwerevisitedbycontacttracerteams.Therearenodi¤erences betweentreatmentandcontrolvillagesinthereceiptofassistancefromNGOsintermsoffood aidorschoolsupplies(excludingBRAC). 3.1.2YoungWomen Table2showsthattreatmentandcontrolsarewellbalancedonindividualcharacteristicsat baseline.PanelAshowsourrespondentsareonaverage 18 yearsold,and 60 %areinarelationship. Whilegirls ageatmarriageiscloseto 16 ,theaverageageatmarriageoftheirhusbandsisalmost double,at 31 .Despitetheiryoungagesandtherecentlyformedmarriages,nearlyhalfhave children.Forthoseinrelationships, 46 %reportbeingsubjecttosomeformofintimatepartner violence.EvenabsenttheEbolacrisis,thesetraitsofearlymarriage,childbearingandexposure toviolenceallhavelongtermconsequencesontheabilitytoacquirehumancapitalandlead nanciallyindependentlives.Indeed,only 23 %ofoursampleofyoungwomenareliterate(based onanassessmentcombiningbasicreadingability,readingcomprehensionandwritingsentences). PanelBshowsbaselinemeasuresofsexualactivity. 75 %aresexuallyactive,withtheirageat debutbeing 15 .Respondentsreportspendingaround 5 hrs/wkwithmen(thattheyareengaged insexualrelationswith).Theminorityofsexuallyactivegirlsreportusingcontraception. 10 % 14 reporthavingexperiencedunwantedsexinthepastyear,and 4 %haveengagedintransactional sex.Figure3Ashowshowsexualactivitiesvaryby
ageatbaseline.Byage 17 ,themajorityof girlsareinrelationshipsandsexuallyactive,withmorebeingsexuallyactivethaninrelationships ateachage.Therightsideaxisshowsthatamongoldergirls,over 10 %ateachageexperience unwantedsex,andthereisaweakgradientinageofengagingintransactionalsex. Theeconomicactivitiesofyoungwomenarebestdescribedasafour-waytypedistribution: 27 %ofyounggirlsincontrolvillagesareinschoolfulltime; 34 %areexclusivelyengagedinincome generatingactivities; 18 %areengagedinbothschoolingandincomegeneration,while 20 %report beingengagedinneitheractivity,hencespendingtheirtimeengagedinhouseholdchoresoras caregivers. 15 Figure3Bshowshowtheseactivitiesvarybyageatbaseline.Themajorityofgirls aged 12 - 14 areinschoolatbaseline,pre-Ebola.Therearenosharpdiscontinuitiesinenrolment ratesatkeycut-o¤stagesoftheeducationsystem(ages 15 and 18 )butrathergradualdeclinesin age,partlyduetograderetention.Thereisasteadyincreaseinspecializationinincomegeneration withage,with 17 beingacriticalcrossoverpoint:upuntilthatagemoregirlsareenrolledin educationfulltimethanareonlyworking,andthesituationreversesthereafter. 16 Giventheseagepro lesofsexualactivityandschoolenrolment,wesplitouranalysisbetween girlsaged 12 - 17 atbaseline,andthoseaged 18 - 25 .Theyoungercohortareourprimaryfocus ofattentionwhenunderstandingthepolicyrelevantimpactsofthesafespaceprovidedbyELA clubs,becauseforthemthechiefconcernisthatshortrunschoolclosurescanleadtolongrun impactsontheirhumancapitalaccumulation. 17 3.2EnrolmentandPregnancyRatesOvertheEpidemic Wepresenttwodescriptiveshighlightinghowschoolenrolmentandpregnancyratesmighthave beenimpactedoverthecourseoftheEbolaepidemic.Wedosofocusingonyounggirlsresiding incontrolvillages,sothatwerenevero¤eredthesafespaceofanELAclub. Figure4Ashowsschoolenrolmentratesbyage:(i)atbaselinein2014(pre-epidemic);(ii)at endlinein2016(post-epidemic)whenallschoolshadreopened;(iii)theirdi¤erence.Aggregate enrolmentfellfrom 45 %to 38 %,withfallsbeingobservedatallages.Ifthiswerejustpickingup acohorte¤ect(asinthecrosssectionatbaselineinFigure3A)thenthemagnitudeofthedrop 15 Incomegeneratingactivitiesatbaselineareformsofself-employment,suc
hassmalltrade/business( 40 %),food processing( 20 %)andhouseholdlandcultivation( 15 %). 16 Thekeyexamschoolstagesareatages15and18.Thetimeallocationsacrosstheagedistributionareplausible: theyoungestgirlsreportspendingaround 60 hrs/wkonallformsoflearningexceptELA.Workhoursrisetojust over 35 hrs/wkforolderwomeninoursample.Ateachage,respondentsreportonaveragespendingatleast 40 hrs/wkengagedinhouseholdchores. 17 ThisagesplitbroadlyalignswiththeseparationbyageinsecretsocietiesinruralSierraLeone:inturnthis mightlimitinformation owsorspilloversbetweenyoungerandolderagecohorts. 15 wouldbeincreasinginagebutthisisnotso:fallsinenrolmentbyagearenon-monotonic,being largestforthoseaged 14 and 15 attheonsetoftheepidemic. Figure4Bshowsreasonsfordropoutchangedthroughtheepidemic.Wecomparereasonsgiven bygirlsaged 12 - 17 thathadalreadydroppedoutatbaseline,tothoseinthesameagebandthat droppedoutbetweenbaselineandendline,i.e.thosethatdidnotre-enrolafterschoolsreopened. Pre-crisisthemostcommonreasongivenfordropoutwascost,followedbypregnancy.Post-crisis, pregnancybecomesthemodalexplanation,applyingtomorethanathirdofdropouts. Withouttheprotectionoftimeinschool,schoolclosuresmighthaveleftyoungwomenmore vulnerabletosexualabuse,limitingtheirabilitytoaccumulatehumancapitalinfuture.Figure 4Cexaminesthisdimensionofvulnerabilitybyshowingevidenceonthelikelihoodgirlsbecame pregnantduringtheepidemic.Thehorizontalaxisrepresentstime,inmonths,fromeitherMay 2014(thestartoftheepidemic)orMay2012(thestartofacounterfactualtwo-yearpre-epidemic period).Forthecrisisperiodsample,the rstverticaldashedlineshowswhenschoolsstartedto reopenandthesecondshowswhenSierraLeonewasdeclaredEbolafree.Attimezero,eachsample includesonlygirlsaged 12 - 17 inthereferenceyearthathadneverbeenpregnant.Thevertical axismeasuresthesharethatbecamepregnantovertimebymonth,ineachsample.Relativeto thepre-epidemiccounterfactual,pregnancyratesincreasesubstantiallyforthesegirls,withthe survivalfunctionsimmediatelystartingtodivergeanddoingsoevenlyoverthecrisis:alog-rank testofequalityofthesurvivalfunctionsrejectsthenulloftheirequality( = 001 ).Inthetwo- yearspre-epidemic, 10 %ofyounggirls
becamepregnant.Inthetwo-yearwindowofthecrisisthis risesbynearlyhalfagain,witharound 15 %becomingpregnant. Figure4Drepeatstheanalysisforgirlsaged 12 - 17 inthereferenceyearthatalreadyhad onechild.Weseeasimilardivergenceinsurvivalratesintheepidemicperiodrelativetothe counterfactual,butthemagnitudeislesspronounced.TakentogetherwithFigure4C,thissuggests theriseinpregnanciesduringtheepidemicwasmoreconcentratedamongyounggirlsthatdidnot havechildrenpre-epidemic. 4ModellingPersistentE¤ectsofSchoolClosures Wedrawtogetherthefeaturesofourcontextandinterventiontodevelopasimplemodelofgirl s timeallocationbetweenschool,ELAclubsandsocializing(includingtimewithmen).Thismakes precisehowschoolclosuresduringanepidemiccanhavepersistentimpactsonschoolingpost- epidemic.Intuitively,theprovisionofsafespacesduringanepidemicreducesthelikelihoodof becomingpregnant,thusincreasingthe intertemporal likelihoodofreenrollingintoschoolpost- 16 epidemic.However,post-epidemic,safespacessubstituteforother contemporaneous timeuses, includingschoolingandsoreduceenrolmentallelseequal.Thetreatmente¤ectofprovidingELA clubsonpost-epidemicschoolenrolmentcanbepositiveornegative,dependingonwhichofthese e¤ectsprevail.Themodelpinsdownmeasurabledimensionsdeterminingthisheterogeneityinthe signoftreatmente¤ectsonschooling,andwemapthisintoourresearchdesign. 4.1SetUp Eachperiod ,girlsallocatetheirtimebetweenschooling ,asafespace(i.e.ELAclubs) ,and socializing .Weuseathreeperiodmodeltomaptoourdatacollection:(i) =0 isthebaseline pre-epidemicperiod;(ii) =1 iswhentheEbolaepidemicoccurs,withschoolclosuresinplace; (iii) =2 istheendlineperiodafterschoolsreopen. Thestatevariable 2f 0 1 g capturesagirl sfertilitystatus. =1 indicatesshebegins period havinghadachild.PreferencesaredescribedbyaCESutilityfunction, ( )=[ + + ] 1 (1) wherethetasteparametersaresuchthat + + =1 andtheelasticityofsubstitutionbetween timeusesis =1 (1 ¡ ) .Weassumegirlsarenotpregnantat =0 ( 0 =0 ),andhaveatime endowmentscaledtounity: 0 + 0 + 0 =1 Attheendofthe rstandlaterperiods,agirlcan becomepregnant.Astimespentsocializingincludestimespentwithmen,eachunitof can resultinapregnancyatthe
endoftheperiodwithprobability 2 [0 1] : P [ +1 =1 j =0]= (2) P [ +1 =1 j =1]=1 wherewemodelpregnancy, =1 ,asanabsorbingstate.Hencebeingpregnantandhavinga childrepresentthesamestate,andwedonotmodelhavingmultiplechildren.Havingachild entailsatimecost 2 [0 1] ,astimeneedstobedevotedtochildcareandhouseholdchores. 18 Hencegivenfertilitystatus ,thetimeconstraintcanberewrittenas: + + =1 ¡ 18 Inourdata,theaveragerespondentwhoissingle,hasnochildrenandisaged 12 - 17 spends 38 hours/weekon householdchores.Thisrisesby 18 hourswhenthe rstchildisborn,butdoeschangewhenasecondandthird childrenareborn.Thisisinlinewithourmodellingassumptionof =1 beinganabsorbingstate,andnotneeding totracktheactualnumberofchildrenintermsoftimecostincurred. 17 4.2Baseline:NoEbolaEpidemic,NoELAClubs We rstensurethemodelmapstobaselinepatternsoftimeinschool(absenttheepidemicshock andELAclubs).Inthisbenchmark =0 8 sothetimeallocationproblemis: ( )=max ( 0 )+ E [ +1 ( +1 )] for =0 1 (3) 2 ( 2 )=max 2 2 ( 2 0 2 ) subjectto + =1 ¡ 2 [0 1] andtheconditionsin(2)Timeallocatedin =2 hasno intertemporalimplications,andsochoicesaredeterminedbytheFOCsettingthemarginalrate ofsubstitutionbetweenschoolingandsocializingequaltotheshadowcostoftime.Timeallocated toschoolandsocializingarethenjustsharesofthetotaltimeavailableat =2 : ¤ 2 ( 2 )= (1 ¡ 2 ) (4) ¤ 2 ( 2 )= (1 ¡ 2 ) wherethesharesarefunctionsoftheunderlyingtasteparameters: = + , =1 ¡ . Similarconsiderationsholdforpregnantgirls/thosewithchildren( =1 ).Giventhisisan absorbingstate,currentchoiceshavenointertemporalvalueandsothetimeallocationsare: ¤ (1)= (1 ¡ ) (5) ¤ (1)= (1 ¡ ) Incontrast,fornon-pregnantgirls( =0 ),timeallocatedtoschoolingin =0 , 1 generates owutilityandbydisplacingtimesocializing thatincludesspendingtimewithmen reduces thelikelihoodofbecomingpregnantandfacingatightertimeconstraintinthenextperiod.The FOCat =0 , 1 makesthisintertemporale¤ectclear: + [ +1 (0) ¡ +1 (1)]= =0 1 (6) Inthisbenchmarkscenario,thetimeallocatedtoschoolingandsocializingcanbesummarized throughthefollowingpolicyfunctionsfor =0 1 : ¤ 0 (0) ¤ 1 (0) ¤ 2 (0)= ¤ 0 (0) ¤ 1 (0) ¤ 2 (0)= (1)=
(1 ¡ ) (1)= (1 ¡ ) (7) Intuitively,themodelpredictsthatabsenttheEbolaepidemicandELAclubs,youngwomen stime 18 allocationtoschoolingdecreasessteadilyovertime,anddropspermanentlyoncetheytransition tomotherhood.The rstimplicationmapscloselytothecrosssectionalfallinenrolmentratesby ageshowninFigure3,andthesecondmapscloselytotheimportanceofpregnancyasareason fordropoutatbaselineshowninFigure4. 4.3Endline:TheEbolaEpidemicandELAClubs Wenowdevelopthemodelallowingfortheepidemicinperiod =1 andtheprovisionofELA clubs.Forexpositionalease,wefocusonthetimeallocationchoiceofgirlswithoutachildatthe onsetoftheepidemic( 1 =0 ),thatistruefor 88% ofoursampleaged 12 - 17 atbaseline. 19 Schoolclosuresduringtheepidemicimply 1 =0 .Toseehowthiscanpersistentlyimpact timeinschool,we rstnotethatduringtheepidemic( =1 ),incontrolvillagesgirlscanonly reallocatetheirtimetowardssocializing,whileintreatedvillages,anotheralternativeistospend timeatELAclubs.With 2f 0 1 g representingtreatmentassignment,theconstraint 2 [0 ] capturestheavailabilityofELAclubs.Agirl stimeallocationproblemcanthusberewrittenas: 1 (0)=max 1 1 (0 1 1 )+ E [ 2 ( 2 )] (8) 2 ( 2 )=max 2 2 2 ( 2 2 2 ) subjectto:(i) 1 + 1 =1 ;(ii) 2 + 2 + 2 =1 ¡ 2 ;(iii) 2 1 2 2 [0 1]; 1 2 2 [0 ] .Demand forsafespacesisdrivenbothbythecontemporaneousutilitygainsfromattendingELAclubs,and theintertemporalvalueofdisplacingtimewithmenandthusreducingthelikelihoodofpregnancy. TheFOCforthetimespentinclubsduringtheepidemicmakesthisclear: 1 + [ +1 (0) ¡ +1 (1)]= 1 (9) Wesolvethistoderivethedemandforschoolingpost-epidemicasafunctionoftreatment assignment,whereweusecapitalletterstodenoteexpecteddemand,andexpectationsaretaken overthedistributionofthestatevariable: Treatment: 2 = E [ ¤ 2 ( 2 ) j =1]= (1 ¡ (1 ¡ 1 )) (10) Control: 2 = E [ ¤ 2 ( 2 ) j =0]= (1 ¡ ) where = + ¸ = + + and = isthepregnancyriskyounggirlsface. 19 As =1 isanabsorbingstate,thetimeallocationchoicesofthosewhohavehadachildexhibitnodynamics andwillnotthereforequalitativelyimpacttheresults. 19 4.3.1TreatmentE¤ectofELAClubsonSchooling Thetreatmente¤ectontimeinschoolpost-epidemicis: = 2 ¡ 2 = ¡ (1 ¡ )( ¡ )+ 1
The rsttermisthe contemporaneous channel,wheretheshareoftimeallocatedtoschoolfalls becausesafespacesareafurthersubstituteforschooling( 0 ).Thesecondtermisthe intertemporal channelwheretheuseofsafespacesduringtheepidemic( 1 )reducesthelikelihood ofbecomingpregnant,relaxingthepost-epidemictimeconstraintandsoincreasingtimeinschool ( 0 ).Thesemovethetreatmente¤ectonpost-epidemicschoolinginoppositedirections. 20 Pregnancyrisk mattersfortimeinschoolinbothtreatmentandcontrolvillages,butto di¤erentextents.Iftheepidemicdi¤erentiallyshocksthedegreeofriskgirlsface,wehavethat: = µ 1 + 1 ¶ +( ¡ ) ¸ 0 Hencethetreatmente¤ectonschoolingincreasesinpregnancyrisk (weseethisfromtheFOCand thefactthat = implies 1 isincreasingin ).Therefore,whilethesignoftreatmente¤ects ofELAclubsontimeinschoolis apriori ambiguous,itincreasesinpregnancyrisk.Thisisakey predictionwetaketothedata. 4.3.2OtherConsiderations Themodelfeaturesdrivingtheresultsare:(i)safespacessubstitutetimefromschoolingand socializing( 0 ),(ii)thelikelihoodofbecomingpregnantisincreasingintheamountoftime spentsocializing( P [ +1 =1 j =0]= );(iii)pregnancyhasatimecostinfutureperiods ( 0 ).Whilethemodelcouldaccommodateadditionalfeatures,thesethreesimpleassumptions aresu¢cienttohighlightakeyresult:thesignoftreatmente¤ectsofELAclubsonschooling post-epidemicis apriori undetermined,butthesee¤ectswillbeincreasinginpregnancyrisk.The remainingassumptionsarestandardorserveexpositionalpurposes. 21 20 Thetreatmente¤ectcanbesignedintheextremecasesofeitherthereisnoriskofpregnancy( =0 )orno timecostassociatedwithhavingchildren( =0 ).Then = ¡ ( ¡ ) 0 ,asclubsonlyactasasubstitute forschooling. 21 Forexample,themodelemphasizestheconsumptionvalueofeachactivity,embodiedinthe tasteparameters. However,schoolingmightalsobeconsideredaninvestmentintohumancapitalgeneratingfuturereturns.Thesame mightalsobetrueforattendingELAclubsgiventheirprovisionoflifeskillsandvocationaltraining.Whilewe havenotmodelledlabormarkets,theseinvestmentreturnswoulde¤ectivelyjustchange( )parameters,but thefundamentaltrade-o¤scapturedinthemodelremain.Wereturntotheseissuesbelowintheempiricalanalysis. 20 5ModeltoDa
taandtheResearchDesign 5.1MappingtheModeltoData Tomapthemodeltodataweneedtoconstructavillage-levelmeasureofthepregnancyrisk younggirlsfaceduringtheepidemic, = .Thisis not thesameasthehealthriskofEbola, butitisnaturallycorrelatedtotheincidenceofEbola.Weconstruct usinginformationfrom thevillageleadersurveyadministeredbetweenJuneandOctober2015,whererespondentswere askedtorecallmonthlyinformationfromJuly2014onwhetherthelocalprimaryhealthunit (PHU)wasclosed,disrupted,andanoverallPHUfunctioningscore.Theywerealsoaskedabout whetherthenearestsecondaryschoolopenedontime(manyopeningsweredelayedfromApril 2015butallschoolswouldhavere-openedbythetimeofourendlinesurvey).Thesecomponents relatedirectlytoincreasesinpregnancyriskforyounggirlsandwomenbecause:(i)PHUclosures reduceaccesstocontraceptives;(ii)prolongedschoolclosuresincreasetheperiodoverwhichgirls areatriskfromunwantedsexualabuseordemandsfrommen. PHUsandsecondaryschoolsservemultiplevillagesoverawideradiusandsothefunctioning ofPHUsandschoolsduringthecrisisisunlikelytobeimpactedbythepresenceofELAclubsin treatedvillages,thatconstituteasmallshareofallvillagesinourfourdistricts(andofcourse,the controlvillagesalsoconstituteonlyasmallshareofallvillagesinoursetting,sothevastmajority ofvillagescoveredbythesamePHUsandschoolsarenotinourevaluationsample). 22 WecombinethesecomponentsintoanindexfollowingAnderson[2008].Table3showsdescrip- tiveevidenceoneachcomponentandtheoverallpregnancyriskindex.Eachcomponentvaries acrossvillages,withtherebeingsubstantialvariationwithinandbetweendistricts(Column2). PanelCshowstheconstructedindex,thatbyde nitionisstandardizedwithmeanzeroandstan- darddeviationone.Wede neavillagepregnancyriskdummyequaltooneiftheindexisabove its 75 thpercentile,andzerootherwise.Werefertotheseashighandlowriskvillages,where 17 % ofvillagesaresode nedtobehighpregnancyrisklocationsforyounggirls. 23 Figure5showstheregionalvariationinourriskindex,contrastingitwithgeographicvariation inEbolacases.ThecentralmapinFigure5showsWHOmeasuredcaseloadsacrossthe 14 districts 22 PHUscomprisethreetiers:communityhealthposts(CHPs)communityhealthcentres(CHCs),and,maternal andchildh
ealthposts(MCHPs).ThelowesttierPHUfacilitytypicallyservesapopulationofupto 5 000 withina 3-mileradiuscovering 10 to 20 villages.Inourevaluationsample,theaveragedistancetothenearestPHUis 1 55 miles.Similarly, 83 %ofvillageshavenosecondaryschoolinthemandthedistancetothenearestone(conditional onnotbeinginthevillage)is 4 91 miles. 23 Togetasenseofwhatthisclassi cationmeans,lowriskvillagesneverhavetheirPHUeverclosed, 18 %have PHUdisruptions,andtheiraveragePHUfunctioningscoreis 94 .Incontrast, 82 %ofhighriskvillageshavetheir PHUeverclosed,allofthemhavethePHUeverdisruptedandthePHUfunctioningscoreis 67 .Inlowrisk locations, 88 %ofsecondaryschoolsopenedontime,whilethisfallsto 62 %inhighrisklocations. 21 (o¢cialdataoncasesisunavailableata nerlevel).ThedistrictswiththehighestEbolacaseloads arePortLoko(closeto,butnotincludingthecapital,Freetown),andKailahun(wheretheinitial outbreakoccurred).TheouterdistrictmapsinFigure5showhowtheriskindexvariesover villages,bytreatmentandcontrol.Withineachdistrict,therearebothhighandlowpregnancy riskvillages.Avariancedecompositionoftheindexrevealsthatwithindistrictvariationaccounts forthemajority( 70 %)oftheoverallvariation. Todemonstratethat measureslocalizedvariationinthenatureofpregnancyriskforgirls, PanelAofFigureA4showswithindistrictpartialcorrelationsbetweentheindexandahost ofvillagelevelcovariates.We ndnosigni cantcorrelationswithnearlyalltheseothervillage characteristics,althoughrichervillagesareassociatedwithgreaterrisk.Allthesevillagecovariates includingthevillagePPIscore areconditionedonthroughout. PanelBofFigureA4showswithindistrictpartialcorrelationsbetweentheindexandhistoric measuresofhealthservicesorknowledge,measuredattheChiefdomlevelusingDHS2013data orthe2007NationalPublicSurvey.Theriskindexisalsouncorrelatedtothesepasthealthre- latedbehaviors,healthoutcomes,accesstohealthfacilities,andstatecapacityinhealth.Welater presentarobustnesscheckthatshowsthedocumentedheterogeneouse¤ectsofELAclubsonlearn- ingarespeci callyrelatedtovillage-levelpregnancyrisk ,andnotothervillagecharacteristics suchasvillagepoverty. Ofcourseourquantitativeresult
swillbesensitivetotheexactconstructionof .Weshow ourcoreresultsarequalitativelyrobustto:(i)exploitingthecontinuousmeasureofpregnancy risktoshowtheintensityoftreatmente¤ectsthemodelhighlights( );(ii)de ningpregnancy risktobewithin-district. 5.2ResearchDesignandIdentifyingAssumptions FollowingthesimpleintuitionofthemodelthattheimpactELAclubshaveonschoolingvarieswith pregnancyrisk,weadopta 2 £ 2 researchdesign,whereonedimensionistherandomassignment ofvillagestoELA, 2f 0 1 g andthesecondisthequasi-randomassignmentofvillagestohigh andlowpregnancyrisk duringtheEbolaepidemic.Foroutcome forindividual invillage indistrict weestimatethefollowingANCOVAspeci cation: = 0 + 1 + 2 + 3 ( £ )+ 0 + 1 + + (11) where 0 istheoutcomeatbaseline(wheneveravailable), isatreatmentdummy, isa dummyequaltooneifvillage iswheregirlsexperienceagreaterincreaseinpregnancyrisk 22 becauseoftheepidemic. and arecharacteristicsof andhervillage , aredistrict xede¤ects(therandomizationstrata),and isanerrorterm. 24 TheAppendixshowstherobustnessofourcore ndingsto:(i)usingrandomizationinference totestthenullofnotreatmente¤ects;(ii)adjustingformultiplehypothesistesting;(iii)only controllingfordistrict xede¤ects. Theparametersofinterestare:(i) b 1 ,thetreatmente¤ectofELAclubsinlowpregnancy riskvillages;(ii) b 1 + b 3 :thetreatmente¤ectofELAinhighpregnancyriskvillages;(iii) b 2 :the impactofresidinginahighpregnancyriskcontrolvillagerelativetoalowpregnancyriskcontrol village.Weneed veassumptionstoholdforthisdesigntoyieldcausalimpacts. First,thatELAclubsarerandomlyassignment.Tables1and2alreadyshowedbalance onobservables.Second,werequireindependenceof and .Toinvestigatethisweregress thepregnancyriskindexonthetreatmentdummy.Column3inTable3showsthesepartial correlations:nonearestatisticallydi¤erentfromzero.Thisresultcontinuestoholdwhen:(i)we alsoconditiononvillagecharacteristics(Column4);(ii)weallowthetreatmentdummytointeract withdistancesfromkeyfacilities(Column5);(iii)weallowformodelselectionusinganElastic Netpenalizedregression(Column6).Asshowninthe nalrowinPanelC,allresultscontinue toholdwhenweusethepregnancyrisk
dummy. Third,werequiretheretobenoselectionbiasduetothenon-randomincidenceofhigh pregnancyrisk.ToinvestigatethepossibilitybeyondwhatwasshowninFigureA4,TablesA4and A5comparecharacteristicsofvillagesandyoungwomenbetweenhighandlowrisklocations.On bothsetsofcharacteristics,thesamplesarewellbalanced.AssuggestedinFigure5,thewithin districtvariationinpregnancyriskisasgoodasrandom. Fourth,werequiretheextentofprogramimplementationtonotvarywithpregnancyrisk.The keyaspectofELAclubsisthattheyareopenandsoprovideasafespaceforgirls,andspend timeawayfrommen.Thisshowstheshareofclubseveropenedorcontinuouslyopened.Figure A5showsELAclubfunctioning,bypregnancyrisk.Inbothhighandlowriskvillages,atleast 60%ofclubsareopen,byDecember2014 closetothepeakoftheepidemic over 90 %ofclubs areopeninbothhighandlowriskvillages. 25 Finally,weneedtobeabletoruleoutastandardselectionongainsconcernarisingfrom 24 ThevillagecontrolsincludethoseshowninPanelAofTable1.Individualcontrols(measuredatbaseline)are age,thehouseholdpovertyscore,householdsizeandwhetherthegirlisilliterate. 25 Ofcourse,therearedi¤erencesinprogramcomponentsdelivered(intherelevanttreatmentarms).Wegenerally seethatlifeskillstrainingwaso¤eredtoagreaterextentinhighriskvillages.Vocationaltrainingappearstohave been rstrolledoutinlowriskvillages,butoncequarantineswerelifted,highriskvillagescaughtup.These di¤erencesarelessimportantgivethekeyrolethatELAclubsplayisintermsofprovidingasafespace,and moreover,bythetimeoftheendlinesurvey,allrelevantprogramcomponentshavebeendeliveredforoverayear inallvillages. 23 thefactthatthereturnstoELAmightdi¤eracrosspregnancyrisk.Onvariationinparticipant characteristics,TableA6comparescharacteristicsofELAparticipantsbypregnancyrisk.We nd noevidenceofdi¤erentialselectionintoELAacrossriskintensities.Participantsacrosshighand lowriskvillagesdonotdi¤erinrelationshipormaritalstatus,whethertheyexperienceintimate partnerviolence(andsomightbepreventedfromattendingELAclubs),thevariousmeasuresof empowerment,humancapitalandengagementineconomicactivitiesatbaseline. 6Results:YoungGirls 6.1TimeUse We rstexaminetimeuseacrosswingbroadactivities:ELA
clubattendance,learning,work- ing(incomegeneratingactivities),householdchoresandsocializing.Learningcorrespondstoall education-relatedactivitiesexceptELA(formalschooling,othervocationaltraining,church-based schools).Toaccountforinterlinkedtimeallocationsacrossactivities,weestimateaSURmodel allowingerrortermstobeclusteredbyvillage.Allestimatesincludezeroesandsoareinterpreted asthetotale¤ectsmargin.Figure6Asummarizestheresultswhereeachactivityshowsthepara- metersofinterest:(i) b 2 (red):thepregnancyriske¤ect,namelytheimpactofresidinginarisk controlvillagerelativetoalowriskcontrolvillage;(ii) b 1 + b 3 (darkblue):thetreatmente¤ect oftheavailabilityofELAclubsinhighriskvillages;(iii) b 1 (lightblue):thetreatmente¤ectof theavailabilityofELAclubsinlowriskvillages.Weshowthe 90 %con denceintervalforeach estimate.Columns1to5inTable4showsthecorrespondingSUREresults.Atthefootofthe tablewereportthep-valueonthenullthat 3 =0 ,soestablishingwhetherthetheprovisionof ELAclubsafespaceshasadi¤erentialimpactbypregnancyrisk( ). The rstcategoryoftimeuseshowninFigure6AistimespentatELAclubs.Inallvillages, younggirlsdevotearound 3 hrs/weekatELAclubs.Thisisaplausiblemagnitudeandinlinewith theearlierresultsonparticipation.Consistentwiththerebeingnoselectionongainsforyoung girls,thetimespentatELAclubsisthesameinlowandhighriskvillages( = 528 ). TheremainingColumnsshowstarkimpactsontimeallocationofresiderinahighpregnancy riskvillage:movingfromalowtoahighriskcontrolvillageisassociatedwithasigni cantdecline intimespentlearning.Themagnitudeofthisis 12 2 hrs/wk,correspondingto 25 %ofthebaseline meaninlowriskcontrolvillages.Thistimeisreallocatedtowardswork/incomegeneration(that increasesby 6 07 hrs/wk,or 40 %),andhouseholdchores(thatincreasesby 5 62 hrs/wk,or 13 %). TheavailabilityofELAclubslargelyreversesthistimereallocationrelatedtohighpregnancy risk:ELAo¤setsthisby 9 84 hrs/wk(or 81 %ofthee¤ectofhigherpregnancyrisk).Thistime 24 comesfromreducedexposureofyounggirlstotimespentonhouseholdchoresby 4 68 hrs/wk. Hence,inhighriskvillagestimespentlearningandatELAclubsmovetogetherpost-pandemic bothincrease.Inthelanguageofthemode
l,theintertemporale¤ectofELAclubsoutweighsthe contemporaneouse¤ectinvillageswithhighpregnancyriskduringtheepidemic.Consistentwith themodel,theresultsinTable6showtheoppositeistrueinlowriskvillages:inthosevillages theevidencesuggeststheavailabilityofELAclubsfurtherreducestimespentlearninginformal institutionspost-epidemic,by 3 03 hrs/wk(or 25 %ofthee¤ectofhigherpregnancyrisk). Couldtheseresultsbepickingupsomeothervillagecharacteristicunrelatedtopregnancy risk?TableA7showsthedocumentedheterogeneouse¤ectsofELAclubsontimespentlearning arespeci callyrelatedtoourmeasureofvillage-levelpregnancyrisk ,andnotothervillage characteristics.Moreprecisely,Column1repeatsourbaselineresultsontimespentlearningusing thevillagelevelpregnancyindex.Columns2to4thenshowsheterogeneouse¤ectsofELAby villagepoverty,whereeachColumnusesanalternativethresholdtodividevillagesintohigh/low poverty.We ndnoevidencethatELAclubshavedi¤erentiale¤ectsontimespentlearningin richerandpoorervillages.TheremainingColumnsshowequallyweakheterogeneousimpactsof theavailabilityofELAclubsontimespentlearningacrossbyvillageinfrastructure,ordistance frominfrastructuresoutsidethevillage.Inshort,theevidencesuggeststhatvillagelevelpregnancy riskisthekeyvillagecharacteristicdrivingheterogeneousimpactsofELAclubs. ThelastcategoryoftimeuseshowninFigure6Arelatestotimespentsocializing.Herewesee thatinallvillages,timespentsocializingsigni cantlydecreases.Someofthiswillbereallocated toELAclubs(wheregirlscansafelymeetothers)butwenextexplorewhetherthisreallocation ofyounggirlstimealsoallowsthemtosubstitutetimeawayfrommen,andultimatelyincreases theiragencyovertheirbodies. 6.2SocializingandTimeSpentwithMen Wesplittimespentsocializingintothatspentwithmen(wherethesurveyquestionrefersprecisely totimewithmentheyaresexuallyactivewith),withfriends,alone,andingroupactivities(such asvolunteering/church).Throughout,thislastcategorymightserveasapartialeuphemismfor secretsocieties,thatcannotbeaskedaboutdirectly.Weestimate(11)acrossformsofsocial activityusingaSURspeci cation.Figure6Bsummarizestheresults,Columns6to9inTable4 showtheunderlyingregressionresults. Weseedramat
icchangesforhowyoungwomenincontrolvillagesspendtimesocializing: centraltothisischangesintimespentwithmen.Inhighriskcontrolvillages,younggirlsare signi cantlymoreexposedtomenforsexualrelations:post-epidemicthetimespentwithmen 25 increasesby 1 27 hrs/wk,a 50 %increaseoverthebaselinemean.ThepresenceofELAclubshelps tolargelyo¤setthisincreasedexposuretomen.Theydososigni cantlyinallvillages,withthe magnitudebeinggreaterinhighriskvillages( = 012 ).Theseo¤settinge¤ectsare 1 86 hrs/wk ( 60 hrs/wk)inhigh(low)riskvillages.These ndingsreinforcethenotionthatELAclubsprovide asafespacethatprovideanalternativetowhereyoungwomenspendingtimewithmenwith whomtheyaresexuallyactive,andthisismoresoforgirlsinhighriskvillages. Theremainingcategoriesofsocializingshowthat:(i)increasedpregnancyriskincontrol villageshasnoimpactonotherformsofsocializingexceptspendingtimewithmen;(ii)inboth highandlowriskvillages,theavailabilityofELAclubsreducetimeyounggirlsspendwithfriends, alone,andingroupactivities(volunteering/church).Ofcourse,asFigure6Ashowed,thesetime reductionsarepartlyo¤setbytimespentwithotheryounggirlsatELAclubsinstead. 6.3SexualActivityandTeenPregnancy Howdoallthesereallocationsoftimetranslateintopost-epidemicoutcomesrelatedtosexual activityandpregnancy?Column1ofTable5showsimpactsonthefrequencyofunprotectedsex (combiningself-reportsonthefrequencyofsexualintercoursewithcontraceptiveuse). 26 Thisrises withourmeasureofhigherpregnancyriskincontrolvillages.However,thisimpactonunprotected sexismostlyo¤setinhighriskvillageswhereyounggirlshaveELAclubsavailabletothem. Fromtheperspectiveoflifetimewelfare,thekeyoutcomeforgirlsisteenpregnancy.Onthe frequencyofpregnancies between baselineandendline,Column2showsthatresidinginahigh riskvillageisassociatedwitha 10 5 ppincreaseinthelikelihoodofbecomingpregnantbyendline. Hencetheepidemicspeedsuptransitionsintochildbearingforteenagersinexactlythoselocations wherehealthserviceprovisionhascollapsed,anddangersduringchildbirthtogirlsarelikelyto beevenmoreseverethanpre-epidemic. Column3showstheincreaseinteenpregnancyisdrivenbyout-of-wedlockpregnancies:absent ELAclubs,theseriseby 7 ppinhighris
kcontrolvillagesrelativetolowriskvillages.However, thisimpactonout-of-wedlockbirthsiscompletelyreversedinhighvillagesinwhichyounggirls havethesafespaceofELAclubsareavailable. 26 Around 25 %ofgirlsreportbeinginarelationshipatbaseline.Hencechangesovertimeonthesemargins re ectboththeconsequencesofnewlyformedrelationships,aswellaschangesinbehavioroverthecrisisfor intactrelationships.Resultsfromthelongtermfollow-upsurveydiscussedbelow,shedmorelightonthechanging characteristicsofpartnersthesegirlsendupinrelationshipswith. 26 6.4SchoolEnrolment Wenowcompletethecausalchainlinkingtemporaryschoolclosuresduringtheepidemictolonger termimpactsontheschoolingofyounggirls,ane¤ectexacerbatedbythepolicythat visibly pregnantgirls couldnotre-enrolevenonceschoolsre-opened.Column4ofTable5con rmsthis link:weseethatmovingfromalowtoahighriskcontrolvillageisassociatedwithadramatic fallinschoolenrollmentratesof 17 pp.Toreiterate,thisfallismeasuredwellafterthecountry isdeclaredEbolafreeandschoolshavereopened.Columns5and6shedlightonthekindsof economicactivitythesegirlsengageininstead:theydonotcombineschoolandworkbutthereisa signi cantriseinexclusivelyengaginginworkby 19 pp.Inshort,absentELAclubs,theepidemic signi cantlyacceleratestheschool-to-worktransitionforyounggirls. 27 ELAclubsdramaticallycountertheschool-to-worktransitionforyoungwomeninhighrisk villages.InvillagesrandomlyassignedanELAclub,thefallinenrolmentpost-epidemicishalved, to 8 pp.Indeed,themagnitudeofthefallinout-of-wedlockpregnancies( 7 5 pp)closelymatches theriseinschoolenrolment( 8 5 pp)inhighriskvillageswithanELAclubsavailable.Thereason whyELAclubshelpgirlsre-enrolpost-crisisinhighriskvillagesisthatthesesafespaceshelp themavoidspendingtimewithmen,avoidout-of-wedlockbirths,andthustheyarenotbarred fromre-enrollinginschoolpost-epidemic.Thisisallinlinewiththeearlierresultsontimespent learning.Inthelanguageofthemodel,theintertemporale¤ectofELAclubsoutweighsthe contemporaneouse¤ectinvillageswithhighpregnancyriskduringtheepidemic. Intreatedhighriskvillages,weobservealowerlikelihoodthatyounggirlstransitionexclusively intoworkpost-epidemic.Moreover
,inthosevillagestheavailabilityofELAclubsalsoenables younggirlstocombineschoolandworkpost-epidemic(Column6) by 9 7 pp. Consistentwiththeearlierresultsontimespentlearning,we ndthatinlowriskvillages, theavailabilityofELAclubsfurtherreducesschoolenrolmentby 5 2 pp.Inthesevillagesthe evidencesuggestsELAclubssubstituteforschoolspost-epidemicsothatthecontemporaneous e¤ectdominatestheintertemporale¤ect. Foralloutcomesrelatedtoenrolment,workingorcombiningthetwo,theimpactofELAclubs signi cantlydi¤ersforhighandlowriskvillages( b 3 6 =0 ineachcase, = 020 , 052 ,and 020 ). 27 Themovementintoincomegenerationforyoungergirlsisdrivenbytheirengagementintoself-employment. Thisisnotsurprising,giventhatmostwomenareengagedinself-employmentactivitiespre-crisis[Casey etal. 2016],andwageemploymentopportunitieswereseverelycurtailedduringthecrisis. 27 6.5OtherOutcomesandRobustness TheAppendixpresentsresultsonotheroutcomesandrobustnesschecks.First,weexamine whetherinhighriskvillagesELAclubsraisetherelativereturnstoschoolattendancebecause theyprovidecomplementaryskillsthroughtheirinterventioncomponents.AsTableA8shows, thepatternofimpactsonliteracyandnumeracyskills bypregnancyrisk,theprovisionofELA clubsandtheirinteraction followexactlythepatternsfoundforschoolenrolment.Incontrast, we ndweakstatisticalevidencethatELAspeci cskills suchasthoserelatedtoentrepreneurial con denceor nancialliteracy areimpacted.ThisnarrowstheinterpretationofELAclubs impactingtheeconomiclivesofyoungwomenoverthecourseoftheEbolaepidemicbecausethey o¤erasafespacetoyounggirls. 28 Second,wesawearlierthattheavailabilityofELAclubsarecauseyounggirlstoreallocate timespentwithfriendsorengagedinsocialactivities.Thiswassoirrespectiveofthepregnancy riskgirlsfacedduringtheepidemic.Thereductionintimespentwithothercanhaveoneof twointerpretations.GirlsmightbesubstitutingthiswithspentatELAclubs,ortheremight beanoverallweakeningofsocialtiesbetweengirlsthatpersistspost-crisis.Giventhedi¤erent implicationsofeach,weexaminethisfurtherbyexploitinganetworksurveymodule eldedat baselineandendline.Thisaskedresponde
ntsaboutthenumber(andidentity)oftiestoothergirls inthevillagealongfourdimensions:friendship,forbusiness(incomegeneration),whointimate topicsarediscussedwith,andforcredit/ nance.TableA9showstheprovisionofELAclubs curbsanylossinyounggirl ssocialtiesalongmultipledimensionsinhighriskvillages. 29 Such maintenanceofsocialstructureshasbothmicroandmacroimplicationsinthelongrun[Fogliand Veldkamp2019].These ndingsaddtoanascentliteratureonhowdevelopmentinterventions impactthestructureofnetworks[He etal. 2020]. Third,weexaminewhetherELAclubshelporganizeandcoordinateactionstocontrolthe epidemic.TableA10examinesimpactsonreportedcasesofEbola.Weelicitedinformationabout whetheranyEbolacaseshadoccurredinthehousehold,extendedfamilyorfriendshipnetwork.We generally ndnegativepointestimatesofELAclubsinlowandhighpregnancyriskvillages,but thereareonlysigni cantdeclinesinreportedEbolacasesbyyoungergirlsamongtheirextended familynetwork(by 4 5 ppor 30 %)andfriendshipnetworks(by 5 0 ppor 28 %). 30 28 Theseskillsmeasuresareconstructedareconstructedfromvariousself-reportedabilities.Thesearethenag- gregatedandrescaledtorangefrom0to1,withthelatterindicatingmoreadvancedpro ciency.Entrepreneurial Con denceisanindexthatmeasuresrespondents self-reportedabilityto:runabusiness,identifybusinessoppor- tunities,obtaincredit,saveandinvest,manage nancialaccounts,bargainprices,manageemployeesandsearch forjobs.FinancialLiteracyisassessedthrougheightsimpleproblemsrelatingtomarketprices,interestrates, borrowingandbudgeting.Thenumberofcorrectanswersisrescaledinanindexrangingfrom0to1. 29 Givennetworktiesarecensoredatzero,weestimate(11)usingaTobitmodel. 30 Theseimpactsareidenti edfromtheendlinecross-sectionbecauseonlythendidweaskrespondentsabout 28 Finally,intheAppendixweexaminetherobustnessofourcore ndingsto:(i)usingran- domizationinferencetotestthenullofnotreatmente¤ects;(ii)adjustingformultiplehypothesis testing;(iii)onlycontrollingfordistrict xede¤ects.Wealsoshowourcoreresultsarequalitatively robusttotwoalternativeconstructionsofthepregnancyriskindex:(i)exploitingthecontinuous measu
reofpregnancyrisktoshowtheintensityoftreatmente¤ectsthemodelhighlights( ); (ii)exploitingonlywithin-districtvariationinpregnancyrisk. 7ExtendedResults 7.1OlderGirlsandWomen Wenowexamineafocusedsetofoutcomesforoldergirlsandwomen,thatwereaged18-25at baselinein2014:whilethiscohortwillbelessimpactedbyschoolclosuresgiventheirlowerlevels ofenrolmentatbaseline(asFigure3Bshowed),theywillberelativelyharderhitbythelarge andrapidlossofeconomicopportunitiesduetotheepidemic.Theyarestillsubjecttothesame heightenedexposuretomenandpregnancyrisksduringtheepidemicasyoungergirls,andcan alsogainfromtheprotectivesafespacesthatELAclubsprovide. Figure7summarizestimeuseimpactsonthisuseforthisoldercohort.FromPanelAwesee thatinallvillages,thiscohortdevotesaround 3 hrs/weekatELAclubs,inlinewiththeresults foryoungergirls.Consistentwiththerebeingnoselectionongains,thetimespentatELAclubs doesnotdi¤erwithvillage-levelpregnancyrisk( = 741 ).TheremainingbarsinPanelAagain showapatternofprotectivee¤ectsofELAinhighriskvillages:theprogramhelpsmaintaintime spentlearning,andkeepsoldergirlsawayfromdevotingtimetohouseholdchores.Oldergirlsin treatedvillagesalsospendlesstimesocializing:thisfallsby 2 97 and 4 01 hrs/wkinlow(high) riskvillages,correspondingto 10 %( 14 %)reductionsrelativetothebaselinemean. PanelBbreaksdownwaysinwhichtheyspendtimesocializing.Aswiththeyoungercohort, incontrolvillages,heightenedpregnancyriskisassociatedwithspendingmoretimewithmen. However,theavailabilityofELAclubsagainprovidesanalternativetimeuse:theseo¤setting e¤ectsare 1 65 hrs/wk( 1 32 hrs/wk)inhigh(low)riskvillages. Table6nextexaminesimpactsonsexualactivityandpregnancy.Linkingdirectlytomen s actions,Columns1and2examinethefrequencyofunwantedortransactionalsex,asreported tohaveoccurredinthepastyear.Thiscoverstheperiodoftheepidemicsopotentiallycapture Ebolacases,andsowedonotcontrolforbaselineoutcomes.Toimproveprecision,thesampleisthesamesetof individualstrackedfrombaselinetoendlineusedthroughout,butwealsoaddinrespondentsfromtherefresher samplecollectedatendline,whowereresidentinthevillagesincebeforetheoutbreak.Theresultsarenearidentical notusingthissecondgroup. 29 changest
hatstartedtooccurduringthecrisis. 31 InhighriskvillageswithELAclubs,oldergirls reportsigni cantlymoreunwantedsexandtransactionalsex.Thesearelargeincreases:unwanted sexincreasesby 5 4 pp,correspondingto 35 %ofthebaselinemean;transactionalsexincreasesalso byexactly 5 4 pp,correspondingto 115 %ofthebaselinemean. Columns3and4showthatELApreventsthesebehavioralchangestranslatingintohigher fertilitybecauseoldergirlsincreasetheiruseoffemalecontrolledcontraceptives.Thisimpact ofELAisobviouslyreassuringintheshortterm,asdelayingchildbearingcanhelpimprovethe lifetimewelfaretrajectoryofwomen.Italsoshowsthatbehavioralchangeamongoldergirlsis consistentwiththemtakingonboardsomethelifeskillsprovidedbyELAinhighdisruption locationsandbargainovertheuseofcontraceptives[Ashraf etal. 2014].Thesechangesin contraceptiveuseallowthemtoo¤setsomeoftherisksfromengagingintransactionalsex. 7.2ExperimenterDemandE¤ects Anobviousconcernisthatresponsesmightbedrivenintreatedvillagesbyexperimenterdemand e¤ects.Thisinterpretationdoesnoteasily ttheresultsbecausethetreatmente¤ectsofELA varyacross:(i)agecohorts;(ii)epidemic-relatedpregnancyriskfordi¤erentoutcomesacross cohorts.FollowingDhar etal. [2020],weaddresstheconcernusingdatacollected(inthelong termfollowup)ontheMarlowe-Crownesocialdesirabilityscale,a13-questionsurveymodule developedbysocialpsychologiststomeasureanindividual spropensitytogivesociallydesirable answers[CrowneandMarlowe1960].Weconvertrespondents socialdesirabilityscore(i.e.,the countofverypositivetraitstheysaytheyhave)intoanindexusingthemethodsetoutin Anderson[2008],thataccountsforthecovariancestructureintheunderlyingquestions.Ahigher indexsigni esanindividualthatseeksmoresocialapproval. TableA12presentstheresultswherewefocusonoutcomesmostsubjecttosocialdesirability bias:(i)timespentsocializingwithmen(forbothcohorts);(ii)theincidenceofunwantedand transactionalsex(fortheoldercohort).Foreach,wereportourbaselinespeci cationincluding theMarlowe-Crownindex(inthesmallersampletrackedinthelongtermfollowup),andhowthe resultsvarybythoseabove/belowthemedianofthesocialdesirabilityindex.Weseethat:(i) thereisno
signi cantrelationshipbetweentheindexandtheseoutcomes;(ii)fortimespentwith mentherealsoappearstobenointeractionofthisindexwithtreatmentforeithercohort;(iii) forunwantedortransactionalsex,itisyoungwomenthathaveabelowmedianindex andso seeklesssocialdesirability thathavethelargerincreasesintheseoutcomesintreatedvillages, 31 Whenaskingabouttransactionalsex,wementionmultipleformsofin-kindgiftsthatmightbeprovidedby partners,includinghelpwithschoolfees.Thishaslongbeenarguedtobepartoftransactionalsexualarrangements inplaceforyoungergirlsinthiscontext[Bledsoe1990]. 30 suggestingifanythingwemightbeunderestimatingchangesinsuchoutcomesamonggirlsthat seektoprovidesociallydesirableanswerstoenumerators. 7.3InterlinkedOutcomesAcrossCohorts Drawingtogether ndingsacrossagecohorts,thecommonimpactofELAclubstoallowgirlsand youngwomentoallocatetimeawayfrommen.Fortheyoungercohortaged12-17,thisleadsto lessunprotectedsex,lowerratesofout-of-wedlockfertility,thatthenmapalmostcompletelyto higherratesofre-enrolmentinschoolpost-epidemic.Fortheoldercohorthowever,theavailability ofELAclubsinhighriskvillagesleadstomoretransactionalsex.AsinDupasandRobinson [2012]andmanyothercrisiscontexts,engagementintransactionalsexbyoldergirlsmightthen representaformofincomegenerationinatimeofaggregatecrisis,wheneconomicopportunities arecurtailedandothercopingmechanismstosmoothconsumptionareunavailable. 32 Anaturalquestioniswhethertheseimpactsacrosscohortsareinterlinked?Thisdepends fundamentallyonwhetherthetypesofmenwhowouldseektransactionalsexwitholderwomen would rstseekitwithyoungerwomen(whowouldnotreportitastransactional).Ifnot,theim- pactsarenotinterlinked,butcouldratherre ectthattheresultsre ectincreasedentrepreneurial activityoranincreaseinrisk-takingbehavioramongtheoldercohortinhighriskvillages. 33 Ontheotherhand,thealternativeinterpretationisthatbyprotectingyoungerwomenin treatedvillages,apotentialconsequenceisthatmenshiftattentiontooldergirls.Thisiscom- poundedbythefactthateconomicopportunitiesformencollapseduringtheepidemic,sothey havefewerworkrelatedactivitiestodevotetimetowards.Theseforcescanplaceupwardpressur
e onthepriceoftransactionalsexinthesevillages.AsFigure4Bshowed,engagementofwomen intransactionalsexoccursevenpre-epidemic,andincreasessteadilywithage.Asthepriceof transactionalsexrises,itbecomesmorepro tableforoldergirlstoengageinsuchbehaviors. Irrespectiveoftheinterlinkageacrosscohorts,twoaspectsrelatedtotheuseoftransactional sexinELAvillagesaresomewhatreassuringthough.First,thisbehaviorismatchedbyincreased contraceptiveuseamongtheoldercohort,thathelpsprotecttheirreproductivehealth[Shah2013]. Second,we ndnoevidencethatyoungerwomenlearn/imitatethebehaviorofolderwomenin movingintotransactionalsex. 34 32 Entryintosexworkhasbeenarguedtobeacopingstrategyforwomenintimesofothereconomiccrisis:such aspost-WW2inGermany,ItalyandJapan[BulloughandBullough1987],duringthe1930sdepressionintheUS [Allen2004]andin1990sRussia[Atlani etal. 2000]. 33 Riskpreferencescanbeimpactedbyaggregateshocks[Callen etal. 2014,CameronandShah2015].Muchof thisliteratureisconsistentwiththeconceptofriskvulnerabilitysoshocksleadto higher levelsofriskaversion,as individualsupdatebeliefsoverthebackgroundrisktheyface[GollierandPratt1996]. 34 AsShah[2013]discusses,researchshowsthatsexworkersinlow-incomesettingsarepaidsubstantialpremia fornon-condomsex[Rao etal. 2003,Gertler etal. 2005,RobinsonandYeh2011],andthisriskpremiumisbest 31 7.4LongTermFollowUp BetweenJune2019andJanuary2020weconductedalongtermfollowupofourrespondents, fouryearsaftertheendoftheEbolaepidemic.Weusethisexaminewhetherthemostimportant impactsdocumentedpersistasthemacroeconomyiswellonthepathofrecoveryfromtheepidemic. Inaddition,thelongtermfollowupgaveustheopportunitytointerviewmenforthe rsttime. Weselectedpartnersoftrackedrespondents,andsoshedlightonwhethertheprovisionofELA clubs,byprotectingwomenduringtheepidemics,hasconsequentimpactsonthecharacteristics ofpartnerstheyformrelationshipswithinthelongrun. Atthetimeofthelongtermfollowup,ELAclubswerenoto¤eringanyskillstraining,although inmanycasestheycontinuedtoexistasaclubwheregirlsandyoungwomencouldspendtime witheachother.Second,amonggirlstrackedtoendline,wehadfundingtosurvey 71 %ofthemin thislongtermfollowup( 3401 respondents).
Wemanagedtotrack 84 %ofthisintendedsample, correspondingto 2852 respondents. 92 %oftheserespondentsreportbeinginarelationship,and wewereabletosurvey 1368 partners.Henceforpartneroutcomes,toimprovepowerwefocuson thelongtermimpactsoftreatmentwithoutconsideringheterogeneitybyvillagepregnancyrisk duringtheepidemic. 7.4.1YoungCohort Table7showsthelongtermimpactsongirlsaged12-17atbaseline,usingaspeci cationanalogous to(11).Wefocusontimeuse,schoolingandpregnancyoutcomes.Remarkably,we ndpersistent impactsincontrolvillagesofpregnancyriskduringtheepidemic.Thoseinhighriskvillages spendalmost 7 hrs/wklessonlearningactivities.However,equallyreassuringisthefactthatthe impactsofELAclubsalsopersistinthelongrun:theELAtreatmente¤ectinhighriskvillages increasestimespentlearningby 7 8 hrs/wk,itdecreasesitforthoseinlowriskvillages,andthe di¤erencebetweentheseissigni cant( = 033 ).Tobeclear,thiscohortofgirlsareaged 17 to 22 inthislongtermfollowup,sotheyarestudyingatthehighesttiersoftheformaleducation system(evenwithstandingayearoflosteducationbecauseofschoolclosures). Table7showstheotherpersistentimpactsofELAclubsisyounggirlsspendinglesstime socializing(Column2),wherethisisdriven veyearslater bythemspendinglesstimewith men,andlesstimeingroupactivities(volunteering/church). Thisalltranslatesintopersistentimpactsonschoolandpregnancyrelatedoutcomes.In particular,thoseinhighriskcontrolvillageshaveenrolmentrates 11 pplowerthanthoseinlow riskvillages,andarealmost 15 ppmorelikelytohavebecomepregnantsincebaseline.This understoodasacompensatingdi¤erentialforincreaseddiseaserisk[ArunachalamandShah2008]. 32 suggestspregnanciesrelatedtotheEbolaepidemicdonotrepresentmerechangesinthetiming ofbirth( tempo e¤ects),atleastoverthewindowtothelongtermfollowup. Thetreatmente¤ectofELAclubsalsopersistentlydi¤ersforgirlsinhighandlowriskvillages: thoseinhighriskvillageshaveenrolmentratesalmost 14 pphigherthanthoseinlowriskvillages ( = 034 ),andtheirpregnancyratesare 13 4 pplower( = 074 ). 35 7.4.2OlderCohort Table8showslongtermoutcomesin2019/20forthoseaged18-25atbaselinein2014.Wefocus onthesameoutcomesrelatedtoriskybehavi
orsandpregnancyshownearlier.Incontrolvillages, therearelongrunimpactsonthelikelihoodofbecomingpregnantofhavingbeeninhighpregnancy riskvillageduringtheepidemic( 11 2 pp).Asforyoungergirls,thissuggestspregnanciesrelated totheEbolaepidemicdonotrepresentmerechangesinthetimingofbirth. We ndthatwhiletherearenopersistentchangesinthefrequencyofunwantedsex,forthose intreatedvillagesthatfacedhigherpregnancyriskduringtheepidemic,thereisstillasigni cantly higherincidenceoftransactionalsexin2019/20,andisremainssigni cantlydi¤erentbetweenhigh andlowriskvillages( = 075 ).Howeverwecontinuetoseethatthisdoesnotleadtoincreased pregnancies,becauseoflonglastingchangesinfemale-controlledcontraceptiveuse. 36 7.4.3Partners Arecurringthemthroughouthasbeenthemanydisadvantagesyounggirlsandwomenfaced inthiscontextevenpre-epidemic eitherthroughseveregenderinequalitiesinlabororhealth (FigureA1),deeprootedprejudicesheldagainstpregnantwomen(Table1),ortheiryoungages ofmarriage,teenpregnancyandbeingsubjecttointimatepartnerviolence(Table2). Inouroriginaldatacollectionweneverplannedtosurveymen,andour ndingscanonly hintathowtheirbehaviorwasimpactedduringtheepidemic,whentheireconomicopportunities disintegrated.Our ndingshintatsomemale-drivenoutcomeschangingoverthecrisis:most obviouslytheincreaseinpregnanciesinhighriskcontrolvillages,andtheincreaseinunwanted andtransactionalsexamongoldergirlsintreatedvillages.Thelongtermfollowupgivesan opportunitytosurveymen(partnersofourrespondents).Givensamplesizes,weincreasepower 35 Togaugewhetherthisisplausible,wereturntoFigure4A,andnotethatatbaseline, 40 %of17yearoldswere enrolledinschool,and 11 %of18+yearoldswereinschool. 36 Earlierworkhasshownthelongrunreturnstovocationaltraininginthiscontent[Alfonsi etal. 2020].For bothcohortswehavethusalsoexaminedwhethertherearelongrunimpactsonincomegeneratingactivities:we ndlittlerobustevidenceofsuchimpactsoneithertheextensiveorintensivemargins.Thisisnotaltogether unsurprisinggiventheevidenceearlierdocumentinghowsuchprogramcomponentswereseverelycurtailedduring theepidemic,andsoELAclubsprimarilyservedassafespaces. 33 byfocusingonestima
tingITTe¤ectsofELAclubs(nothowtheyvarywithpregnancyriskfor womenduringthecrisis).Wedonotcontrolforbaselineoutcomesastheseareunavailable. Table9showsthecharacteristicsofpartnersatthelongrunfollowup,ofbothcohortsofgirls andyoungwomen.Onimportantdimensions,partnersintreatedvillageshaveimprovedtraits thanthosepartneredtogirlsandyoungwomenincontrolvillages:theyarebettereducatedand moreaversetogenderbasedviolence.However,theyarenodi¤erentintermsofageorgender norms(fortheyoungercohortofgirls),ormightevenbemoreconservative(fortheoldercohort). Whileonlysuggestive,theseresultshintatthematchingprocesschangingasaresultofthe availabilityofELAclubsduringthecrisis,whenwomenweremostatrisk.Thismightrepresent anadditionalmarkerofhigherlongrunwelfarefortreatedgirlsandyoungwomen. 37 8Discussion Virusesareamajorthreattohumanhealth:overthelastcentury,moredeathshavebeencaused byvirusesthanallarmedcon ictcombined[Adda2016].Giventhelongrunincidenceofhighly infectiousdiseasesisdeterminedbyurbanizationdrivingclosercontactbetweenhumanandanimal populations,andrisingglobaltemperatures,wecanexpectthemtoremainaglobalthreatfor theforeseeablefuture.Understandinghowsuchepidemicshocksimpacttheeconomiclivesofthe vulnerableisoffundamentalimportance.Inthecontextofgenderequality,hard-earnedgainsin women sempowermentcanbequicklyerasedbyaggregateeconomicshocks,anditisintimesof greatestcrises,thatgenderdi¤erentialsinoutcomesaremostlikelytoopenup[Du o2012] an issueattheforeofpolicydiscussionsinallcountriesinthecurrentpandemic.Inthispaper,given thenatureoftimingbetweenourpre-plannedevaluationofELA,andthecoincidentaloutbreak ofEbolajustasourbaselinesurveywascompleting,ourstudypresentsauniqueopportunityto understandthemicroeconomicmechanismsthroughwhichthiskindofaggregateepidemicshock impactstheeconomiclivesofyoungwomen.Ourstudyprovidesthreebroadlessonsforpolicy. First,akeyreasonwhyschoolclosureshavepersistentimpactsongirlsisbecauseofthepolicy thatpregnantgirlscannotre-enrolbackintoschool.Stigmatizationanddiscriminationagainst pregnantgirlsremainsapervasivebarriertothemresumingeducationthroughoutSubSaharan Africa.CountriessuchasUgandaand
Tanzaniahavesuchexplicitrestrictionsinplace,while othersretainambiguouspolicystatementsontheissue[Birungi etal. 2015].Itisthereforehugely signi cantthatasSierraLeonewasstruckbyitssecondmajorviralepidemicin veyears,the 37 We ndnoevidenceofanysofteningofattitudestowardspregnantgirlsinthelongrunfollowupamongvillage elders.Inarepeatcommunitysurveytovillageeldersin2019/20,we ndasimilarpatternofattitudesaswe reportedinTable1:forexample, 95 %ofthemstillagreethatvisiblypregnantgirlsareabadin uence,andthat theyshouldnotbeallowedbacktoschool. 34 governmentannounceditwouldoverturnthelawbarringpregnantgirlsfromgoingtoschool.We hopeothersfollowsuit. 38 Second,beyondspendinglesstimewithmen,wehavethroughoutfoundrobustevidencethat ELAclubsalsoreducethetimethatyounggirlsandwomenspendengagedinsocialgroups (volunteering/church).Thisappliesequallytobothcohortsandirrespectiveofthepregnancy risksfacedduringtheepidemic.Thismighthintatthereducedrolethat secretsocieties play inthelivesofyoungwomenwithaccesstoELAclubs.Thiscouldbeimportantbecausean importantfunctionofsecretsocietiesistoreinforcetraditionalnormsandreproductivebehaviors amongyoungergirls.Understandingthecoevolutionofinformalsecretsocieties,formalschooling anddevelopmentinterventionsremainsalargeopenquestionforfutureresearch. Third,itisnaturaltoaskwhetherasimilarprogramwouldbeequallyprotectiveinanother contextandtypeofaggregatecrisis.PreliminaryanalysisfromanimpactevaluationoftheELA programinSouthSudan(alsoimplementedbyBRAC)suggestssimilarcrisis-o¤settinge¤ects, eventhoughthenatureoftheaggregateshock,con ict,isdi¤erent,andthecon ictoccurred afteramoresustainedperiodofclubimplementation[Buehren etal. 2018].Nonetheless,the analysisindicatesthatmanyoftheperniciouse¤ectsofcon ictonyoungwomenareo¤setby havingparticipatedinELAclubs. WeendbyreiteratingthelinktothecurrentCOVID-19pandemic.WhiletodateSierraLeone hasexperiencedfewerthan 2000 casesand 100 deaths,thepotentialforthecountrytofaceanother signi canthealthandeconomicshockisreal.Thecohortofgirlsstudiedhereareexperiencing theirsecondsuchhealthcrisis:inourlongtermfollowup
,respondentswereinterviewedupuntil January2020,whenthecountrywasontheroadtorecoveryfromEbola,butontheeveofwhen itwouldbestruckbyCOVID-19.Understandingthedynamicsacrossthesecrisisfortheseyoung womenremainsatthetopofourresearchagenda. AAppendix A.1Attrition AlthoughgeographicmobilityishighinSierraLeone,wedonothavesevereattritionevengiven theepidemic: 83 %ofourrespondentsweretrackedfrombaselinetoendline( 4790 ). 81 %( 3865 ) oftrackedrespondentsresidedinthesamevillage.Forthoseoriginallyinatreatedvillageand 38 ThegovernmentofSierraLeonehadoriginallybeenchallengedoverthepolicyinalegalcasebroughttothe EconomicCommunityofWestAfricanStates CommunityCourtofJustice:inDecember2019theyruledthatthe banshouldberevoked.ThecasechallengingthebanwasbroughtbySierraLeoneanNGO(WAVES)inpartnership withEqualityNowandtheInstituteforHumanRightsandDevelopmentinAfrica. 35 thentrackedtoanother,wecanusetheapproximatedateoftheirmovetounderstandtheextent towhichtheywereexposedtoELAclubs.Atleast 60 %oftrackedmovershavebeenpartially exposedtoELAclubsintheiroriginalvillage.Finally, 922 ( 16 %)ofgirlsattritedoutofthesample (only 12 wereduetodeath). 39 TableA3presentscorrelatesofattrition.Columns1and2showthattreatmentassignment,and exposuretohighpregnancyriskdonotseparatelypredictattrition.Column3showsthiscontinues toholdwhencontrollingforbothtogether,andtheirinteraction.Thisalsocontinuestoholdwhen: (i)weadditionallycontrolforcharacteristicsofgirls,householdsandvillages(Column4);(ii) allowtheretobedi¤erentialattritionbetweentreatment,pregnancyrisk,andtheirinteraction withbaselinecharacteristicssuchasenrollment,employment,ageandhouseholdpoverty.The nalColumnofTableA3presentscorrelatesofattritionbetweenendlineandthelongtermfollow up:weshowthattreatmentassignmentandpregnancyriskdoesnotpredictattrition,andnor doestheirinteraction. 40 A.2Robustness Weexaminetherobustnessofourcoreresultsforeachagecohort:TableA10showstheoutcomes thatarecentraltounderstandtheimpactoftheEbolashockontheeconomiclivesofyoung women,andthemechanismsthroughwhichELAclubsmitigatetheseimpacts.Theoutcomes shownareweeklyhoursspentonlearningactivitiesfortheyoungercohort(PanelA),t
imespent withmenforbothcohorts(PanelB),outofwedlockpregnanciesfortheyoungeragecohort(Panel C),andthesupplyoftransactionalsexfortheolderagecohort(PanelD).Foreachparameter ofinterest,weshowthebaselineestimateaspreviouslyreported.Wethenconsiderthefollowing robustnesschecks. RandomizationInference,MultipleHypothesisTesting,Controls First,weuserandom- izationinferencetotestthenullofnotreatmente¤ects,followingthemethodssetoutinYoung [2019]andHe etal. [2020].Theresultingp-valuesareshowninbracesinTableA11:weseethat allofthe 11 signi cantcoe¢cientsinthebaselinespeci cationsremainstatisticallysigni cantat conventionallevelsonceweaccountforrandomizationinference.Thisisreassuringandsuggests our ndingsarenotdrivenbyoutliers. 39 Attriterswerereplacedbyarefreshersampleof 1415 girlssurveyedonlyatendline.Around 44 %ofthissample residedinthesamevillageatbaselineandendline. 40 Ofcourseindividualcharacteristicsofgirlsjointlypredictattrition(suchasbaselineeconomicactivitiesengaged in,maritalstatusandhouseholdsize),butnotdi¤erentiallysointreatedvillages,inhighriskvillages,andvillages withboth(theF-statisticsatthefootofeachColumnreportthistobethecase). 36 Second,giventhelargenumberofoutcomesconsidered,someadjustmentformultiplehypoth- esistestingcanbeconsidered.Tobeclear,theeconomicactivitiesoutcomesinPanelAsum toonebyconstruction(sowearenotmeasuringimpactsonalatentoutcomewithalternative proxies),andthetotale¤ectstimeallocationresultsinPanelBarealreadyestimatedusingaSUR modelaccountingforcorrelationacrossoutcomes.Thelargerconcernisthatwehavemultiple ITTestimatesineachspeci cation:atthefootofeachColumninTableA11wethusalsoshow thep-valueontheF-testofthejointsigni canceoftheparametersofinterest.Infouroutof ve casestheseremainhighlysigni cant. Toaddressmultiplehypothesistestinginthemostconservativeway,wealsoreportp-values computedusingthestep-downprocedureofRomanoandWolf[2016](basedon 1 000 bootstrap iterations).Fivecoe¢cientsremainsigni cant:moreover,fortheITTestimateofELAinhigh pregnancyriskvillages( b 1 + b 3 ),threeofthe vecoe¢cientsremainsigni cant.Thisistobe exp
ectedgiventhefactorialdesignestimatedgoeswellbeyondtheoriginalevaluationdesign. Wenextaddresstheconcernthatthenumberoftreatmentandcontrolvillagesisrelatively small,andthebalancingtestsinTables1and2mightnotbeespeciallypowerful.Indeed,our baselinespeci cationcontrolledforthevillagecharacteristicsshowninPanelAofTable1to addressanypotentialimbalance.TableA10thenshowshowourresultsvaryifweonlycontrolfor district xede¤ects(therandomizationstrata):ofthe 11 signi cantcoe¢cientsinthebaseline speci cations, 9 remainstatisticallysigni cant. AlternativeMeasuresofPregnancyRisk Ourfourthsetofchecksexaminealternativemea- suresofpregnancyriskduringtheepidemic:(i)exploitingthecontinuousmeasureofpregnancy risktoshowtheintensityoftreatmente¤ectsthemodelhighlights( );(ii)exploitingonly within-districtvariationinpregnancyrisk. Ourbaselineresultsexploitedthedummyindexmeasureofpregnancyrisk,de nedhighrisk villagestobeatthe 75 thorhigherpercentilesofouroverallpregnancyindex(Table3).The modelhighlightsthattimeallocationsareimpactedcontinuouslythroughpregnancyrisk.To investigatethisdirectly,weexaminehowtheresultsvaryusingdi¤erentthresholdsoftherisk index.FigureA6showstheresults,byagecohortandfortwosummaryoutcomes:(i)hrs/wk spentinproductiveactivitiesoflearning,incomegenerationorELAclubs(PanelA);(ii)hrs/week spentwithmen(PanelB).Theomittedcategoryarevillagesinthe rstsixdecilesofthedisruption index.Theimpactsontimeinlearningactivitiesandtimespentwithmenbothvarydepending ontheintensityofrisk,whereweshowcuto¤satthe60th,70th,80th,90thpercentileoftheindex. Reassuringly,forbothoutcomes,weobserveincreasingimpactsofpregnancyriskacrossdeciles, 37 andincreasinglyo¤settingimpactsoftheavailabilityofELAclubs,asthemeasuredintensityof pregnancyriskvaries. Tomotivateasecondapproachtoconstructingtheindexofpregnancyrisk,wenotethatthe mainmeasureusedisde nedinabsoluteterms,andsomeasuresthehighestriskvillagesinany ofthefourdistrictsinourevaluationsample.AsFigure5shows,thisleadstomosthighrisk villagesbeinginPortLoko.Forourresearchquestions,itistheabsolutelevelofpregnancyrisk thatmatters.However,thisraisesthec
oncernofwhethertheimpactsofhighversuslowrisk insteadpickupsomethingspeci ctoPortLokorelativetootherdistricts.Toprobethisfurther werede nethehighriskdummytobedistrictspeci c,sothatvillagesineachdistrictthatlie inthehighestquartileofthecontinuouspregnancyriskindexareclassi edashighrisk.Thisis the nalrowofresultsshownforoutcomesinTableA11:ofthe 11 signi cantcoe¢cientsinthe baselinespeci cations, 9 remainstatisticallysigni cantatconventionallevelsonceweconsider highrisktobewithin-districtratherthaninabsolutetermsacrosstheevaluationsample(with somepointestimatesbeinglargerinabsolutevaluethaninthebaselinespeci cation). References [1] adda.j (2016) EconomicActivityandtheSpreadofViralDiseases:EvidencefromHigh FrequencyData, QuarterlyJournalofEconomics 131:891-941. [2] agüero.j.mandt.beleche (2017) HealthShocksandTheirLong-lastingImpacton HealthBehaviors:Evidencefromthe2009H1N1PandemicinMexico, JournalofHealth Economics 54:40-55. [3] alfonsi.l. , o.bandiera , v.bassi , r.burgess , i.rasul , m.sulaimananda.vitali (2020) TacklingYouthUnemployment:EvidencefromaLaborMarketExperimentinUganda, Econometrica ,forthcoming. [4] allen.h (2004) Prostitution, inR.S.McElvaine(ed.) EncyclopediaoftheGreatDepression , Vol.2.,MacmillanReferenceUSA,NewYork. [5] amnestyinternational (2015)ShamedandBlamed:PregnantGirls RightsatRiskin SierraLeone,AmnestyInternationalPublications,London. [6] anderson.m.l. (2008) MultipleInferenceandGenderDi¤erencesintheE¤ectsofEarly Intervention:AReevaluationoftheAbecedarian,PerryPreschool,andEarlyTraining Projects, JournaloftheAmericanStatisticalAssociation 103:1481-95. 38 [7] archibong.bandf.annan (2020)SchoolinginSicknessandinHealth:TheE¤ectsof EpidemicDiseaseonGenderInequality,mimeo,BarnardCollege. [8] arunachalam.randm.shah (2018) ProstitutesandBrides?, AmericanEconomicRe- viewPapers&Proceedings 98:516-22. [9] ashraf.n,e.fieldandj.lee (2014) HouseholdBargainingandExcessFertility:AnEx- perimentalStudyinZambia, AmericanEconomicReview 104:2210-37.
[10] ashraf.n,n.bau,c.lowandk.mcginn (2020) NegotiatingaBetterFuture:HowIn- terpersonalSkillsFacilitateInter-GenerationalInvestment, QuarterlyJournalofEconomics 135:1095-151. [11] atlani.l,m.carael,j.brunet,t.frasca,andn.chaika (2000) SocialChangeand HIVintheFormerUSSR:TheMakingofaNewEpidemic, SocialScience&Medicine 50: 1547-56. [12] baird.s.j,c.t.mcintoshandb.ozler (2011) CashorCondition:EvidencefromaCash TransferExperiment, QuarterlyJournalofEconomics 126:1709-53. [13] baird.s,e.chirwa,j.dehoop and b.ozler (2014)GirlPower:CashTransfersandAdoles- centWelfare,EvidencefromaCluster-RandomizedExperimentinMalawi,NBERChapters, in AfricanSuccesses :HumanCapital,NBER. [14] bandiera.o , n.buehren , r.burgess , m.goldstein , s.gulesciandm.sulaiman (2020) Women sEmpowermentinAction:EvidencefromaRandomizedControlTrialinAfrica, AmericanEconomicJournal:AppliedEconomics 12:210-59. [15] bennett.d,c.f.chianganda.malani (2015) LearningDuringaCrisis:TheSARSEpi- demicinTaiwan, JournalofDevelopmentEconomics 112:1-18. [16] birungi.hetal. (2015)EducationSectorResponsetoEarlyandUnintendedPregnancy: AReviewofCountryExperiencesinSub-SaharanAfrica,STEPUPandUNESCOResearch Report. [17] bledsoe.c (1990) SchoolFeesandMarriageProcessforMendeGirlsinSierraLeone, in BeyondtheSecondSex ,Sanday.PandR.Goodenough(eds.),UniversityPennsylvaniaPress. [18] bowles.j,j.hjort,t.melvinande.werker (2016) Ebola,JobsandEconomicActivity inLiberia, JournalofEpidemiologyandCommunityHealth 70:271-7. 39 [19] buchmann.n,e.field,r.glennerster,s.nazneen,s.pimkinaandi.sen (2018)Power vsMoney:AlternativeApproachestoReducingChildMarriageinBangladesh,aRandomized ControlTrial,mimeo,DukeUniversity. [20] buehren.n,s.chakravarty,m.goldstein,v.slavchecskaandm.sulaiman (2018) Girls EmpowermentandCon ict:EvidencefromSouthSudan,mimeoWorldBank. [21] bullough.vandb.bullough (1987) WomenandProstitution:ASocialHistory , PrometheusBooks,Bu¤alo. [22] callen.m,m.isaqzadeh,j.d.longandc.sprenger (2014) ViolenceandRiskPrefer- ence:ExperimentalEvidencefromAfghanistan,
AmericanEconomicReview 104:123-48. [23] cameron.landm.shah (2015) Risk-TakingBehaviorintheWakeofNaturalDisasters, JournalofHumanResources 50:484-515. [24] casey.k,r.glennersterandt.suri (2017)TheEconomicImpactsofEbolaonFirmsin SierraLeone,IGCReportF-39204-SLE-1. [25] christensen.d,o.dube,j.haushofer,b.siddiqiandm.voors (2020)BuildingResilient HealthSystems:ExperimentalEvidencefromSierraLeoneandthe2014EbolaOutbreak, mimeo,Chicago. [26] corno.l,n.hildebrandtanda.voena (2019) AgeofMarriage,WeatherShocks,and theDirectionofMarriagePayments, Econometrica ,forthcoming. [27] crowne.dandd.marlowe (1960) ANewScaleofSocialDesirabilityIndependentof Psychopathology, JournalofConsultingPsychology 24:349-54. [28] dhar.d,t.jainands.jayachandran (2020)ReshapingAdolescents GenderAttitudes: EvidencefromaSchool-basedExperimentinIndia,mimeo,Northwestern. [29] doepke.m,m.tertiltanda.voena (2012) TheEconomicsandPoliticsofWomen s Rights, AnnualReviewofEconomics 4:339-72 [30] duflo.e (2012) WomenEmpowermentandEconomicDevelopment, JournalofEconomic Literature 50:1051-79. [31] duflo.e , p.dupasandm.kremer (2015) Education,HIV,andEarlyFertility:Experi- mentalEvidencefromKenya, AmericanEconomicReview 105:2757-97. 40 [32] dupas.p (2011) DoTeenagersRespondtoHIVRiskInformation?EvidenceforaField ExperimentinKenya, AmericanEconomicJournal:AppliedEconomics 3:1-34. [33] dupas.p,andj.robinson (2012) The(hidden)CostsofPoliticalInstability:Evidence fromKenya s2007ElectionCrisis, JournalofDevelopmentEconomics 99:314-29. [34] edmonds.e,b.feigenbergandj.leight (2020)AdvancingtheAgencyofAdolescent Girls,NBERWP27513. [35] evans.d.k,m.p.goldsteinanda.popova (2015) Health-careWorkerMortalityandthe LegacyoftheEbolaEpidemic, LancetGlobalHealth 3:e439-40. [36] ferreira.f.h.gandn.schady (2009)SocialConsequencesoftheGlobalFinancialCrisis inLatinAmerica:SomePreliminary,andSurprisinglyOptimistic,Conjectures,WorldBank OtherOperationalStudies10957. [37] field.eanda.ambrus (2008) EarlyMarriage,AgeofMenarche,andFemaleSchooling
AttainmentinBangladesh, JournalofPoliticalEconomy 116:881-930. [38] fluckiger.m,m.ludwiganda.sinaonder (2019) EbolaandStateLegitimacy, Eco- nomicJournal 129:2064-89. [39] fogli.aandl.veldkamp (2019) Germs,SocialNetworksandGrowth, ReviewofEco- nomicStudies ,forthcoming. [40] gertler.p,m.shahands.m.bertozzi (2005) RiskyBusiness:TheMarketforUnpro- tectedCommercialSex, JournalofPoliticalEconomy 113:518-50 [41] glennerster.r,t.suriands.bhogale (2016)TheImplicationsoftheEbolaOutbreak onMarkets,Traders,andFoodSecurityinSierraLeone,IGCPolicyBrief. [42] gollier.candj.w.pratt (1996) RiskVulnerabilityandtheTemperingE¤ectofBack- groundRisk, Econometrica 64:1109-23. [43] hes.s,d.jaimovichandm.scheundeln (2020) DevelopmentProjectsandEconomicNet- works:LessonsFromRuralGambia, ReviewofEconomicStudies ,forthcoming. [44] himelein.k,m.testaverde , a.turayands.turay (2015)TheSocio-economicImpacts ofEbolainSierraLeone:ResultsfromaHighFrequencyCellPhoneSurvey(RoundThree), WashingtonDC:TheWorldBank. 41 [45] huttner.aetal. (2018) DeterminantsofAntibodyPersistenceAcrossDosesandConti- nentsafterSingle-doserVSV-ZEBOVVaccinationforEbolaVirusDisease:AnObservational CohortStudy, LancetInfectiousDiseases ,April,DOI:https://doi.org/10.1016/S1473- 3099(18)30165-8. [46] jacoby.h.gande.skoufias (1997) Risk,FinancialMarkets,andHumanCapitalina DevelopingCountry, ReviewofEconomicStudies 64:311-35. [47] jayachandran.s (2015) TheRootsofGenderInequalityinDevelopingCountries, Annual ReviewofEconomics 7:63-88. [48] jensen.r. (2012) DoLaborMarketOpportunitiesA¤ectYoungWomen sWorkandFamily Decisions?ExperimentalEvidencefromIndia, QuarterlyJournalofEconomics 127:753-92. [49] lauthartejunior.iandi.rasul (2020)TheAnatomyofaPublicHealthCrisis:Household ResponsesOvertheCourseoftheZikaEpidemicinBrazil,mimeo,UCL. [50] maccormack.c.p (1979) Sande:ThePublicFaceofaSecretSociety, in TheNewReligions ofAfrica ,B.Jules-Rosette(ed.),Praeger. [51] maffioli.e (2020)ThePoliticalEconomyofHealthEpidemics:EvidencefromtheEbola Outbre
ak,mimeoMichigan. [52] mcormack-hale.f (2018) SecretSocietiesandWomen sAccesstoJusticeinSierraLeone: BridgingtheFormalandInformalDivide, Stability:InternationalJournalofSecurity& Development 7:1-17. [53] rangel.m.a,j.noblesanda.hamoudi (2020) Brazil sMissingInfants:ZikaRisk ChangesReproductiveBehavior, forthcoming, Demography . [54] rao.v,i.gupta,m.lokshinands.jana (2003) SexWorkersandtheCostofSafeSex: TheCompensatingDi¤erentialforCondomUseAmongCalcuttaProstitutes, Journalof DevelopmentEconomics 71:585-603. [55] rasul.i (2020) TheEconomicsofViralOutbreaks, AmericanEconomicAssociationP&P 110:265-68. [56] robinson.jande.yeh (2011) TransactionalSexasaResponsetoRiskinWesternKenya, AmericanEconomicJournal:AppliedEconomics 3:35-64. 42 [57] romano.j.pandm.wolf (2016) E¢cientComputationofAdjustedp-valuesfor Resampling-basedStepdownMultipleTesting, Statistics&ProbabilityLetters 113C:38-40. [58] shah.m (2013) DoSexWorkersRespondtoDisease?EvidencefromtheMaleMarketfor Sex, AmericanEconomicReview:Papers&Proceedings 103:445-50. [59] thomas.a , t.nkuzimana,a.perezhoyos,andf.kayiyakire (2014)ImpactoftheWest AfricanEbolaVirusDiseaseOutbreakonFoodSecurity,EuropeanCommission,JointRe- searchCentre. [60] thomas.m.r,g.smith , f.h.g.ferreira,d.evans,m.maliszewska , m.aliszewska , m.cruz , k.himeleinandm.over (2015)TheEconomicImpactofEbolaonsub-Saharan Africa:UpdatedEstimatesfor2015,WashingtonDC:WorldBank. [61] undp (2015)Socio-EconomicImpactofEbolaVirusDiseaseinWestAfricanCountries:A CallforNationalandRegionalContainment,RecoveryandPrevention,UNDG,Westernand CentralAfrica. [62] unicef (2014)SierraLeoneHealthFacilitiesSurvey2014:AssessingtheImpactoftheEVD OutbreakonSierraLeone sHealthSystem,GovernmentofSierraLeoneMinistryofHealth andSanitation,Freetown. [63] worldbank (2011) WorldDevelopmentReport2012:GenderEqualityandDevelopment , WashingtonDC:TheWorldBank. [64] worldbank (2014) TheEconomicImpactofthe2014EbolaEpidemic:ShortandMedium TermEstimatesforWestAfrica(English) ,Washington,DC:WorldBankGroup. [65] young.a (2019)
ChannellingFisher:RandomizationTestsandtheStatisticalInsigni cance ofSeeminglySigni cantExperimentalResults, QuarterlyJournalofEconomics 134:557-98. 43 Table1:VillageCharacteristics Means,robuststandarderrorsfromOLSregressionsinparentheses P-valueoft-testofequalityofmeanswithrelevantcontrolgroupinbraces ControlTreatmentDifference Normalized Difference (1)(2)(3)(4) Numberofdwellings 167108{.362}-.121 [462][148] Numberofsampledadolescentgirls 28.628.4{.943}-.013 [13.4][9.30] Povertyscore(Meanacrosshouseholds) .353.346{.394}-.094 [.055][.054] DistancefromFreetown(miles) 52.852.6{.291}-.005 [25.7][24.2] DistancefromKailahun(miles) 78.478.6{.248}.005 [19.6][18.5] DistancefromnearestPHU(miles) 1.891.86{.940}-.012 [1.71][1.77] Distancefromnearestsecondaryschool(miles) 3.534.23{.258}.115 [3.34][5.10] B.VillageLeaderSurvey "Girlswhoarevisiblypregnanthaveabad influenceontheirnon-pregnantpeers" [=1ifstronglyagree] .960.967{.873}.025 "Girlsshouldbeallowedtocontinuetheir educationwhilepregnant" [=1ifstronglyagree] .120.073{.365}-.111 Pregnantgirlsallowedtositexamsinthenearest secondaryschool .239.319{.270}.132 C.PolicyResponses Villagewasquarantined .060.040{.595}-.065 Villagevisitedbycontacttracingteam .960.933{.455}-.084 ReceivedRelieffromNGO .780.873{.139}.174 Villagereceivedfoodaid .260.213{.504}-.077 Villagereceivedschoolsupplies(excl.BRAC) .220.207{.818}-.023 Notes: DatasourcesaretheVillageCensus(PanelA),andtheVillageLeaderSurveys(distancemeasuresinPanelA,alloutcomesin PanelsBandC).Column3reportsp-valuesfromatestofequalityofmeanscarriedoutbyOLSregressionofeachcharacteristicona dummyforassignmenttotreatment.Allregressionsincludestrata(district)dummiesandcalculaterobuststandarderrors.Normalized differencesinColumn4arecomputedfollowingImbensandWooldridge[2009].ThePovertyScore(PPI)iscalculatedthrough scorecards:highervaluesindicatealowerprobabilitythatthehouseholdispoor.DistancefromFreetownandKailahunarecomputed fromGPSdata. A.BaselineBalanceonVillageCharacteristics Table2:BaselineBalanceforIndividualCharacteristics Means,clusteredstandarderrorsfromOLSregressionsinparentheses P-valueoft-testofequalityofmeanswith
relevantcontrolgroupinbraces ControlTreatmentDifference Normalized Difference (1)(2)(3)(4) A.BasicCharacteristics NumberofAdolescentGirls 1,1983,592 Age 17.717.5{.412}-.025 [3.76][3.74] Inanyrelationship .596.596{.893}-.000 Married .283.283{.825}.001 Ageatmarriage 16.116.4{.378}.060 [2.82][2.87] Ageofhusbandatmarriage 31.031.8{.095}.078 [6.88][7.42] HasChildren .492.486{.820}-.007 Ifinrelationship:experiencedany formofintimatepartnerviolence .464.486{.491}.032 Skills:Literacy[0-100] 22.622.0{.689}-.016 [27.5][27.2] Sexuallyactive .747.712{.103}-.056 Ifactive:Ageatsexualdebut 14.714.6{.348}-.026 [2.26][2.09] 5.085.14{.705}.009 [5.33][5.46] Ifactive:Usescontraceptive .440.422{.390}-.026 (any,excludingcondoms) [.497][.494] Ifactive:Everusedcondoms .104.095{.375}-.021 Unwantedsexoverpastyear .106.101{.739}-.012 Transactionalsexoverpastyear .041.035{.346}.022 Notes: Column3reportsp-valuesfromatestofequalityofmeanscarriedoutbyOLSregressionofeachcharacteristiconadummyforassignment totreatment.Regressionsincludestrata(district)dummiesandstandarderrorsareclusteredatthevillagelevel.Column4reportsnormalized differencesarecomputedfollowingImbensandWooldridge[2009].Intimatepartnerviolenceisdefinedasthethreatoruseofphysicalviolencefrom therespondent'spartner.Timeallocationdatawascollectedbothatbaselineandendline.Respondentswereprovidedasetof25beadsandaboard withsixcirclesrepresenting:"Education","IGA","Leisure","HouseholdChores","Sleep"and"Other".TheEducationcategoryincludesschooling, vocationaltrainingandstudytime."IGA"includespaidandunpaidworkofanykind.Respondentswerethenaskedtoallocatebeadsintoeachcircle inawaythatrepresentstimeallocationinanaverageweek.Dataonleisuretimeallocationwascollectedinasimilarway.Therecordedcategories forleisureare:"Friends","Men","Alone","Church","Volunteer"and"Other".Theexactphrasingforthe"Men"categoryis"Withboysormenyouhave asexualrelationshipwith".Respondentswerethenaskedtoallocatebeadsintoeachcircleinawaythatrepresentstimeallocationinanaverage week.Thedatapointswerelaterconvertedintoweeklyhoursusingrecordedtotalleisuretimefromthefirstexercise.UnwantedSexisdefinedas, "Beeninvolvedinanysexualintercourset
hatyouwerenotwillingtodo".TransactionalSexisdefinedas,"Receivinganythingsuchasmoney,gifts, helpwithschoolworkorsomethingelseinexchangeforsexualintercourse".Literacyisassessedbasedonrespondent'sabilitytoreadsimplethings likelabelsoncontainers(basic),andreadingcomprehensionandwritingofcompletesentence(advanced).Thescorerepresentingrespondent's proficiencywiththesetasksisthenrescaledtorangebetween0and100. Leisureactivity:Engagedinsexual activitieswithmen(weeklyhours) B.TimeUseandSexualActivity Table3:PregnancyRiskIndex Standarddeviationsinbrackets,standarderrorsinparentheses A.HealthPolicyComponents (1)(2)(3)(4)(5)(6) PHUEverClosed .140.001-.000.046.076 [.348](.049)(.050)(.075)(.075) PHUEverDisrupted .320.000-.006-.077-.069 [.468](.065)(.065)(.098)(.096) PHUFunctioningScore 89.6-1.19-1.02-1.00-.342 [19.8](2.58)(2.55)(3.90)(3.98) B.SchoolPolicyComponent NearestSecondarySchoolRe-OpenedonTime .825.085.106.072.080 [.381](.061)(.059)(.079)(.081) C.Index PregnancyIndex 0-.096-.132.094.111 [1](.132)(.129)(.235)(.248) PregnancyRiskDummy .170.429-.039-.042-.013.009 [=1ifindexgreaterthan75thpercentile](.050)(.051)(.103)(.108) N(villages) 200200200200200200 Notes: AlldatacomesfromtheVillageLeadersSurvey,collectedinOctober2015.ThebetweengroupvariationreportedinColumn2iscomputedthroughaone-wayANOVAanalysisofthedependentvariable acrossdistricts.Foreachmeasureofdisruptions,Columns3throughto6reportestimatedcoefficientsonassignmenttotreatmentfromregressionsthatuseincrementallylargersetsofvillage-levelcovariates.All regressionsincludedistrictfixedeffectsandcalculaterobuststandarderrors.Column3reportsthecoefficientonELAfromaregressionoftheEbolameasureofinterestontreatmentassignment.Thecoefficientin Column4isobtainedcontrollingfor:numberofdwellings,whetherthevillageisapoliticalstronghold(i.e.itistheresidenceofachief),numberofNGOsactivepre-Ebola,averagePPIscore,shareofChristians, distancefromFreetownanddistancefromKailahun(wherethefirstEbolacasewasrecorded).InColumn5distancefromthenearestfacilityofinterest(PHUinPanelA,SecondarySchoolinPanelB,bothinPanel C)andinteractionbetweenELAassignmentandthesedistancesareaddedasregressors.Noneofthecoefficien
tsontheinteractionbetweenELAanddistancefromeachfacilityisstatisticallysignificant(not shown).Column6includesallregressorsandinteractionsemployedinColumn5,plusfeaturesselectedbyapenalizedregression(ElasticNet)ofthePolicyindex(dummy)onallvillagecharacteristicsandtheir interactions.ThePrimaryHealthUnitfunctioningscoreisassessedonamonthlybasisbetweenJuly2014andSeptember2015,andlateraggregatedintoanindexrangingbetween0and100.Secondaryschools wereconsideredashavingre-openedontimeiftheywereopeninApril2015.ThePregnancyRiskIndexisconstructedfollowingAnderson[2008].ItaggregatesvariablesinPanelAandB,andassignshighervalues tohigherriskvillages. .302 .451 Betweendistrict variation (shareoftotal) .400 .371 .445 ConditionalcoeffonTreatment withdistanceinteractionand modelselection Conditionalcoeffon Treatmentwith distanceinteraction Mean Unconditional coeffon Treatment Conditional coeffon Treatment Table4:OverallTimeUse,andTimeSpentSocializing Sample:Girlsaged12-17atbaseline(2014) Outcomesmeasuredpost-epidemic(2016) SURestimates,standarderrorsinparentheses ELAClubLearningWorking Household Chores SocializingMenAloneFriends Volunteer/ Church (1)(2)(3)(4)(5)(6)(7)(8)(9) PregnancyRisk --12.2***6.07**5.62***1.761.27***.641.251-.058 (2.37)(3.10)(1.89)(1.63)(.444)(.458)(.430)(.660) ELATreatment|HighPregnancyRisk 3.02***9.84***-2.35-4.68**-4.70***-1.86***-1.55***-1.42***-2.48*** (.265)(2.62)(3.26)(1.91)(1.51)(.452)(.443)(.403)(.585) ELATreatment|LowPregnancyRisk 3.19***-3.03**1.72.480-2.77***-.602***-1.11***-1.11***-2.43*** (.157)(1.30)(1.20)(.790)(.810)(.211)(.233)(.240)(.334) DifferenceTreatmentEffects[ 3 ,p-value] {.528}{.000}{.239}{.012}{.262}{.012}{.385}{.507}{.942} ControlMeanatBL|LowPregnancyRisk -48.815.142.326.42.526.356.5710.8 Observations 2,3812,3812,3812,3812,3812,3792,3792,3792,379 Socializing(hours/week) Outcomes: Timeallocationdatawascollectedbyaskingrespondentstoallocateasetof25beadsonaboardwith6circlesrepresenting:"LearningActivities","IGA","socializing", "HouseholdChores","Sleep"and"Other".TheLearningcategoryincludesschooling,vocationaltrainingandstudytime."IGA"includespaidandunpaidworkofanykind.Respondentwere thenaskedtoalloc
atebeadsintoeachcircleinawaythatrepresentstimeallocationinanaverageday,anddatapointswerelaterconvertedintoweeklyhours.Asimilarprocedurewas implementedtorecordallocationofsocializingtimeacrossthefollowingactivities:"Friends","Men","Alone","Church","Volunteer"and"Other".Theexactphrasingforthe"Men"category is,"Withboysormenyouhaveasexualrelationshipwith.".Usingthenumberofhoursspentontotalsocializingtimefromthepreviousstep,theseallocationswerelaterconvertedinto weeklyhours. Notes: ***,**and*denotesignificanceatthe1%,5%,and10%levels.Controlvariablesinclude:age,PPIscore,householdsize,illiteracy,villagesize(nrofdwellings),villageaveragePPI score,distancesformkeyfacilities(clinic,secondaryschoolandmarket),adummyequaltooneifthevillageisapoliticalstronghold(i.e.hasaresidentparamountand/orsectionchief), thenumberofNGOsactivewithinthevillagebeforetheEbolaoutbreak(excludingBRAC),shareofChristianhouseholds,distancefromFreetownanddistancefromKailahun.All specificationsincludedummiesfortherandomizationstrata(district)anderrorsareclusteredattheunitofrandomization(village). Table5:SexualActivity,PregnancyandSchoolEnrolment Sample:Girlsaged12-17atbaseline(2014) Outcomesmeasuredpost-epidemic(2016) OLSestimates,standarderrorsinparentheses,p-valuesinbraces Frequencyof UnprotectedSex Pregnant SinceBL PregnantsinceBL, Out-of-Wedlock InSchool FullTime InSchooland Working Only Working (1)(2)(3)(4)(5)(6) PregnancyRisk 1.93***.105***.070*-.167*** -.060.188*** (.624)(.036)(.039)(.045) (.042)(.061) ELATreatment|HighPregnancyRisk -1.48**-.040-.072*.087*.097**-.129** (.677)(.039)(.038)(.050)(.037)(.063) ELATreatment|LowPregnancyRisk .157.006-.010-.051*.009.024 (.328)(.020)(.016)(.030)(.027)(.019) DifferenceTreatmentEffects[ 3 ,p-value] {.027}{.289}{.135}{.020}{.052}{.020} ControlMeanatBL|LowPregnancyRisk 3.28.115.090.520 .296.080 Observations 1,4122,3842,3842,3842,3842,384 SexualActivityandPregnancySchoolingandWork Outcomes: Frequencyofsexismeasuredovera30dayperiod.Frequencyofunprotectedsexisdefinedasintercoursefrequencyforrespondentsthatdonotuseanyform ofcontraceptive,andequaltozeroforthosethatdo(toprovideconservativeestimates,respondentsthateverusecon
domsareassumedtoengageinprotectedsex). PregnancyoutcomesrefertoconceptionsthattookplaceafterAugust2014,oncebaselinedatacollectionwascompleted.IncomeGeneratingActivities(IGA)includeboth self-employmentandwagelabor. Notes: ***,**and*denotesignificanceatthe1%,5%,and10%levels.ThemainresultsareestimatedwithanANCOVAspecification.Controlvariablesinclude:age,PPI score,householdsize,illiteracy,villagesize(nrofdwellings),villageaveragePPIscore,distancesformkeyfacilities(clinic,secondaryschoolandmarket),adummyequalto oneifthevillageisapoliticalstronghold(i.e.hasaresidentparamountand/orsectionchief),thenumberofNGOsactivewithinthevillagebeforetheEbolaoutbreak (excludingBRAC),shareofChristianhouseholds,distancefromFreetownanddistancefromKailahun.Columns2through6controlforalsoforbaselinevaluesofthe relevantoutcomevariable.Allspecificationsincludedummiesfortherandomizationstrata(district)anderrorsareclusteredattheunitofrandomization(village). Table6:SexualActivityandPregnancy,OlderGirlsandWomen Sample:Girlsaged18-25atbaseline(2014) Outcomesmeasuredpost-epidemic(2016) OLSestimates,standarderrorsinparentheses,p-valuesinbraces Unwanted Sex TransactionalSex FemaleControlled ContraceptiveUse Pregnancy, sinceBL (1)(2)(3)(4) PregnancyRisk -.038-.002-.027.017 (.024)(.018)(.062)(.049) ELATreatment|HighPregnancyRisk .054**.054***.137**-.047 (.023)(.020)(.063)(.048) ELATreatment|LowPregnancyRisk .004.017.034.008 (.013)(.011)(.030)(.028) DifferenceTreatmentEffects[ 3 ,p-value] {.063}{.125}{.150}{.333} ControlMeanatBL|LowPregnancyRisk .154.047.468.844 Observations 2,2432,2432,3142,400 Notes: ***,**and*denotesignificanceatthe1%,5%,and10%levels..Columns(1)doesnotcontrolforbaselineoutcomevalues,whileallotherspecifications do.Controlvariablesforallspecificationsinclude:age,PPIscore,householdsize,illiteracy,villagesize(nrofdwellings),villageaveragePPIscore,distances formkeyfacilities(clinic,secondaryschoolandmarket),adummyequaltooneifthevillageisapoliticalstronghold(i.e.hasaresidentparamountand/orsection chief),thenumberofNGOsactivewithinthevillagebeforetheEbolaoutbreak(excludingBRAC),shareofChristianhouseholds,distancefromFreetownand distancefromKailahun
.Allspecificationsincludedummiesfortherandomizationstrata(district)anderrorsareclusteredattheunitofrandomization(village). Outcomes: UnwantedSexisdefinedas,"Beeninvolvedinanysexualintercoursethatyouwerenotwillingtodo".TransactionalSexisdefinedas,"Receiving anythingsuchasmoney,gifts,helpwithschoolworkorsomethingelseinexchangeforsexualintercourse" . FemaleControlledContraceptives include: contraceptivepills,injectionsorimplants,IUDs,vasectomyandfemalesterilization.PregnancyoutcomesrefertoconceptionsthattookplaceafterAugust2014, oncebaselinedatacollectionwascompleted. Table7:LongTermFollowUp,YoungGirls Sample:Girlsaged12-17atbaseline(2014) Outcomesmeasuredpost-epidemic(2019/20) Standarderrorsinparentheses,p-valuesinbraces LearningSocializingWorkingChoresMenAloneFriends Volunteer/ Church EnrolmentOnly Pregnancy, sinceBL (1)(2)(3)(4)(5)(6)(7)(8)(9)(10) PregnancyRisk -6.80*.2484.723.28.371.547-.665.213-.111*.146** (3.58)(1.41)(2.05)(2.98)(.603)(.522)(.426)(.685)(.059)(.073) ELATreatment|HighPregnancyRisk 7.81**-4.36***-4.13-1.42-1.14*-1.56***-.373-1.34*.093-.083 (3.49)(1.49)(3.07)(2.92)(.634)(.447)(.367)(.768)(.057)(.068) ELATreatment|LowPregnancyRisk -1.07-2.563.95*-.470-.895***-.260-.361-.943*-.045.051 (2.29)(.968)(2.05)(1.58)(.337)(.317)(.347)(.492)(.032)(.033) DifferenceTreatmentEffects[ 3 ,p-value] {.033}{.321}{.031}{.779}{.733}{.018}{.982}{.661}{.034}{.074} ControlMeanatBL|LowPregnancyRisk 48.826.415.142.32.526.356.5710.8.520.115 Observations 1,2801,2801,2801,2801,2801,2801,2801,2801,2941,294 Notes: ***,**and*denotesignificanceatthe1%,5%,and10%levels.ThemainresultsareestimatedwithanANCOVAspecification.Controlvariablesinclude:age,PPIscore,householdsize,villagesize (nrofdwellings),villageaveragePPIscore,distancesformkeyfacilities(clinic,secondaryschoolandmarket),adummyequaltooneifthevillageisapoliticalstronghold(i.e.hasaresidentparamount and/orsectionchief),thenumberofNGOsactivewithinthevillagebeforetheEbolaoutbreak(excludingBRAC),shareofChristianhouseholds,distancefromFreetownanddistancefromKailahun.All specificationsincludedummiesfortherandomizationstrata(district)anderrorsareclusteredattheunitofrandomization(village).All
regressionsincludethebaselinevalueoftheoutcomevariablesasa control. SchoolingandPregnancy Outcomes: Timeallocationdatawascollectedbyaskingrespondentstoallocateasetof25onaboardwith6circlesrepresenting:"LearningActivities","IGA","socializing","HouseholdChores","Sleep"and "Other".TheLearningcategoryincludesschooling,vocationaltrainingandstudytime."IGA"includespaidandunpaidworkofanykind.Respondentwerethenaskedtoallocatebeadsintoeachcircleina waythatrepresentstimeallocationinanaverageday,anddatapointswerelaterconvertedintoweeklyhours.Asimilarprocedurewasimplementedtorecordallocationofsocializingtimeacrossthefollowing activities:"Friends","Men","Alone","Church","Volunteer"and"Other".Theexactphrasingforthe"Men"categoryis,"Withboysormenyouhaveasexualrelationshipwith.".Usingthenumberofhoursspent ontotalleisurefromthepreviousstep,theseallocationswerelaterconvertedintoweeklyhours.PregnancyoutcomesrefertoconceptionsthattookplaceafterAugust2014,oncebaselinedatacollectionwas completed. TimeUse(hours/week)Socializing(hours/week) OLSSURSUR Table8:LongTermFollowUp,YoungWomen Sample:Womenaged18-25atbaseline(2014) Outcomesmeasuredpost-epidemic(2019/20) OLSestimates,standarderrorsinparentheses,p-valuesinbraces Unwanted Sex Transactional Sex FemaleControlled ContraceptiveUse Pregnancy, sinceBL (1)(2)(3)(4) PregnancyRisk .013 -.046-.198**.112** (.048) (.034)(.077)(.048) ELATreatment|HighPregnancyRisk .028.069*.191***-.045 (.046)(.036)(.049)(.049) ELATreatment|LowPregnancyRisk .025-.014-.031.024 (.025)(.025)(.039)(.027) DifferenceTreatmentEffects[ 3 ,p-value] {.945}{.075}{.007}{.230} ControlMeanatBL|LowPregnancyRisk .154 .047.468.844 Observations 142314231,4931,553 Notes: ***,**and*denotesignificanceatthe1%,5%,and10%levels.ThemainresultsareestimatedwithanANCOVAspecification.Control variablesinclude:age,PPIscore,householdsize,villagesize(nrofdwellings),villageaveragePPIscore,distancesformkeyfacilities(clinic, secondaryschoolandmarket),adummyequaltooneifthevillageisapoliticalstronghold(i.e.hasaresidentparamountand/orsectionchief), thenumberofNGOsactivewithinthevillagebeforetheEbolaoutbreak(excludingBRAC),shareofChristianhouseholds,distancefrom Fr
eetownanddistancefromKailahun.Allspecificationsincludedummiesfortherandomizationstrata(district)anderrorsareclusteredatthe unitofrandomization(village).Allregressionsincludethebaselinevalueoftheoutcomevariablesasacontrol. Outcomes: UnwantedSexisdefinedas,"Beeninvolvedinanysexualintercoursethatyouwerenotwillingtodo".TransactionalSexisdefined as,"Receivinganythingsuchasmoney,gifts,helpwithschoolworkorsomethingelseinexchangeforsexualintercourse".Pregnancyoutcomes refertoconceptionsthattookplaceafterAugust2014,oncebaselinedatacollectionwascompleted. FemaleControlledContraceptives include: contraceptivepills,injectionsorimplants,IUDs,vasectomyandfemalesterilization Table9:Partners,LongTermFollowUp Outcomesmeasuredpost-epidemicin2019/20 OLSestimates,standarderrorsinparentheses,p-valuesinbraces Age Education Index Gender Empowerment Index Aversionto GBV Age Education Index Gender Empowerment Index Aversionto GBV (1)(2)(3)(4)(5)(6)(7)(8) ELATreatmentEffect -.942.271**-.019.064***-.147.036*.055-.044** (1.02)(.112)(.029)(.028)(.729)(.020)(.080)(.020) MeaninControl 29.7.243.324.77835.5.056-.157.334 Observations 494492496488871872871872 PartnersofGirlsAged12-17atBaselinePartnersofWomenAged18-25atBaseline Notes: ***,**and*denotesignificanceatthe1%,5%,and10%levels.Allcontrolsrefertothefemalepartner.ThemainresultsareestimatedwithanANCOVAspecification.Control variablesinclude:age,PPIscore,householdsize,villagesize(nrofdwellings),villageaveragePPIscore,distancesformkeyfacilities(clinic,secondaryschoolandmarket),a dummyequaltooneifthevillageisapoliticalstronghold(i.e.hasaresidentparamountand/orsectionchief),thenumberofNGOsactivewithinthevillagebeforetheEbolaoutbreak (excludingBRAC),shareofChristianhouseholds,distancefromFreetownanddistancefromKailahun.Allspecificationsincludedummiesfortherandomizationstrata(district)and errorsareclusteredattheunitofrandomization(village). Outcomes: Alloutcomesrefertothemalepartner. ThePartnerEducationIndex isaninverse-covarianceweightedindex(Anderson,2008)of:whetherthepartnerhaseverattended school,whetherthepartneriscurrentlyinschool,andwhetherhehascompletedhighschool.Forthe GenderEmpowermentIndex ,respond
entswereaskedwhethermen,womenor bothshouldberesponsibleforthefollowingactivities:earnmoneyforthefamily;haveahigherlevelofeducation;dowashing,cleaningandcooking;fetchwaterifthereisnowater pumportap;feedingandbathingchildren;helpthechildrenintheirstudiesathome;lookafterillpersons.Theindexistheshareofquestionstowhichtheanswerwasboth/same. Therefore,highervaluesrepresentmoreegalitariangendernorms.Partnerswerealsoaskedwhetherahusbandisjustifiedinhittinghiswifeinfivescenarios.The GBVAversion Index istheshareofnegativeanswerstothesefiveIPVscenario.Thescenariosare:"Ifshegoesourwithouttellinghim?";"Ifsheneglectsthechildren?","Ifsheargueswithhim?", "Ifsherefusestohavesexwithhim?","Ifsheburnsthefood?". Figure1:TimelineoftheEbolaEpidemicinSierraLeone Notes: DataretrievedfromWorldHealthOrganization'sSituationReports(lastupdate11May2016).Confirmedcasesrefertolabtestedpatients,while probablecasesrefertocasesdiagnosedbyclinicalstaffandbutnottested. 20202019 Figure2:StudyTimeline 2013201420152016 DATA COLLECTION Endlinesurvey (Feb-May2016) Census (Oct-Nov2013) Baselinesurvey (Feb-May2014) ELAMonitoringand CommunityLeader surveys (Jun-Oct2015) Vocationaltraining round2 (Q12016) Clubsestablished, lifeskillstrainingstarts (Aug2014) Vocationaltraining+ microcreditround1 (Q12015) FirstcasesofEbolain SierraLeone,schools close (May-Jun2014) Ebolareachesall 14Districts (Oct2014) Schools Reopen (Apr2015) Outbreak reaches peak (Dec2014) SierraLeone declaredEbola-free (Mar2016) ELA INTERVENTION EBOLA OUTBREAK LongTermFollowUp Women/Men/Communities (Jun2019-Jan2020) Figure3:AgeProfilesforSexualActivity,SchoolEnrolmentandOther EconomicActivities,Pre-Crisis A.SexualActivity Notes: PanelA:UnwantedSexisdefinedas,"Beeninvolvedinanysexualintercoursethatyouwerenotwillingtodo".TransactionalSexisdefinedas, "Receivinganythingsuchasmoney,gifts,helpwithschoolworkorsomethingelseinexchangeforsexualintercourse".Bothpanelsusetheentire baselinesampleofyoungwomen.PanelB:SchoolreferstoformalschoolingandIGAreferstobothwageemploymentandself-employment. B.SchoolEnrolmentandOtherEconomicActivities Sexually Active,AnyRelationship Unwantedsex,transactionalsex Particip
ationRate Notes: PanelAandBreportaveragesforthesampletrackedbetweenBLandEL . Amongthereasonsfordroppingout,the Preference categorycollectsallthoseanswers,categoricalorqualitative,indicatingthattherespondent chosetoleaveschool.Commonanswersinthiscategoryare"didnotfinditinteresting"or"toodifficult".Thecategory Health/HHShock includesallinstancesofnon-financialshocksthataffectedrespondents,suchassicknessor familycircumstancesthatforcedtherespondentoutofschooling.Thecategory HHPreference includesallthoseanswerspointingtothedecisionofleavingschoolhavingbeentakenbytherespondents'parentsorguardians. PanelCandDdepictKaplan-Meyersurvivalfunctions.PanelCfocusesonwomenaged12-17atthebeginningofthestatedperiods(May'12orMay'14)whodidnotexperienceanypregnancybefore.Respondents'pregnancy historiesareusedtogenerateapseudo-panelwithmonthlyobservations,whereeachindividual's failure variableswitchesto1fromthemonthinwhichtheyoungwomenbecomespregnantforthefirsttime.PanelDrepeatsthis analysisforyoungwomenwhoexperiencedatleastonepregnancybeforethestatedperiods(May'12orMay'14). Figure4:SchoolingandPregnancyOvertheCourseoftheEpidemic A.SchoolEnrolmentRatesbyAge PreandPost-epidemic,ControlVillages B.ReasonsforSchoolDropoutbyDropoutDate GirlsAged12-17atBaseline,ControlVillages TimetoConception, 2ndPregnancy ControlGroup C.TimetoConception,FirstPregnancyD.TimetoConception,SecondPregnancy ControlVillagesControlVillages -0.2 0 0.2 0.4 0.6 0.8 1 All12131415161718+ Pre Post Difference 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 CostPregnancyPreferenceHealth/HH Shock MarriageHH Preference Other upto2014 2014-216 schools reopen Ebolafree schools reopen Ebolafree N=45,PRindexmean(sd)=.065(.891)N=40,PRindexmean(sd)=.102(.632) N=82,PRindexmean(sd)=-.541(.426) N=33,PRindexmean(sd)=1.13(1.44) Figure5:PregnancyRisk(PR)forYoungGirlsandWomen Notes: DataforthecentralmapisretrievedfromWorldHealthOrganization'sSituationReports(lastupdated11May2016),showingconfirmedandprobablecasesper100,000.Confirmedcasesrefertolabtestedpatients,whileprobable casesrefertocasesdiagnosedbyclinicalstaffandbutnottested.Intheouterdistrictmaps,foreachdistrict,weshowthenumberofsamplevillages,andt
hemeanandstandarddeviationoftheEbola-relatedvulnerabilityindex.The villagesmostexposedtoEbola-relatedpregnancyrisk(inthetopquartileoftheindex)aredepictedinred. HighPregnancyRisk LowPregnancyRisk Control Treatment Figure6:TimeUseatEndline,GirlsAged12-17atBaseline Notes: BarsrepresentestimatedparametersfromasystemofSURusingalltimeusecategoriesexcluding "other".Controlvariablesforallspecificationsinclude:age,PPIscore,householdsize,illiteracy,villagesize(nrof dwellings),villageaveragePPIscore,distancesformkeyfacilities(clinic,secondaryschoolandmarket),adummy equaltooneifthevillageisapoliticalstronghold(i.e.hasaresidentparamountand/orsectionchief),thenumber ofNGOsactivewithinthevillagebeforetheEbolaoutbreak(excludingBRAC),shareofChristianhouseholds, distancefromFreetownanddistancefromKailahun.Allequationsincludebaselinevalueoftheoutcomeas independentvariable,dummiesfortherandomizationstrata(district)anderrorsareclusteredattheunitof randomization(village).Errorbarsrepresent90%confidenceintervals. Outcomes: Timeallocationdatawascollectedbyaskingrespondentstoallocateasetof25beadsonaboard with6circlesrepresenting:"LearningActivities","IGA","socializing","HouseholdChores","Sleep"and"Other". TheLearningcategoryincludesschooling,vocationaltrainingandstudytime."IGA"includespaidandunpaidwork ofanykind.Respondentwerethenaskedtoallocatebeadsintoeachcircleinawaythatrepresentstime allocationinanaverageday,anddatapointswerelaterconvertedintoweeklyhours.Asimilarprocedurewas implementedtorecordallocationofsocializingtimeacrossthefollowingactivities:"Friends","Men","Alone", "Church","Volunteer"and"Other".Theexactphrasingforthe"Men"categoryis,"Withboysormenyouhavea sexualrelationshipwith.".Usingthenumberofhoursspentontotalsocializingtimefromthepreviousstep,these allocationswerelaterconvertedintoweeklyhours. A:OverallTimeUse WeeklyHours,90%ConfidenceIntervals B:TimeSpentSocializing WeeklyHours,90%ConfidenceIntervals Outcomes: Timeallocationdatawascollectedbyaskingrespondentstoallocateasetof25beadsonaboard with6circlesrepresenting:"LearningActivities","IGA","socializing","HouseholdChores","Sleep"and"Other". TheLearningcategoryincludesschooling,vo
cationaltrainingandstudytime."IGA"includespaidandunpaidwork ofanykind.Respondentwerethenaskedtoallocatebeadsintoeachcircleinawaythatrepresentstime allocationinanaverageday,anddatapointswerelaterconvertedintoweeklyhours.Asimilarprocedurewas implementedtorecordallocationofsocializingtimeacrossthefollowingactivities:"Friends","Men","Alone", "Church","Volunteer"and"Other".Theexactphrasingforthe"Men"categoryis,"Withboysormenyouhavea sexualrelationshipwith.".Usingthenumberofhoursspentontotalsocializingtimefromthepreviousstep,these allocationswerelaterconvertedintoweeklyhours. Notes: BarsrepresentestimatedparametersfromasystemofSURusingalltimeusecategoriesexcluding "other".Controlvariablesforallspecificationsinclude:age,PPIscore,householdsize,illiteracy,villagesize(nrof dwellings),villageaveragePPIscore,distancesformkeyfacilities(clinic,secondaryschoolandmarket),adummy equaltooneifthevillageisapoliticalstronghold(i.e.hasaresidentparamountand/orsectionchief),thenumber ofNGOsactivewithinthevillagebeforetheEbolaoutbreak(excludingBRAC),shareofChristianhouseholds, distancefromFreetownanddistancefromKailahun.Allequationsincludebaselinevalueoftheoutcomeas independentvariable,dummiesfortherandomizationstrata(district)anderrorsareclusteredattheunitof randomization(village).Errorbarsrepresent90%confidenceintervals. Figure7:TimeUseatEndline,GirlsAged18-25atBaseline A:OverallTimeUse WeeklyHours,90%ConfidenceIntervals B:TimeSpentSocializing WeeklyHours,90%ConfidenceIntervals TableA1:ELALifeSkillsModules TableA2:ParticipationinELAClubs Means,standarddeviationsinbrackets P-valueoft-testofequalityofmeanswithcontrolgroupinbraces ControlTreatmentDifferenceControlTreatmentDifferenceControlTreatmentDifference (1)(2)(3)(4)(5)(6)(7)(8)(9) A.Membership NumberofpotentialELAmemberspervillage 136.6129.9{.738} [132.1][9.5] (ClubMembers)/(popaged12-25) .307 [.175] B.Participation HaveyoueverheardaboutELAclubs? .272.890{.000}.2410.883{.000}.301.897{.000} HaveyoueverparticipatedinanyELAclubactivities? .041.708{.000}.041.761{.000}.041.657{.000} .824.832.815 ParticipatedinLifeSkillstrainingatleastweekly .772.804.734 .512.488.540 Didyoueverreceivedtraininginfin
ancialliteracy? .247.217.281 .337.319.358 DidyoutakeamicrofinanceloanfromBRAC? .127.088.169 Observations 1,1973,5925901,7966081,796 AllAge12-17atBaselineAge18-25atBaseline Notes: DataonpotentialELAclubmembersineachvillageisobtainedfromthevillagecensusadministeredpriortotheintervention,whiledataonclubmembershipwascollectedduringtheELA MonitoringSurveyin2015.EqualityofmeansistestedbyOLSregressionofthevariableofinterestontreatmentassignmentanddistrictfixedeffects,withstandarderrorsclusteredatthevillagelevel. Dataonparticipationinfinancialliteracytraining,livelihoodskillstrainingandmicrofinanceisconditionalonbeingassignedtothetreatmentarmthatofferedthosespecificprograms. HaveyoueverparticipatedinLifeSkillstraining organizedthroughtheclub? Couldrecountatleast4majortopics(outof8)thatwere coveredintheLifeSkillscurriculum HaveyoueverparticipatedinVocationalTraining organizedthroughtheclub? TableA3:Attrition DependentVariable=1ifgirlistrackedfrombaselinetoendline OLSestimates,standarderrorsinparentheses P-valuesofjoint-significancetestinbraces EndlinetoLong TermFollowUp (1)(2)(3)(4)(5)(6) ELATreatment .007.011.001.035-.023 (.016)(.018)(.018)(.101)(.023) PregnancyRisk .024.040.029.015-.052 (.023)(.029)(.029)(.161)(.040) -.021-.013.014.033 (.032)(.032)(.190)(.052) IndividualControls NoNoNoYesYesYes F-Test {.000}{.000}{.000} VillageControls NoNoNoYesYesYes F-Test {.065}{.062}{.000} Interactions NoNoNoNoYesNo F-Test {.180} Meanofoutcomevariable .673 AdjustedR-squared .004.005.005.012.018.034 Observations 5,7345,7345,7345,7345,7343,183 .829 ELATreatmentxPregnancyRisk BaselinetoEndline Notes: ***denotessignificanceat1%,**at5%,and*at10%.DataforColumns1through5isfromthebaselinesurvey,exceptPPIscoreswhichwere compiledduringthevillagecensuspriortocollectionofthebaselinesurvey.Allregressionsincludedummiesforrandomizationstrata(district)anderrors areclusteredattheunitofrandomization(village).Individualcontrolsinclude:age,enrolmentatbaseline,employment,PPIscore,maritalstatus, householdsize.Villagecontrolsinclude:villagesize(nrofdwellings),distancefromsecondaryschool,distancefromPHU,whetherthevillageisapolitical stronghold,numberofNGOsactive,av
eragePPIscore,shareofChristians,whetherthevillagereceivedfoodorschoolrelief,distancefromFreetownand fromKailahun.Interactionsinclude:enrolment,employment,ageandPPIscore.InColumn6,thelongtermfollowupaimedattracking71%ofendline respondents.Trackingbeganfromarandomsubsampleofendlinerespondents,withresamplingtakingplaceincaserespondentscouldnotbetracked. Attritionanalysisfocusesonrespondentsthatwerepartofthefirstdraw,thusexcludingresampledwomen.TheresultsinColumn6arefromaregression analogoustoColumn4. TableA4BalancebyPregnancyRisk Means,clusteredstandarderrorsfromOLSregressionsinparentheses P-valueoft-testofequalityofmeanswithrelevantcontrolgroupinbraces LowRiskHighRiskDifference (1)(2)(3) A.VillageCharacteristics Numberofdwellings 13560{.191} [287][49.0] Povertyscore(Meanacrosshouseholds) .345.359{.035} [.053][.059] Numberofpre-existingNGOs 3.082.44{.252} [1.92][1.81] DistancefromFreetown(miles) 56.633.5{.271} [24.6][11.8] DistancefromKailahun(miles) 75.4193.8{.597} [18.5][10.9] DistancefromnearestPHU(miles) 1.812.13{.453} [1.70][1.96] Distancefromnearestsecondaryschool(miles) 4.322.77{.578} [5.00][2.72] B.VillageLeaderSurvey "Girlswhoarevisiblypregnanthaveabad influenceontheirnon-pregnantpeers" [=1ifstronglyagree] .970.941{.413} "Girlsshouldbeallowedtocontinuetheir educationwhilepregnant" [=1ifstronglyagree] .096.029{.045} Pregnantgirlsallowedtositexamsinthenearest secondaryschool 28.337.9{.938} C.PolicyResponses Villagewasquarantined .030.118{.119} Villagevisitedbycontacttracingteam .9281{.088} ReceivedRelieffromNGO .849.853{.103} Villagereceivedfoodaid .199.353{.986} Villagereceivedschoolsupplies(excl.BRAC) .355.500{.690} ReceivedRelieffromGovernment .687.706{.140} Notes: DatasourcesaretheVillageCensus(PanelA),andtheVillageLeaderSurveys(distancemeasuresinPanelA,alloutcomesin PanelsBandC).Column3reportsp-valuesfromatestofequalityofmeanscarriedoutbyOLSregressionofeachcharacteristicona dummyforassignmenttotreatment.Allregressionsincludestrata(district)dummiesandcalculaterobuststandarderrors.ThePoverty Score(PPI)iscalculatedthroughscorecardsanditsvalue,rangingfrom0to100,representsthelikelihoodofahouseholdbeingbelow thepoverty
line.Thenumberofpre-existingNGOsincludesallorganizationsapartfromBRAC.DistancefromFreetownandKailahun arecomputedfromGPSdata. TableA5:BalancebyPregnancyRisk,IndividualCharacteristics Means,clusteredstandarderrorsfromOLSregressionsinparentheses P-valueoft-testofequalityofmeanswithrelevantcontrolgroupinbraces LowRiskHighRiskDifference (1)(2)(3) A.BasicCharacteristics NumberofAdolescentGirls 1,1983,592 Age 17.517.8{.127} [3.74][3.80] Inanyrelationship .578.681{.313} Married .272.339{.505} Ageatmarriage 16.316.2{.973} [2.91][2.65] Ageofhusbandatmarriage 31.731.3{.634} [7.46][6.68] HasChildren .486.497{.505} Ifinrelationship:Experiencedanyformof conjugalviolence .4770.498{.200} B.EmpowermentandAspirations EmpowermentIndex[0-100] 17.115.2{.776} [20.6][20.7] Sexuallyactive .715.745{.648} Ifactive:Ageatsexualdebut 14.614.6{.061} [2.17][2.02] Ifactive:Usescontraceptive .434.396{.163} (any,excludingcondoms) .496.489 Ifactive:Everusedcondoms .100.084{.351} Leisureactivity:Men(weeklyhours) 5.005.73{.501} [5.43][5.36] Timeuse:LearningActivities(weeklyhours) 29.224.0{.058} [34.2][3.8] Unwantedsexoverpastyear .107.0805{.076} Transactionalsexoverpastyear .037.036{.944} Skills:Literacy[0-100] 22.321.3{.248} [27.5][26.2] Enrolledonly .295.189{.017} EngagedinIncomegeneratingactivityonly .326.371{.276} Engagedinboth .159.210{.721} C.HumanCapitalandEconomicActivities Notes: Column3reportsp-valuesfromatestofequalityofmeanscarriedoutbyOLSregressionofeachcharacteristic.Regressions includestrata(district)dummiesandstandarderrorsareclusteredatthevillagelevel.Intimatepartnerviolenceisdefinedasthe threatoruseofphysicalviolencefromtherespondent'spartner.FortheEmpowermentIndex,respondentswereaskedwhethermen, womenorbothshouldberesponsibleforthefollowingactivities:earnmoneyforthefamily;haveahigherlevelofeducation;do washing,cleaningandcooking;fetchwaterifthereisnowaterpumportap;feedingandbathingchildren;helpthechildrenintheir studiesathome;lookafterillpersons.Theindexistheshareofquestionstowhichtheanswerwasboth/same.Therefore,higher valuesrepresentmoreegalitariangendernorms.Timeallocationdatawascollectedbothatbaselineandendline.Respondentswere provided
asetof25beadsandaboardwithsixcirclesrepresenting:"Education","IGA","Leisure","HouseholdChores","Sleep"and "Other".TheEducationcategoryincludesschooling,vocationaltrainingandstudytime."IGA"includespaidandunpaidworkofany kind.Respondentswerethenaskedtoallocatebeadsintoeachcircleinawaythatrepresentstimeallocationinanaverageweek. Dataonleisuretimeallocationwascollectedinasimilarway.Therecordedcategoriesforleisureare:"Friends","Men","Alone", "Church","Volunteer"and"Other".Theexactphrasingforthe"Men"categoryis"Withboysormenyouhaveasexualrelationship with".Respondentswerethenaskedtoallocatebeadsintoeachcircleinawaythatrepresentstimeallocationinanaverageweek. Thedatapointswerelaterconvertedintoweeklyhoursusingrecordedtotalleisuretimefromthefirstexercise.UnwantedSexis definedas,"Beeninvolvedinanysexualintercoursethatyouwerenotwillingtodo".TransactionalSexisdefinedas,"Receiving anythingsuchasmoney,gifts,helpwithschoolworkorsomethingelseinexchangeforsexualintercourse".Literacyisassessed basedonrespondent'sabilitytoreadsimplethingslikelabelsoncontainers(basic),andreadingcomprehensionandwritingof completesentence(advanced).Thescorerepresentingrespondent'sproficiencywiththesetasksisthenrescaledtorangebetween 0and100. TableA6:ELAParticipants,byPregnancyRisk Means,clusteredstandarderrorsfromOLSregressionsinparentheses P-valueoft-testofequalityofmeanswithrelevantcontrolgroupinbraces LowRiskHighRiskDifference (1)(2)(3) A.BasicCharacteristics Age 17.317.2{.594} [3.73][3.79] Inanyrelationship .565.631{.982} Married .268.272{.324} Ageatmarriage 16.316.6{.415} [2.82][2.64] Ageofhusbandatmarriage 31.8231.29{.553} [7.62][5.86] Haschildren 0.4780.434{.275} Ifinrelationship:Experiencedany formofintimatepartnerviolence 0.4920.535{.128} GenderEmpowermentIndex[0-100] .175.145{.497} [.204][.209] Sexuallyactive .696.671{.364} Ifactive:Ageatsexualdebut 14.514.6{.223} [2.21][1.84] Ifactive:Usescontraceptive .425.408{.383} (any,excludingcondoms) [.495][.492] Ifactive:Everusedcondoms .093.098{.584} Leisureactivity:Men(weeklyhours) 4.895.20{.380} [5.27][5.19] 29.427.2{.804} [34.2][30.1] Unwantedsexoverpastyear .121.086{.138} Transactionalsexoverpastyear .03
9.043{.941} Skills:Literacy[0,1] .209.232{.818} [.262][.254] Enrolledonly .289.223{.363} Engagedinincomegenerationonly .334.325{.606} Engagedinboth .166.240{.152} Notes: Column 3 reports p-values from a test of equality of means carried out by OLS regression of each characteristic. Regressions include strata(district)dummiesandstandarderrorsareclusteredatthevillagelevel.Intimatepartnerviolenceisdefinedasthethreatoruseofphysical violencefromtherespondent'spartner.FortheEmpowermentIndex,respondentswereaskedwhethermen,womenorbothshouldbe responsibleforthefollowingactivities:earnmoneyforthefamily;haveahigherlevelofeducation;dowashing,cleaningandcooking;fetchwater ifthereisnowaterpumportap;feedingandbathingchildren;helpthechildrenintheirstudiesathome;lookafterillpersons.Theindexisthe shareofquestionstowhichtheanswerwasboth/same.Therefore,highervaluesrepresentmoreegalitariangendernorms.Timeallocationdata wascollectedbothatbaselineandendline.Respondentswereprovidedasetof25beadsandaboardwithsixcirclesrepresenting:"Education", "IGA","Leisure","HouseholdChores","Sleep"and"Other".TheEducationcategoryincludesschooling,vocationaltrainingandstudytime."IGA" includespaidandunpaidworkofanykind.Respondentswerethenaskedtoallocatebeadsintoeachcircleinawaythatrepresentstime allocationinanaverageweek.Dataonleisuretimeallocationwascollectedinasimilarway.Therecordedcategoriesforleisureare:"Friends", "Men","Alone","Church","Volunteer"and"Other".Theexactphrasingforthe"Men"categoryis"Withboysormenyouhaveasexualrelationship with".Respondentswerethenaskedtoallocatebeadsintoeachcircleinawaythatrepresentstimeallocationinanaverageweek.Thedata pointswerelaterconvertedintoweeklyhoursusingrecordedtotalleisuretimefromthefirstexercise.UnwantedSexisdefinedas,"Been involvedinanysexualintercoursethatyouwerenotwillingtodo".TransactionalSexisdefinedas,"Receivinganythingsuchasmoney,gifts, helpwithschoolworkorsomethingelseinexchangeforsexualintercourse".Literacyisassessedbasedonrespondent'sabilitytoreadsimple thingslikelabelsoncontainers(basic),andreadingcomprehensionandwritingofcompletesentence(advanced).Thescorerepresenting respondent'sproficiencywiththesetasksis
thenrescaledtorangebetween0and100. B.EmpowermentandAspirations C.HumanCapitalandEconomicActivities Timeuse:LearningActivities(weekly hours) TableA7:OtherFormsofVillageHeterogeneity Sample:Girlsaged12-17atbaseline(2014) Outcomesmeasuredpost-epidemic(2016) SURestimates,standarderrorsinparentheses PregnancyRisk (mainresult) Topquartile Bottom Quartile Above Median Top Quartile Above Median TopQuartile Above Median TopQuartile (1)(2)(3)(4)(5)(6)(7)(8) HeterogeneityDummy -12.2***-2.223.31-.1395.07**3.17-6.33***-3.92* (2.37)(2.90)(2.28)(2.80)(2.30)(3.04)(2.14)(2.18) ELATreatment|HDummy==1 9.84***-3.60.106.734-3.69*-2.221.74.270 (2.62)(2.70)(1.72)(2.56)(2.11)(2.81)(1.77)(2.18) ELATreatment|HDummy==0 -3.03**.206-.812-1.421.36-.237-3.71*-.947 (1.30)(1.45)(1.87)(1.50)(1.64)(1.42)(1.90)(1.57) DifferenceTreatmentEffects[ 3 ,p-value] {.000}{.222}{.721}{.470}{.061}{.530}{.041}{.663} ControlMeanatBL|HDummy==0 48.851.247.248.148.648.353.651.8 Observations 2,3812,3812,3812,3812,3812,3812,3812,381 Definitions: Timeallocationdatawascollectedbyaskingrespondentstoallocateasetof25beadsonaboardwith6circlesrepresenting:"LearningActivities","IGA","Socializing","HouseholdChores", "Sleep"and"Other".TheLearningcategoryincludesschooling,vocationaltrainingandstudytime."IGA"includespaidandunpaidworkofanykind.Respondentwerethenaskedtoallocatebeadsinto eachcircleinawaythatrepresentstimeallocationinanaverageday,anddatapointswerelaterconvertedintoweeklyhours.Asimilarprocedurewasimplementedtorecordallocationofsocializing timeacrossthefollowingactivities:"Friends","Men","Alone","Church","Volunteer"and"Other".Theexactphrasingforthe"Men"categoryis,"Withboysormenyouhaveasexualrelationshipwith.". Usingthenumberofhoursspentontotalsocializingtimefromthepreviousstep,theseallocationswerelaterconvertedintoweeklyhours.AProgressoutofPovertyIndex(PPI)wascollectedforeach householdsinstudyvillagesduringthecensusthattookplacebeforethebaselinesurvey.Thisusedtoconstructthe VillagePoverty measure.Dataonvillageinfrastructurewascollectedduringthe communityleaderssurveythattookplacein2015.Thetypesofinfrastructuretakenintoconsiderationare:Constructedwaterwell,Telecentre/chargings
tation,Villagebarray,Marketstructure,Primary school,Secondaryschool,Vocationaltrainingcenter,Healthcenter,Publictoilet,Communitybank,Mobilebankingagent,Dryingfloor.Foreachtypeofinfrastructure,datawascollectedonwhetherthe villagehasoneand,ifnot,howdistanttheclosesfacilityofthattypeis.Theformerwasusedtoconstructa VillageInfrastructureIndex ,thelatterwasusedtoconstructan IndexofClosenessto Infrastructure( imputingadistanceofzeroifthetypeofinfrastructureispresentwithinthevillage).Bothindicesareinverse-covarianceweightedfollowingAnderson(2008). Notes: ***,**and*denotesignificanceatthe1%,5%,and10%levels.Controlvariablesinclude:age,PPIscore,householdsize,illiteracy,villagesize(nrofdwellings),villageaveragePPIscore, distancesformkeyfacilities(clinic,secondaryschoolandmarket),adummyequaltooneifthevillageisapoliticalstronghold(i.e.hasaresidentparamountand/orsectionchief),thenumberofNGOs activewithinthevillagebeforetheEbolaoutbreak(excludingBRAC),shareofChristianhouseholds,distancefromFreetownanddistancefromKailahun.Allspecificationsincludedummiesforthe randomizationstrata(district)anderrorsareclusteredattheunitofrandomization(village). TimeinLearningActivities(hrs/wk) HeterogeneityDummy: VillagePovertyVillageInfrastructure Closenessto Infrastructure TableA8:Skills Sample:Girlsaged12-17atbaseline(2014) Outcomesmeasuredpost-epidemic(2016) OLSestimates,standarderrorsinparentheses,p-valuesinbraces Literacy [0,1] Numeracy [0,1] Entrepreneurial Confidence[0,1] Financial Literacy[0,1] (1)(2)(3)(4) PregnancyRisk -.120***-.071***.003-.050* (.033)(.025)(.023)(.030) ELATreatment|HighPregnancyRisk .087***.068***.014.047 (.031)(.024)(.021)(.030) ELATreatment|LowPregnancyRisk -.050**-.020.010-.015 (.021)(.015)(.014)(.017) DifferenceTreatmentEffects[ 3 ,p-value] {.000}{.005}{.854}{.073} ControlMeanatBL|LowPregnancyRisk .246.424.642.588 Observations 2,3822,3822,3812,382 Outcomes: LiteracyandNumeracyareself-reportedabilitiestoperform:"Readingsimplethingslikelabelsoncontainers"; "Readingcomprehension,writingcompletesentencesorlongerpassages";"Basiccounting,simpleaddition/subtractions,and measurement";"Workingwithfractions,multiplyinganddividing,doin
galgebraorbasicbookkeeping".Answerstotheskillself- assessmentsarethenaggregatedandrescaledinameasurethatrangesfrom0to1,withthelatterindicatingmoreadvanced proficiency.EntrepreneurialConfidenceisanindexthatmeasuresrespondents'self-reportedabilityto:runabusiness,identify businessopportunities,obtaincredit,saveandinvest,managefinancialaccounts,bargainprices,manageemployeesandsearch forjobs.FinancialLiteracyisassessedthrough8simpleproblemsrelatingtomarketprices,interestrates,borrowingand budgeting.Thenumberofcorrectanswersisrescaledinanindexrangingfrom0to1. Notes: ***,**and*denotesignificanceatthe1%,5%,and10%levels.ThemainresultsareestimatedwithanANCOVA specification.Controlvariablesinclude:age,PPIscore,householdsize,illiteracy,villagesize(nrofdwellings),villageaveragePPI score,distancesformkeyfacilities(clinic,secondaryschoolandmarket),adummyequaltooneifthevillageisapolitical stronghold(i.e.hasaresidentparamountand/orsectionchief),thenumberofNGOsactivewithinthevillagebeforetheEbola outbreak(excludingBRAC),shareofChristianhouseholds,distancefromFreetownanddistancefromKailahun.AllColumns controlforalsoforbaselinevaluesoftherelevantoutcomevariable.Allspecificationsincludedummiesfortherandomization strata(district)anderrorsareclusteredattheunitofrandomization(village). TableA9:SocialNetworks Sample:Girlsaged12-17atbaseline(2014) Outcomesmeasuredpost-epidemic(2016) SURestimates,standarderrorsinparentheses Numberoftiestoothersofthistypeof relationship: FriendshipBusiness TalkAbout IntimateTopics Credit/Finance (1)(2)(3)(4) PregnancyRisk -.210-.093.026-.207 (.163)(.195)(.208)(.306) ELATreatment|HighPregnancyRisk .337*.459**.199.579* (.175)(.189)(.207)(.312) ELATreatment|LowPregnancyRisk .169*.104.046.048 (.097)(.129)(.113)(.142) DifferenceTreatmentEffects[ 3 ,p-value] {.401}{.131}{.527}{.122} ControlMeanatBL|LowPregnancyRisk 2.17.724.803.876 Observations 1,5991,5991,5991,599 Outcomes: Networkdegreesofeachtypewerecomputedfromanswerstothefollowingquestions: Friends "whoareyourclosestfriends?"; Business "Ifyouwantto talkaboutissuesrelatedtoincome-generatingactivities,forexampleconcerningyouremployer,yourbusiness,agriculture,useofresou
rcesetc.whomdoyoutalk to?"; IntimateTopics "Whodoyoutalktoaboutintimatetopicssuchasrelationshipswithboysandmen(husband,boyfriend,partner),gender-basedviolence, personalhygiene,etc.?"; Credit/Finance "Whodoyoutalktoaboutissuesrelatedtofinanceandaccesstocredit?".Inordertohelprespondentswiththistask,they wereprovidedwith(orreadoutloud)alistofyoungwomenresidingintheirvillage,whichwerecompiledaspartofthecensusesthattookplacebeforebaselineand midline. Notes: ***,**and*denotesignificanceatthe1%,5%,and10%levels.Controlvariablesinclude:age,PPIscore,householdsize,illiteracy,villagesize(nrof dwellings),villageaveragePPIscore,distancesformkeyfacilities(clinic,secondaryschoolandmarket),adummyequaltooneifthevillageisapoliticalstronghold (i.e.hasaresidentparamountand/orsectionchief),thenumberofNGOsactivewithinthevillagebeforetheEbolaoutbreak(excludingBRAC),shareofChristian households,distancefromFreetownanddistancefromKailahun.Allspecificationsincludedummiesfortherandomizationstrata(district)anderrorsareclusteredat theunitofrandomization(village). TableA10:EbolaCases Outcomesmeasuredpost-epidemic(2016) OLSestimates,standarderrorsinparentheses,p-valuesinbraces AnyEbolaCaseWithin:HouseholdFamilyNetwork Friends Network HouseholdFamilyNetwork Friends Network (1)(2)(3)(4)(5)(6) PregnancyRisk -.008 -.007-.004.004 .043 .055 (.026) (.071)(.074)(.023) (.057) (.061) ELATreatment|HighPregnancyRisk .011-.008-.014.006-.018-.025 (.033)(.075)(.079)(.029)(.057)(.063) ELATreatment|LowPregnancyRisk -.023-.045*-.050*-.016-.020-.021 (.014)(.025)(.028)(.017)(.032)(.035) DifferenceTreatmentEffects[ 3 ,p-value] {.365}{.645}{.675}{.535}{.978}{.952} ControlMeanatBL|LowPregnancyRisk .034.152.178.032.144.139 Observations 2,6572,8942,8942,6172,6172,617 Aged12-17atBaseline(2014) Notes: ***,**and*denotesignificanceatthe1%,5%,and10%levels.Controlvariablesinclude:age,PPIscore,householdsize,illiteracy,villagesize(nrofdwellings),villageaveragePPI score,distancesformkeyfacilities(clinic,secondaryschoolandmarket),adummyequaltooneifthevillageisapoliticalstronghold(i.e.hasaresidentparamountand/orsectionchief),the numberofNGOsactivewithinthevillagebeforetheEbolaoutbreak(excl
udingBRAC),shareofChristianhouseholds,distancefromFreetownanddistancefromKailahun.Allspecifications includedummiesfortherandomizationstrata(district)anderrorsareclusteredattheunitofrandomization(village). Aged18-25atBaseline(2014) Outcomes: ThesampleincludesthoseindividualstrackedfromBaselinetoEndlineplusrespondentsresampledatendlinethathavebeenresidinginthesamevillagesincebeforethe Ebolaoutbreak. TableA11:Robustness Coefficientestimates,standarderrorsinparentheses,p-valuesinbraces A.Learning Activities (hrs/week) C.Outof Wedlock Pregnancy D.Transactional Sex AgeCohort: 12-1712-1718-2512-1718-25 (1)(2)(3)(4)(5) PregnancyRisk -12.2***1.27***.620.072*-.002 (2.35)(.444)(.551)(.039)(.018) Baseline,RI {.003}{.037}{.404}{.064}{.949} Baseline,MHT {.002}{.108}{.617}{.349}{.925} Nocontrols -12.7***.966.529.064-.014 (2.81)(.595)(532)(.042)(.017) Baseline,withindistrictpregnancyrisk -8.51**2.01***.032.100.000 (3.42)(.621)(.546)(.040)(.022) ELATreatment|HighPregnancyRisk 9.69***-1.86***-1.68***-.075**.054*** (2.61)(.452)(.475)(.038)(.021) Baseline,RI {.016}{.003}{.004}{.043}{.098} Baseline,MHT {.015}{.007}{.029}{.793}{.130} Nocontrols 8.72***-1.59***-1.75***-.066*.054** (3.02)(.533)(.466)(.040)(.022) Baseline,withindistrictpregnancyrisk 7.46**-2.59**-1.32**-.131***.047** (3.41)(.661)(.530)(.041)(.023) ELATreatment|LowPregnancyRisk -2.96**-.602***-1.32***-.010.018 (1.30)(.211)(.328)(.017)(.011) Baseline,RI {.062}{.012}{.000}{.583}{.151} Baseline,MHT {.201}{.108}{.007}{.303}{.449} Nocontrols -4.12**-.505*-1.33***.001.014 (1.72)(.269)(.326)(.018)(.011) Baseline,withindistrictpregnancyrisk -1.41-.662***-1.43***-.010.021** (1.41)(.188)(.306)(.016)(.011) F-statistic[p-value],Baseline {.000}{.000}{.000}{.168}{.022} Observations 2,3452,3792,4012,3822,244 B.TimeSpentwith Men(hrs/week) Outcomes: Alloutcomesmeasuredatendline.Respondentswereprovidedasetof25beadsandaboardwith6circlesrepresenting:"Education","IGA", "socializing","HouseholdChores","Sleep"and"Other".TheLearningcategoryincludesschooling,vocationaltrainingandstudytime.Theexactphrasingfor the"Men"categoryis,"Withboysormenyouhaveasexualrelationshipwith".TransactionalSexisdefinedas,"Rece
ivinganythingsuchasmoney,gifts,help withschoolworkorsomethingelseinexchangeforsexualintercourse." Notes: ***,**and*denotesignificanceatthe1%,5%,and10%levels.Controlvariablesinclude:age,PPIscore,householdsize,illiteracy,villagesize(nrof dwellings),villageaveragePPIscore,distancesformkeyfacilities(clinic,secondaryschoolandmarket),adummyequaltooneifthevillageisapolitical stronghold(i.e.hasaresidentparamountand/orsectionchief),thenumberofNGOsactivewithinthevillagebeforetheEbolaoutbreak(excludingBRAC), shareofChristianhouseholds,distancefromFreetownanddistancefromKailahun.Allspecificationscontrolforbaselinevaluesoftheoutcome,withColumn 4controllingforanypregnancyatbaseline.Allspecificationsincludedummiesfortherandomizationstrata(district),anderrorsareclusteredatthecommunity level.RandomizationInferencepreformedbycomparingtheparametersfrombenchmarkspecificationstotheestimateddistributionofeachparameterunder originaltreatmentallocation,andparametersestimatedusingtheseplacebotreatments(1,000permutations).Reportedp-valuestestthenullhypothesisofthe parametersbeingzerousingthisempiricallyestimatedparameterdistribution.P-valuesadjustedformultiplehypothesistestingarecomputedusingthestep- downprocedureofRomanoandWolf[2016](1,000bootstrapiteration).The NoControls rowsincludebaselineoutcomevaluesasindependentvariables, districtFEandstandarderrorsareclusteredatthevillagelevel.Thelastrobustnesscheckemploysadifferentmeasuresofpregnancyrisk.This"within district"measureisadummyequalto1ifthevillageisinthetopquartileofthedistributionofthepregnancyriskindexwithineachdistrict.TheF-testatthe TableA12:SocialDesirability Outcomesmeasuredpost-epidemic(2016) InColumns1to6,SURestimates,standarderrorsinparentheses InColumns7to12,OLSestimates,standarderrorsinparentheses Marlowe-Crowne Index Below Median Above Median Marlowe-Crowne Index Below Median Above Median Marlowe-Crowne Index Below Median Above Median Marlowe-Crowne Index Below Median Above Median (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12) PregnancyRisk .611*.317.960*.317-.048.739-.016-.054.077.021.010.046 (.344)(.407)(.563)(.341)(.394)(.501)(.036)(.049)(.086)(.023)(.030)(.050) ELATreatment|HighPregn
ancyRisk -1.13***-1.35***-.877*-1.40***-1.70***-1.14**.064*.113**-.028.055**.069***.035 (.366)(.465)(.485)(.387)(.422)(.491)(.037)(.049)(.081)(.027)(.026)(.057) ELATreatment|LowPregnancyRisk -1.04***-1.26***-.774**-1.67***-2.02***-1.36.001-.003.015.017.019.018 (.272)(.405)(.356)(.319)(.401)(.370)(.019)(.032)(.019)(.012)(.023)(.017) Marlowe-CrowneIndex .041-.187*-.009.002 (.104)(.108)(.008)(.006) DifferenceTreatmentEffects[ 3 ,p-value] {.297}{.801}{.825}{.423}{.383}{.609}{.125}{.042}{.611}{.210}{.176}{.781} ControlMeanatBL|LowPregnancyRisk 2.342.522.187.697.697.68.158.148.166.045.037.053 Observations 1,2916596321,5527637891,4407057351,440705735 Notes: ***,**and*denotesignificanceatthe1%,5%,and10%levels.Controlvariablesinclude:age,PPIscore,householdsize,illiteracy,villagesize(nrofdwellings),villageaveragePPIscore,distancesformkeyfacilities(clinic,secondaryschooland market),adummyequaltooneifthevillageisapoliticalstronghold(i.e.hasaresidentparamountand/orsectionchief),thenumberofNGOsactivewithinthevillagebeforetheEbolaoutbreak(excludingBRAC),shareofChristianhouseholds,distance fromFreetownanddistancefromKailahun.Allspecificationsincludedummiesfortherandomizationstrata(district)anderrorsareclusteredattheunitofrandomization(village). TransactionalSex GirlsAged18-25atbaseline UnwantedSex GirlsAged18-25atbaseline TimeSocializingwithMen(hrs/wk) GirlsAged18-25atbaseline TimeSocializingwithMen(hrs/wk) GirlsAged12-17atbaseline Outcomes: Timeallocationdatawascollectedbyaskingrespondentstoallocateasetof25beadsonaboardwith6circlesrepresenting:"LearningActivities","IGA","Socializing","HouseholdChores","Sleep"and"Other".TheLearningcategory includesschooling,vocationaltrainingandstudytime."IGA"includespaidandunpaidworkofanykind.Respondentwerethenaskedtoallocatebeadsintoeachcircleinawaythatrepresentstimeallocationinanaverageday,anddatapointswerelater convertedintoweeklyhours.Asimilarprocedurewasimplementedtorecordallocationofsocializingtimeacrossthefollowingactivities:"Friends","Men","Alone","Church","Volunteer"and"Other".Theexactphrasingforthe"Men"categoryis,"With boysormenyouhaveasexualrelationshipwith.".Usingthenumberofhoursspentontotal
socializingtimefromthepreviousstep,theseallocationswerelaterconvertedintoweeklyhours.Duringthelong-termfollow-upin2019,trackedrespondents wereadministered13questionsformtheMarlowe-CrowneSocialDesirabilitytest.Answersfromthesequestionswherelateraggregatedinanainverse-covarianceweightedindex(Anderson,2008)ofsocialdesirability. Notes PanelA:Source:UNDP.TheGenderInequalityIndexaggregatesinformationon:maternalmortalityrates,adolescentfertilityrates,educationbygender,femaleheldparliamentaryseats,andinequalityinlabormarket participation.Theindexrangesfrom0to1,withavalueof0indicatingperfectequality. PanelB:SourceWorldBankWDI PanelC:SourceWorldBankWDI,modelledestimatesofmaternalmortalityper100livebirths. PanelD:SourceWHO,Brightercoloredbarsrepresentnursesandthedarkerbarsrepresentdoctorsper1,000population.*Datafor2008,**Datafor2009,theremainingdatapointsarefor2010. FigureA1:SierraLeoneonContext PanelA.GenderInequalityIndex2013 PanelB.AdolescentFertilityRate,Ages15-19,2013(%) PanelC.MaternalMortalityRatio,2013(%)PanelD.PhysiciansandNursesper1,000population FigureA2:EbolainSubSaharanAfrica Notes: AllpanelsreportdatafromtheELAClubMonitoringSurveyadministeredinJuneandJuly2015toclubmentors.PanelCreportsthenumberofgirlsineachvillagethatregisteredasELAmemberswhentheclubfirstopened. InPanelD,attendanceismeasuredatthetimeofthemonitoringsurvey. FigureA3:ELAImplementation A.ShareofELAClubsContinuously,EverOpen B.ELAProgramDelivery(Shareofclubsofferingaparticularservice, conditionalonbeingopenandtreatmentassignment) C.ELAClubMembership,byVillage D.ELAAttendance/MemberRatio,byVillage QuarantinesLifted Note :Eachpartialcorrelationisestimatedbyregressingthevillage-levelpregnancyriskdummyoneachvariableofinterest,withdistrict(strata)fixed effectsandrobuststandarderrors.Datafromthe2013DemographicandHealthSurvey,anationallyrepresentativesurveyadministeredto12,629 respondents,andfromthe2007NationalPublicServiceSurvey,anationallyrepresentativesurveycollectedbytheDecentralizationSecretariattomonitor satisfactionwithpublicserviceprovision.Thesampleincludes6,300households. FigureA4:Ebola-RelatedPregnancyRiskandLocalAreaCharacteristics B:ChiefdomCharacteri
stics A:VillageCharacteristics FigureA5:ELAClubFunctioning,byPregnancyRisk Notes: DatafromtheELAClubmonitoringsurveywerecollectedinOctober2015.InJanuary2015,quarantineswereofficiallyliftedbythe SierraLeoneanGovernment. quarantineslifted quarantineslifted LowPregnancyRisk HighPregnancyRisk WeeklyHours,90%ConfidenceInterval ReferenceGroup:Bottom60%ofDisruptions'Distribution Outcomes: TimeallocationdatawascollectedbothatBaselineandEndline.Respondentswereprovidedasetof25beadsandaboardwith 6circlesrepresenting:"Education","IGA","Socializing","HouseholdChores","Sleep"and"Other".TheEducationcategoryincludesschooling, vocationaltrainingandstudytime."IGA"includespaidandunpaidworkofanykind.Respondentwerethenaskedtoallocatebeadsintoeach circleinawaythatrepresentstimeallocationinanaverageday.SocializingtimeallocationdatawascollectedbothatBaselineandEndlinein asimilarway.Thecategoriesrecordedwere:"Friends","Men","Alone","Church","Volunteer"and"Other".Respondentswerethenaskedto allocatebeadsintoeachcircleinawaythatrepresentstimeallocationinanaverageweek.Thedatapointswerelaterconvertedintoweekly hoursusingrecordedtotalsocializingtimeformthefirstexercise. Notes :DecilesofthedistributionofEbola-induceddisruptionsareontheX-axis.Errorbarsrepresent90%confidenceintervals.Bars representcoefficientestimatesviaSURregressiononalltimeuse/socializingcategoriesexcluding"other".Controlvariablesinclude:age,PPI score,householdsize,illiteracy,villagesize(nrofdwellings),villageaveragePPIscore,distancesformkeyfacilities(clinic,secondaryschool andmarket),adummyequaltooneifthevillageisapoliticalstronghold(i.e.hasaresidentparamountand/orsectionchief),thenumberof NGOsactivewithinthevillagebeforetheEbolaoutbreak(excludingBRAC),shareofChristianhouseholds,distancefromFreetownand distancefromKailahun.Allspecificationsincludedummiesfortherandomizationstrata(district)anderrorsareclusteredattheunitof randomization(village).Errorbarsrepresent90%confidenceintervals. FigureA6:ImpactsbyIntensityofPregnancyRisk GirlsAged12-17atBaseline A:TimeSpentinLearning,IncomeGenerationorELA WeeklyHours,90%ConfidenceInterval ReferenceGroup:Bottom60%ofDisruptions'Distribution PanelB:T