Community-based reintegration in Aceh : assessing the impacts of BRA-KDP

53714  IndonesianSocialDevelopmentPapers Since1998,Indonesiahasbeenundergoingamomentouspoliticalandeconomictransition.The fall of the New Order, the economic crisis, and radical decentralization have changed the political, economic and social context. Within this new context, power relations are in flux, identities are being renegotiated, and institutions are changing. Changes in incentives, and in the role of formal and informal institutions at various levels, have altered the ways in which individualsandgroupsrelatetoeachotherandthestate.Understandingthisnewcontext,and the ways in which various actors (national and international) can promote progressive social changeisimportant. The Indonesian Social Development Papers series aims to further discussion on a range of issues relating to the current social and political context in Indonesia, and to help in the generation of ideas on how democratic and peaceful transition can be supported. The series willcoverarangeofissuesincludingconflict,development,corruption,governance,theroleof thesecuritysector,andsoon.Eachpaperpresentsresearchonaparticulardimensionofsocial developmentandofferspragmaticpolicysuggestions.Papersalsoattempttoassesstheimpact ofvariousinterventions--fromlocalandnationalactors,aswellasinternationaldevelopment institutions--onpreexistingcontextsandprocessesofchange. The papers in the series are works in progress. The emphasis is on generating discussion amongst different stakeholders--including government, civil society, and international institutions--rather than offering absolute conclusions. It is hoped that they will stimulate further discussions of the questions they seek to answer, the hypotheses they test, and the recommendationstheyprescribe. PatrickBarron(serieseditor)pbarron@worldbank.org  CommunityBased Reintegration in Aceh AssessingtheImpactsofBRAKDP PatrickBarron MacartanHumphreys LauraPaler JeremyWeinstein December2009 IndonesianSocialDevelopmentPaperNo.12 PapersintheIndonesianSocialDevelopmentseriesarenotformalpublicationsoftheWorldBank.They arepublishedinformallyandcirculatedtoencouragediscussionandcommentbetweenthoseinterested inIndonesiandevelopmentissues.Thefindings,interpretations,judgments,andconclusionsexpressed in the paper are those of the authors and should not be attributed to: the World Bank and affiliated organizations; members of the World Bank's Board of Executive Directors or the governments they represent;oranyofthefundingagencies. ThefullrangeofpublicationsassociatedwiththebroaderstudyoflocalconflictinIndonesia(ofwhich thisreportisaproduct)isavailableonlineatwww.conflictanddevelopment.org. Emailaddressesforcorrespondence: pbarron@worldbank.org macartan@gmail.com lbp2106@columbia.edu jweinst@stanford.edu Copiesofthispaperareavailablefrom: PNPMSupportFacility JalanDiponegoroNo.72 Jakarta10310Indonesia Tel:+62(0)213148175 Fax:+62(0)2131903090  Preface TheendoftheconflictinAcehledtothearrivalofarangeofprogramsaimedat`reintegrating' formercombatantsandprovidingassistancetoconflictaffectedgroups.TheBRAKDPprogram wasaninnovativeattemptbylocalandnationalgovernmentstoemploylessonslearnedfrom successful past communitydevelopment work to postconflict Aceh. The program, the design andimplementationofwhichwassupportedbytheWorldBank,deliveredoverUS$20million to conflictaffected villages, aiming to support the welfare of conflict victims while building socialcohesionandtrustinthestate. Diditwork?Whatimpactsdidithaveandwhy?WhatcantheexperienceofBRAKDP,andthe broaderreintegrationeffortsinAceh,tellusabouthowgovernments,donorsandcivilsociety cansupportpeaceconsolidationinpostconflictarenas? This paper aims to contribute towards answers to these questions. It outlines the results of surveys of households, former combatants and village heads conducted across Aceh, which were designed to determine project impacts. It should be read in conjunction with a complementary paper, Delivering Assistance to ConflictAffected Communities: the BRAKDP Program in Aceh (Indonesian Social Development Paper number 13), which in addition to further analysis of the Aceh Reintegration Livelihood Surveys (ARLS) data set discussed here, provides evidence from the project's Monitoring Information Supervision (MIS) system, from supervision missions and from qualitative fieldwork. In addition to providing a deeper understanding of the BRAKDP programs and its impacts, the papers aim to contribute to a small but growing literature evaluating postconflict programs. We hope that the BRAKDP experience will serve as a laboratory through which broader processes of change can be understoodinasocietytransitioningfromconflict. PatrickBarron Conflict&Developmentteam WorldBank,Indonesia i  ExecutiveSummary This paper describes the results of an evaluation of the CommunityBased Reintegration Assistance for Conflict Victims program (BRAKDP) funded by the Aceh PeaceReintegration Agency (Badan ReintegrasiDamai Aceh, or BRA) and implemented with support from the MinistryofHomeAffairsandtheWorldBank.BRAKDPwasdesignedtoassistconflictvictims acrossAcehaspartofthereintegrationprogramemergingfromtheHelsinkipeacedeal,which brought to an end a 30year conflict between the separatist GAM movement and the GovernmentofIndonesia.US$21.7millionoffundswerechanneledthroughtheGovernment's KecamatanDevelopmentProgram(KDP),whichhasbeenoperatinginAceh,andelsewherein Indonesia,since1998.Blockgrantsweredeliveredtocommunitieswhodecidedonhowfunds should be spent. As with other communitydriven development (CDD) projects, the program emphasized participation, local ownership, and transparency in giving communities control overthechoiceandimplementationofprojects. BydesigntheprogramwastobeimplementedthroughoutAcehintwophases.Intheevent, however,onlythefirstphasewasimplemented.ThisphasewasimplementedbetweenAugust 2006and2007,andtargeted1,724villages,aroundonethirdofthetotalinAceh,in67rural subdistricts. First phase subdistricts (kecamatan) were selected on the basis of a rule which took account of theextentto which areas wereaffectedbyconflict and how effectivelythey haddisbursedKDPmoniesinthepast;allvillagesinselectedsubdistrictsreceivedblockgrants. The program had both economic and social goals. It aimed to deliver quick assistance to conflictaffectedvillagersinordertoimprovematerialwellbeingintheshortterm.Inaddition, itsoughttopromotesocialcohesion,strengthenvillageleveldecisionmakinginstitutions,and cultivategreaterfaithingovernmentalinstitutionsintheaftermathoftheconflict. This evaluation uses data from a household and village head survey conducted after the programtoexaminetheextenttowhichthesegoalshavebeenmet.Theevaluationemploysa secondbestapproachtoestimatetheimpactofBRAKDPonmaterialwellbeing,socialcohesion, and trust in government. Because the program was not randomized, simple comparison of outcomes in project and control areas are unlikely to produce valid inferences of program effects. Indeed our data suggests that biases arising from selection effects would be considerable.Insteadtheevaluationreliesonaneconometricstrategythat(a)usesaformof matchingtoidentifysuitablesubdistrictsthatdidnotreceivetheprogramtoserveasacontrol group, (b) accounts for variables used in the selection process, and (c) uses an instrumental variablesapproachtodealwithissuesofnoncompliancewithtreatmentassignment. The evaluation finds that there were a large number of beneficiaries. An estimated 530,000 individuals live in households that directly received assistance. There was widespread participationintheprogram(over200,000peopleattendedBRAKDPmeetings)andpoorerand femaleheaded households were as likely to attend as others. The program was successful to some extent in reaching out to more marginalized groups in villages. Targeting of conflict victimswashoweverlimited.Overall24percentofconflictvictimsinthestudyareareceived benefits; within project areas, 44 percent of conflict victims were supported through the ii program,arateonlymarginallyhigherthanthatfornonvictims(40percent).Withinvillages, conflictvictimswerenomorelikelythannonvictimstoreceivebenefits.Howeverbecausethe designoftheprogramprovidedmorefundstoareaswithmoreexposuretoconflict,onaverage, conflict victims who received cash received 13 percent more than nonvictims. Particular categoriesofconflictvictimswerelikelytoreceivehigheramounts:the`mostconflictaffected' received19percentmorethannonvictims. Theprogramwasalsowellreceived.Peopleweremorelikelytobeawareoftheprogramthan regular KDP and reported it to be very popular. Ninetyfour percent of villagers in treatment areassaidtheprogramwashelpful,andthisfigureroseto96percentofconflictvictimsand97 percentforthemostconflictaffected. We find evidence that BRAKDP is associated with a strong set of welfare gains and improvements in perceptions of wellbeing. The participation of villages in the program is associatedwithan11pointdeclineintheshareclassifiedas"poor"asreportedbyvillageheads. Typically in programs of this form block, grants are used for community projects selected by villages. In this case, however, the majority of villages prioritized the provision of cash to individuals and groups, primarily for economic purposes. Our data suggest that these cash disbursements are associated with an increased ownership of assets (notably engines and motorcycles) among households in general and conflict victims in particular. The program is also associated with an increase in the farming of productive land (corresponding approximately to a doubling in land use for conflict victims). Conflict victims in areas that receivedtheprogramaresignificantlymorelikelytoreporttheirwelfarehasimprovedinthe pastyearthanarevictimsinvillagesthatdidnotgettheprogram.Thereis,however,nodirect evidenceofwelfareimpactsinotherareassuchasschoolattendance,healthandemployment levels. Unsurprisingly, given the small proportion of project funds used for public goods, the programisnotassociatedwithchangesinthelevelofcommunityinfrastructure. Theevidenceforimprovementsinsocialcohesionandstrongerrelationsbetweencitizensand governmentisweak.Levelsofsocialacceptanceofreturninggroups,reportedsocialtensions, divisionsandconflictandcommunityefficacyaresimilarbetweenthosevillagesthatreceived theprogramandthosethatdidnot.BRAKDPisassociatedwithanincreaseinparticipationin women's groups, but there is no evidence of an overall increase in associational activities. ThereissomeevidencethatBRAKDPresultsinlowerlevelsofacceptanceofexcombatantsby conflict victims. This could be because excombatants received funds that were meant for civilianconflictvictims,althoughthereisnoevidencethatthey`captured'theprogram.Itmay alsobearesultoffrustrationsfromexcombatantsthattheycouldnotbenefitmoreorbecause ofempowermentofcommunitiestostanduptoexcombatantdemands.Finally,thereisonly minimal evidence that exposure to BRAKDP resulted in higher levels of trust in village and higherlevelgovernments. While there is some evidence in the broader literature that CDD programs improve social cohesionandincreaseacommunity'scollectivecapacity,wedonotfindevidencethatBRAKDP had these effects. One reason may be that the program only ran for one year, limiting the iii extenttowhichsuchgains,whichtendtobuildovermultipleprogramcycles,couldeventuate. Anotherreasonmaybebecausecommunitydecisionstoemphasizeprivategoodslimitedthe opportunities for collective implementation. This latter account highlights a possible tension withintheCDDmodel.ManyofthegoalsofCDDmaydependuponprocessesthatarebrought into play conditional on particular types of activities (joint selection of projects, community oversight of implementation, etc.). Yet certain kinds of activities that communities might choose are less likely to encourage interaction, limiting some of the social gains that CDD projectsmightpurporttohave.CDDprogramscanalsoleadtoincreasedtensionsbypromoting competitionoverfiniteresources.Inthelongrun,thismayleadtoastrongerbasisforpeace, by empowering groups and building local institutions, but it can also create divisions in the shortrun.Inpostconflictcontexts,itisnecessarytoweighthese(potential)effects. iv  TABLEOFCONTENTS Preface.............................................................................................................................................i ExecutiveSummary ........................................................................................................................ii TableofContents ...........................................................................................................................v TableofTables ............................................................................................................................. vii TableofFigures ............................................................................................................................. vi Acknowledgements ..................................................................................................................... viii 1 Introduction..............................................................................................................................1 1.1 ApproachestoReintegration.............................................................................................1 1.2 CommunityBasedReintegrationAssistanceforConflictVictimsinAceh.........................2 1.3 TheSelectionofProjectAreas...........................................................................................4 1.4 AnalyticalFrameworkandMethods..................................................................................6 Selectingappropriatecomparisonunitsexpost .....................................................................8 Datacollection ......................................................................................................................11 Estimationstrategy ...............................................................................................................12 2 ImplementingBRAKDP ..........................................................................................................14 2.1 WhoReceivedAssistancethroughBRAKDP? .................................................................14 Whoisaconflictvictim?........................................................................................................14 AreconflictvictimsconcentratedintheareasthatreceivedBRAKDP? ...............................16 Doconflictvictimsbenefitmorethanothersintreatmentareas?........................................19 Conclusionsontargeting.......................................................................................................21 2.2 HowWereBRAKDPFundsSpent? ..................................................................................22 Directbenefits .......................................................................................................................23 Projects..................................................................................................................................26 2.3 ParticipatinginBRAKDP .................................................................................................28 2.4 PerceptionsofandProblemswithBRAKDP ...................................................................29 3 ImpactsonWelfare ................................................................................................................34 3.1 PovertyProfile .................................................................................................................34 3.2 AssetIndex ......................................................................................................................36 3.3 HouseholdInfrastructure ................................................................................................38 3.4 LandUse ..........................................................................................................................39 3.5 EmploymentandWages..................................................................................................41 3.6 EducationandHealth ......................................................................................................42 3.7 PublicGoods ....................................................................................................................43 3.8 WelfarePerceptions ........................................................................................................44 3.9 ConclusionsonWelfare ...................................................................................................45 4 ImpactsonSocialCohesion ....................................................................................................46 4.1 SocialAcceptance ............................................................................................................46 4.2 SocialTensions.................................................................................................................50 4.3 ConflictResolution...........................................................................................................51 4.4 CollectiveEfficacy ............................................................................................................52 4.5 AssociationalLife .............................................................................................................53 4.6 ConclusionsonSocialCohesion.......................................................................................55 v 5 ImpactsonTrustinLocalGovernmentandStateSocietyRelations......................................56 5.1 TrustinCommunityDecisionMaking .............................................................................56 5.2 TrustinGovernment........................................................................................................58 Behavioralmeasures .............................................................................................................58 Attitudinalmeasures .............................................................................................................59 Knowledgeofgovernment ....................................................................................................60 5.3 AttitudesaboutGovernance ...........................................................................................61 5.4 ConclusionsonTrustinLocalGovernmentandStateSocietyRelations.........................62 6 CommunityDevelopmentandExCombatantReintegration.................................................63 6.1 ExCombatantCaptureofBRAKDPFunds?.....................................................................63 6.2 AlternativeExplanations..................................................................................................64 7 Conclusions.............................................................................................................................66 References....................................................................................................................................69 TABLEOFFIGURES FIGURE1.1:CONFLICTINTENSITYINACEH ............................................................................................................................. 5 FIGURE1.2:PROPENSITYSCORESFORTREATMENTANDCOMPARISONUNITS ............................................................................... 9 FIGURE1.3:CHARACTERISTICSOFSUBDISTRICTSSELECTEDINTOSTUDY ................................................................................... 10 FIGURE1.4:LOCATIONOFSUBDISTRICTSSELECTEDFORTHESTUDY......................................................................................... 11 FIGURE2.1:NUMBEROFCONFLICTVICTIMSINPROJECTANDCOMPARISONVILLAGES ................................................................. 18 FIGURE2.2:DISTRIBUTIONOFCONFLICTVICTIMSANDBENEFICIARIESWITHINVILLAGES .............................................................. 20 FIGURE3.1:SELECTIONINTOBRAKDPANDEFFECTSONPOVERTY ......................................................................................... 35 FIGURE3.2:MOTORCYCLEHOLDINGSANDSELECTIONINTOBRAKDP..................................................................................... 38 FIGURE3.3:LANDUSE .................................................................................................................................................... 40 FIGURE4.1:CONFLICTVICTIMSACCEPTANCEOFEXCOMBATANTS .......................................................................................... 48 FIGURE4.2:VILLAGEHEADACCEPTANCEOFEXCOMBATANTS ................................................................................................ 48 vi  TABLEOFTABLES TABLE1.1:VILLAGEBLOCKGRANTALLOCATIONS ................................................................................................................... 3 TABLE2.1:CORRELATIONBETWEENSUBJECTIVEANDOBJECTIVEMEASURESOFVICTIMHOOD ........................................................ 16 TABLE2.2:CONFLICTVICTIMPRIORITIZATION ...................................................................................................................... 17 TABLE2.3:BRAKDPBENEFICIARIES ................................................................................................................................. 21 TABLE2.4:USEOFFUNDS(MISDATA) .............................................................................................................................. 22 TABLE2.5:BRAKDPGOODS(TREATMENTSAMPLEONLY) ................................................................................................... 24 TABLE2.6:CASHBENEFITSBYCATEGORYOFCONFLICTVICTIM ................................................................................................ 25 TABLE2.7:HOWWEREBENEFITSUSED(TREATMENTSAMPLEONLY)....................................................................................... 26 TABLE2.8:PROJECTSAPPROVEDANDSUPPORTED ........................................................................................................... 27 TABLE2.9:PROJECTSSUPPORTED ...................................................................................................................................... 28 TABLE2.10:BRAKDPAWARENESS&PARTICIPATIONI(PROJECTAREASONLY)........................................................................ 28 TABLE2.11:BRAKDPAWARENESSANDPARTICIPATIONII(PROJECTAREASONLY)................................................................... 29 TABLE2.12:BRAKDPCONDUCT(TREATMENTAREASONLY)................................................................................................. 30 TABLE2.13:AWARENESSOFDEVELOPMENTPROJECT ........................................................................................................... 31 TABLE2.14:PROBLEMSINDEVELOPMENTPROJECTS ............................................................................................................. 32 TABLE2.15:HARMFUL/HELPFUL ....................................................................................................................................... 33 TABLE3.1:AGGREGATEMEASURESOFCOMMUNITYWELLBEING(BYVILLAGEHEADS) ................................................................ 34 TABLE3.2:ASSETINDEX .................................................................................................................................................. 36 TABLE3.3:ASSETSBYCATEGORY ....................................................................................................................................... 37 TABLE3.4:QUALITYOFHOUSING ...................................................................................................................................... 39 TABLE3.5:WATERSOURCE .............................................................................................................................................. 39 TABLE3.6:LANDUSE ...................................................................................................................................................... 40 TABLE3.7:EMPLOYMENT................................................................................................................................................. 41 TABLE3.8:AVERAGEDAILYWAGESOFLABORERS ................................................................................................................ 42 TABLE3.9:SICKNESS ....................................................................................................................................................... 42 TABLE3.10:INSCHOOL(INDIVIDUALS25YEARSOLD) ........................................................................................................ 43 TABLE3.11:COMMUNITYPUBLICGOODS ........................................................................................................................... 44 TABLE3.12:SUBJECTIVEPERCEPTIONSOFWELLBEING .......................................................................................................... 45 TABLE4.1:SOCIALACCEPTANCE ........................................................................................................................................ 47 TABLE4.2:GROUPSTHATBENEFITMORETHANOTHERS ........................................................................................................ 49 TABLE4.3:SOCIALTENSIONS ............................................................................................................................................ 50 TABLE4.4:ESCALATINGTOVIOLENCE ................................................................................................................................. 51 TABLE4.5:CONFLICTRESOLUTION ..................................................................................................................................... 52 TABLE4.6:COLLECTIVEEFFICACY ....................................................................................................................................... 52 TABLE4.7:COMMUNITYLEADERSHIPINPUBLICGOODSPRODUCTION ...................................................................................... 53 TABLE4.8:ASSOCIATIONALLIFE(BYVILLAGEHEADS) ............................................................................................................ 54 TABLE4.9:INVOLVEMENTINASSOCIATIONALLIFE ................................................................................................................ 55 TABLE5.1:SATISFIEDWITHDECISIONS................................................................................................................................ 56 TABLE5.2:VILLAGERS'ROLEINDECISIONMAKING .............................................................................................................. 57 TABLE5.3:POLITICALEFFICACY ......................................................................................................................................... 58 TABLE5.4:TRUSTINDISTRICTGOVERNMENT ...................................................................................................................... 59 TABLE5.5:TRUSTINVILLAGEAPPARATUS ........................................................................................................................... 59 TABLE5.6:CONFIDENCEINEFFECTIVENESSOFSUBDISTRICT,DISTRICT,ANDPROVINCIALGOVERNMENT ........................................ 60 TABLE5.7:AWARENESSOFGOVERNMENT .......................................................................................................................... 61 TABLE5.8:SUPPORTFORDEMOCRACY ............................................................................................................................... 61 TABLE6.1:SHAREOFEXCOMBATANTSANDCIVILIANSRECEIVINGBENEFITS(PROJECTAREASONLY) ............................................. 64 vii  ACKNOWLEDGEMENTS Theworkonthispaperwasfundedfrommultiplesources.AgrantfromtheUKDepartmentfor InternationalDevelopment(DFID)paidforthemainsurveysused.Analysiswassupportedbya grantfromtheWorldBank'sPostConflictFundandfromtheEmbassyoftheNetherlands.The Swedish government, through the Folke Bernadotte Academy, provided generous financial support, which made possible the participation of Humphreys, Paler and Weinstein in this research. Many other individuals and institutions have provided support. Makiko Watanabe, Adrian Morel and Yuhki Tajima helped develop the evaluation framework, the surveys and ensured theireffectiveimplementation.NeilsenIndonesiaoversawthesurveys'implementation.Cecilia Mohelpedinthedevelopmentandpilottestingoftheinstruments.MuslahuddinDaud,Susan Wong and Rob Wrobel provided critical feedback throughout. Scott Guggenheim provided useful comments on a draft of this paper. Pak Rusli (and his team from KDP) played an important role in supporting the evaluation. Key staff from the Ministry of Home Affairs, Bappenas and the KDP's National Management Committee were also supportive. Wawan HerwandiandRajyasriGayatrihelpedwithresearchandlogisticalassistance. TheauthorswouldalsoliketothankBRA'sleadership,bothPakIslahuddin(theagency'sformer head)whowasinstrumentalinimplementingtheprogram,andcurrentdirectorPakNurDjuli. viii  1. INTRODUCTION This paper describes the impact of Aceh's CommunityBased Assistance for Reintegration of Conflict Victims program (BRAKDP). The program, implemented between August 2006 and August 2007, sought to address the needs of conflict victims in a way that would not only generate improvements in welfare, but also increase community cohesion and strengthen relations between citizens and government. This evaluation assesses whether BRAKDP was successfulinachievingtheseobjectives. 1.1 ApproachestoReintegration As international donors accumulate lessons from nearly two decades of work in postconflict settings,thereisastrongconsensusthateffortstodisarm,demobilize,andreintegrateformer fighters are essential to the creation of a stable peace. But an initial, exclusive focus on the needsoffightersinpostconflictprogramshasgraduallygivenwaytoabroadervisionofthe reintegrationprocess--onethatemphasizestheneedsofconflictvictimsaswell. To better reflect the needs and interests of those who must accept fighters back into their communities and rebuild their own lives, governments and donors have begun to support communitybased approaches to facilitate reintegration as a complement to traditional DDR programs. 1 These communitybased processes are designed to engage perpetrators and victims alike in the process of community rebuilding in the hope that such efforts aid in the consolidationofpeaceandthepromotionofsocialcohesion.Suchprogramscanbeeffectivein reconstructing conflictaffected infrastructure, while strengthening the wartorn social fabric (Cliffe, Guggenheim and Kostner 2003). The conception and design of BRAKDP reflected this newemphasisoncommunitybasedapproachestoreintegration,withaparticularfocusonthe welfareofcivilianconflictvictims. Yet,despitethenewconsensusontheimportanceofcommunitybasedprocesses,andaswith programstargetedatformercombatants,therehavebeenfewattemptstoassessempirically theefficacyoftheseprogramsortotestthehypothesesimplicitinthisapproach(Humphreys andWeinstein2007;Muggah2008).2Studieshavetendedtotaketheformoflessonslearned fromindividualcasestudies.Withoutvariationonthekeyindependentvariable(i.e.,whether there was a program), analysts have been poorly positioned to say much about programs' causalimpacts.Hereweseektoundertakeasystematicinvestigationofjustthis,toexamine theimpactsofBRAKDPandprovideadeeperanalysisofwhatworked,forwhom,andwhy. 1 SeeSwedishMinistryofForeignAffairs(2006).TherecentCartagenaCongress,hostedbytheGovernmentof Colombia,broughttogethermorethan1,500academics,practitionersandpolicymakerstodiscusshow communitybasedapproachestoreintegrationcanbestbedesignedandimplemented. 2 MansuriandRao(2004)arguethatevidenceontheimpactsofcommunitydrivendevelopment(orCDD)projects lagsbehindtheextenttowhichtheyarebeingimplemented. 1 1.2 CommunityBasedReintegrationAssistanceforConflictVictimsinAceh Foralmost30years,separatistrebelswiththeFreeAcehMovement(GAM)andGovernmentof Indonesia(GoI)securityforcesengagedinamilitaryconfrontationinAceh.Whiletheconflict occurred in several stages during that period, civilians frequently suffered the brunt of hostilities. GoI forces adopted a strategy of trying to undercut popular support for GAM by terrorizingsuspectedciviliansupporters.GAM,too,killedorintimidatedsomewhorefusedto support the movement. This resulted in an unconfirmed number of instances of murder, torture,rape,displacementandpropertydestruction,from1990onwards.3 TheHelsinkipeaceagreementsignedbyGAMforcesandtheIndonesiangovernmentinAugust 2005 contained provisions for reintegration assistance to former combatants, pardoned political prisoners and conflict victims. The central government provided substantial sums to support this. A government body, the Aceh Peace Reintegration Agency (Badan Reintegrasi DamaiAcehorBRA)wasestablishedtoadministergovernmentfundsandtocoordinatewith donors providing assistance. After an initial failed attempt to manage individuallysubmitted proposals from victims, BRA opted for a communitybased approach to supporting conflict victims. BRA channeled its funds through the existing Government of Indonesia (and World Bank supported), Kecamatan Development Program (KDP), which had been operating across theprovincesince1998--producingBRAKDP.4AswithregularKDP,theprogramhadan`open menu'withonlyaverylimitednumberoftypesofactivitiesineligibleforfunding.Communities could choose to use their funds on private or public goods, or a mix. The KDP program was adaptedtoprovidefundsdirectlytovillages(KDPnormallyprovidesgrantstosubdistricts,with villages competing over funds) and facilitators were tasked with helping communities to identifyconflictvictimswhocouldbenefit.BRAKDPallocatedapproximatelyUS$21.7million to1,724villages,aroundonethirdofthetotalinAceh,in67subdistrictsinall17ruraldistricts inAceh,atargetareawithapopulationof1.1million. Accordingtoprogramdocumentation: TheprimaryfocusofBRAKDPistoassistconflictaffectedcommunitiesinimprovingtheirliving conditionsthroughprovisionofsmallprojectsthataccordwiththeirneeds.Italsoencourages peopletoovercomemistrustofgovernmentthatisaresultoftheconflictbydeliveringtangible outputsthatfitwithcommunities'priorities.Equallyimportantistheprocessbywhichvillagers identify, prioritize and implement their projects. A World Bank study on the efficacy of KDP suggests that by applying the principles of participation, transparency, local choice and accountability, communitydriven development programs help improve intergroup and state societyrelations,helpingareastobemoreimmunetoviolentconflicts.BRAKDPisanattempt to apply the communitydriven development approach for reintegration with the hope that it will improve relations between different groups such as exGAM combatants, IDPs [internally 3 SeeIOMandHarvardUniversity(2007),whichshowslevelsofabuseandtraumainAcehofasimilarscaleto AfghanistanandBosnia. 4 In2007,theGovernmentchangedthenameofKDPtotheNationalProgramforCommunityEmpowermentin RuralAreas(PNPMRural)andadaptedittobecomeitsflagshipcommunitypovertyreductionprogram.Funding forPNPMRuralisnowUS$1.2billion.Asof2009,PNPMRuralcoversall6,408ruralsubdistrictsinIndonesia. 2  displaced persons], antiseparatist front members and communities, and to improve relations between these groups and other villagers and the local state (Terms of Reference, BRAKDP EvaluationandGAMLivelihoodsStudy). Targetingwastobeachievedinthreeways.Firstsubdistrictswithhigherlevelsofconflictwere prioritized(asecondcriterionwasalsoused:agoodperformancerecordinpriorKDPgrants). Second,thesizeofthegrantallocatedtoavillagedependedonestimatesofsubdistrictconflict exposureandvillagepopulationsize(seeTable1.1).Finally,itwashopedthatintheirdecisions regarding how to deploy funds villagers would take special account of the needs of conflict victims. It was originally envisioned that a second round of BRAKDP would follow, providing assistance to remaining villages, but BRA decided to revert to their initial program of directly targetingindividuals.5 TABLE1.1:VILLAGEBLOCKGRANTALLOCATIONS Population Large Medium Small (>=700persons) (300699persons) (<299persons) Confl High 170,000,000 150,000,000 120,000,000 6 ict (US$19,000) (US$17,000) (US$13,000) Inten Medium 120,000,000 100,000,000 80,000,000 sity (US$13,000) (US$11,000) (US$8,900) Low 80,000,000 70,000,000 60,000,000 (US$8,900) (US$7,800) (US$6,700) Morethansimplyofferingfunds,participationintheBRAKDPprogramrequiredvillagestohold aseriesofcommunitymeetingsinwhichvillagersthemselvesdecidedhowthemoneyshould be allocated.7Communities were required to hold at least four meetings, run by trained KDP facilitators.Thefirstmeetingoccurredatthesubdistrictlevelandwasprimarilyinformational, emphasizingthatconflictvictimswereintendedtobetheprimaryrecipientsandoutliningbasic program procedures. Subsequent meetings took place at the village level. In the second meeting,villagersestablishedcriteriaforidentifyingconflictaffectedvillagersandtodetermine who in that village should be considered a conflict victim. Proposals were put forward and voted on by the community at the third meeting. Any villager could submit a proposal for consideration. Proposals had to be for a `project', outlining how funds would be used, by individual households, groups, or (for public goods) the wider community. The fourth `accountability' meeting took place after the funds had been spent, with village officials and facilitators reporting on how the funds had been spent. BRAKDP thus emphasized local ownership,puttingdecisionmakingauthorityandoversightinthehandsofvillagers,including conflictvictims. 5 SeeBarronandBurke(2008)andICG(2007)forbackground. 6 USdollarvaluesarethoseatthetimetheprogramwasimplemented. 7 SeeMorel,WatanabeandWrobel(2009)forafulleroutlineofhowtheprogramworked. 3 1.3 TheSelectionofProjectAreas Everyvillageinselectedsubdistrictsreceivedtheprogram.TheruleemployedbyBRAtoselect subdistricts into the program was complex but still relatively welldefined (see the full description in Box 1). For all rural subdistricts, BRA assigned subdistricts to receive the programonthebasisoftwovariables:conflictintensityandspendingcapacity.Theirgoalwas totreatatargetnumberofthemostconflictaffectedsubdistrictsinadistrict,conditionalona subdistrictsurpassingaspendingcapacitythreshold. When assignment is not done randomly, understanding the selection rule by which communitiesareassignedtoreceiveaprogramisimportantsincedifferenttypesofselection can produce different sorts of bias. This in turn affects how we can interpret differences in outcomesbetweenprojectandcomparisonareas.Selectingthe`neediest'places,forexample, canresultintreatmentareasthatareexanteworseoffthancontrolareas,whichcouldleadto an underestimation of program effects if selection is not accounted for. Selecting the most `capable' places, on the other hand, can result in treatment areas that are ex ante better off thancontrolareas,whichcouldleadtoanoverestimationofprogrameffectsifselectionisnot accounted for. In the case of BRAKDP, one selection variable (subdistrict conflict exposure) reflectsa`need'criterion.Theotherselectionvariable(spendingcapacity)reflectsa`capacity' criterion.Theneteffectofthesetwopotentiallyoffsettingcriteriaonbiasisambiguous. Forthefirstselectionvariable,BRAusedaWorldBankproducedmeasureofconflictintensity, generatedthroughafactoranalysisofnineindicators.Indicatorsincludeddataonthenumber of conflict victims in each of the three years preceding the end of hostilities; the number of reported clashes between GAM and GoI forces; and perceptions of conflict intensity from surveydata.Factoranalysiscreatesacontinuousmeasureofconflictintensity,whichBRAused todividesubdistrictsintolow,mediumandhighconflictintensitygroupssothateachgroup hadapproximatelythesamenumberofsubdistricts.Thisdivisionwasthenusedtoselectthe total number of subdistricts to treat in each district, as well as to rank subdistricts within a districtbytheirlevelofconflictaffectedness.Figure1.1showsthegeographiclocationofsub districtsinAcehbyconflictintensity.Highconflictintensitysubdistrictsclusterprimarilyonthe coast,especiallyneartheoilandgasrichregionsonthenortherncoast,whilethelowintensity regionsareprimarilyinthehighlandareastothesouth. Thesecondassignmentvariable,spendingcapacity,requiredsubdistrictstohavespentatleast 60percentoftheir2005KDPfundsatthetimeoftreatmentassignmentweredeemedeligible toparticipateintheprogram.Thisrenderedineligiblethosesubdistrictsthatwerenotableto handleeffectivelytheinflowoffundsfromBRAKDP. 4  FIGURE1.1:CONFLICTINTENSITYINACEH Source:Conflict&Developmentprogram,WorldBank BRAaimedtousethesetwovariablestoprioritizethemostconflictaffectedsubdistrictsina district(withaminimumofoneperdistrict),conditionalonsubdistrictsmeetingthespending capacity threshold. This was done by using an assignment `rule' that combines the conflict intensityandspendingcapacityvariablesinanonlinearwaytoselectsubdistrictsforBRAKDP. First, BRA selected a target number of subdistricts to treat in each district. That target was determinedbythenumberofhighconflictaffectedsubdistrictsinadistrict.Iftherewereno high conflict affected subdistricts, then the target was set equal to the number of medium conflictaffectedsubdistricts.Iftherewerenohighormediumaffectedsubdistricts,however, the target was set to one low conflictaffected subdistrict. This was in order to meet the requirementthatatleastonesubdistrictbetreatedineachdistrict.Afterthetargetnumber wasselected,subdistrictswererankedineachdistrictbyconflictintensity,fromhightolow. Subdistricts were then selected for treatment up to the target, conditional on their meeting the60percentspendingcriterion.Ifnosubdistrictinadistrictmetthespendingcriterion,then the highest spending capacity subdistrict was selected.8Box 1.1 summarizes the assignment ruleBRAused. 8 ThisonlyoccurredinthedistrictofSimeulue. 5  Box1.1:TheTreatmentAssignmentRule 1. Setatargetnumberofsubdistrictstobetreatedineachdistrictequalthenumberofhigh conflictaffected subdistricts in the district. If there are no high conflict affected, set the target equal the number of medium conflict affected districts; if there are no high or medium,letthetargetequalone. 2. Select the most conflict affected subdistricts in each district up to the target in each district,conditionalonasubdistrictmeetingthe60percentspendingcriterion. 3. If insufficient subdistricts meet the spending criterion, select subdistricts with the best spendingperformance. Using the data originally employed by BRA, we applied this rule to reproduce the treatment assignmentprocess.Thisexercisecorrectlyclassifies207(or92percent)ofall225ruralsub districtseligibleforthisstudy. Theclassificationrateisreassuringlyhigh.Yet,aremainingeightpercentdidnot`comply'with thetreatmentassignmentinthattheyshouldhavereceivedtheprogram(accordingtotherule) anddidnotreceiveit,ortheyshouldnothavereceiveditandtheydid.Inparticular,tensub districtsthatshouldhavebeenassignedtotreatmentdidnotreceiveit,andeightthatshould not have been assigned to treatment did receive it. This noncompliance could result from humanerrorinapplicationoftherule,oritcouldreflectonthespotdecisionsthathadtobe madeduringtherolloutprocess,oritcouldreflectotherelementsoftheassignmentrulethat have not been as clearly codified.9The danger of nonrandom noncompliance is that it can introduce bias in the form of factors that we cannot account for but nonetheless impact the estimationofprogramimpacts.SincewecannotbeconfidentthatnoncomplianceinBRAKDP israndom,wetakestepstoaddressthisissueinouranalysis. 1.4 AnalyticalFrameworkandMethods BRAKDP is hypothesized to impact three families of outcomes: material wellbeing, social cohesion, and trust in government. In this paper, we focus on nine explicit hypotheses about theimpactofBRAKDP.ThesearedetailedinBox1.2. 9 Whileweusethelanguageof`compliance'toatreatment,thereisnoimplicationthatthedecisiontotakepart ornotwasmadebysubdistricts.Indeedthemostlikelyreasonfornoncomplianceisthatfurthercriteriawere usedbyprogramofficerswhenmakingselections.Inparticularfromcorrespondence,weknowthatinsomecases forbudgetaryreasons,someformallynoneligiblesubdistrictsreplacedthosethatmettheselectioncriteriain ordertoensurethatblockgrantsmatchedwiththeoverallprogrambudget.Wedonothoweverhavedetailson thispartoftheassignment. 6  Box1.2:Hypotheses 1Welfare 1.1 SocioeconomicwelfarewillbehigherandwelfareimprovementswillbegreaterinBRA KDPtargetvillagesthaninvillageswithoutBRAKDP. 1.2 In particular, socioeconomic welfare levels and gains among conflict victims will be higherinBRAKDPtargetvillagesthaninvillageswithoutBRAKDP. 2.SocialCohesion 2.1 Social, economic, and political reintegration of exGAM combatants, militias, and IDPs willbegreaterinvillagesinwhichBRAKDPhasestablishedprograms. 2.2 CommunitieswillbelessresentfulofbenefitstargetedatexGAMcombatants,militias, and IDPs in villages in which BRAKDP has worked and more accepting of their participationinthesocial,economic,andpoliticallifeofthevillage. 2.3 Disputes will be less likely to escalate in villages in which BRAKDP has implemented programs. 2.4 Communities will be better able to solve local collective action problems in villages in whichBRAKDPprogramshavebeenestablished. 2.5 AssociationallifewillbemoredevelopedinvillagesinwhichBRAKDPhasoperated. 3.TrustinGovernment 3.1 TrustinthedecisionmakingprocessesofvillagegovernmentswillbehigherinBRAKDP communities. 3.2 Trust in the ability of local government to deliver services/benefits will be greater in villagesinwhichBRAKDPhasbeeninoperation. Source:Programdocuments We use a second best strategy to estimate the causal effects of the programs on these objectives.Anoptimalapproachwouldbetorelyonsomeformofrandomizationinorderto generate comparison (`control') groups that are every way identical to program (`treatment') groups.Becauseoftheneedtoprioritizesubdistrictsintherolloutoftheprogram,BRAKDP did not use a randomized design. This results in treatment areas that are systematically different(moreconflictaffectedandmoreefficientatdisbursingKDP)thancomparisonareas. Nevertheless,theselectionofareasintotheprogramgenerallyfollowedthewelldefinedand transparentprocessjustdescribed.Bysystematicallyaccountingforthisprocess,weareableto produceacontrolgroupassimilartothetreatmentgroupaspossibleandtoestimateprogram effects,eventhoughcomparisonareasexhibitsystematicdifferenceswithcontrolareas. Ourempiricalstrategyhasthreemaincomponents: a) astrategyforidentifyinganappropriatecomparison(control)groupofsubdistrictsex post; b) aprotocolfordatacollection;and c) an estimation strategy to account for systematic differences between the comparison and treatment groups and the fact that not all communities that should have been assignedtotreatmentwereandviceversa. Wedescribethesethreecomponentsnext. 7 Selectingappropriatecomparisonunitsexpost Therewasnosetof`control'unitsidentifiedbeforeprogramimplementation;inaddition,the treatment units were selected according to a deterministic criterion and subdistricts that received the program differ systematically from those that did not. These features make it difficulttoidentifyanappropriatecomparisongroupofsubdistricts.Identifyinganappropriate comparisongroup,however,isessentialifwearetoestimatewhatmighthavehappenedinthe absenceoftheprogramandinthiswaytoascertaintheprogram'simpact. Ourstrategywastoidentifyapoolofsubdistrictsthatweremostsimilartothosethatreceived the program at the time the treatment assignment decision was made. Conceptually, our strategy is similar to determining the `propensity' for assignment to treatment, or the probabilitythatanygivensubdistrictwouldhavereceivedthetreatmentgiventherule.10For example,inBRAKDPhighconflictandhighcapacitysubdistrictsshouldhavehadaveryhigh probability of receiving treatment, whereas low conflict and low capacity subdistricts would not.Wethenselectcomparisonunitsthathavepropensitiessimilartothoseofunitsthatwere infactassignedtotreatmentundertheassignmentrule. Thedifficultyweface,however,isthattheactualassignmentruleproducesbinarypropensities. In other words, under the rule and conditional upon the data, all potential comparison sub districts either were or were not assigned to treatment. Thus on the surface it appears that some subdistricts had 0 percent probability of receiving the program and some had a 100 percentprobabilityofreceivingtheprogram.Toselectcontrolsubdistricts,weneedinsteada continuous measure of propensity. To do this, we draw on the knowledge that, since continuousmeasureswereusedtoassigntreatment,someuntreatedunitsare,infact,`closer' to being treated than others. Our challenge was to translate these notions of proximity into usablenotionsofassignmentprobability. An obvious approach is to generate a propensity score by a logistic or probit regression of actualtreatmentonthetwoassignmentvariables.Whilethisapproachgeneratesacontinuous propensity score, it does not correctly capture the known assignment process. A measure of treatment propensity should model not only the assignment variables but also the known selectionruleinascertainingtheproximityofcontrolstobetreated. Instead we use an approach in which we generate a continuous measure of treatment propensity under the assumption that a deterministic rule is applied to data that is fundamentally`noisy'(i.e.theassignmentvariables--likemostmeasures--havesomerandom error in them). The advantage of this approach is that it allows us to make full use of the assignmentruleanditprovidesanaturalwaytoincorporatecomplexdependenciesbetween unitsintheassignmentprocess(forexampleifvaluesforonesubdistrictdeterminethetarget numberofsubdistrictsinadistrict). 10 RosenbaumandRubin(1983,1984)haveshownthat,ifassignmentismadeonthebasisofobservablevariables andallassignmentvariablesareaccountedfor,selectingtreatmentandcontrolgroupswithsimilarpropensity scorescreatescomparablegroupsforcausalinference.SeealsoMorganandWinship(2007:99). 8  FIGURE1.2:PROPENSITYSCORESFORTREATMENTANDCOMPARISONUNITS Source:Authors'calculations In practice, we simultaneously apply a small independent shock (distributed with mean zero andvarianceequaltohalfthestandarddeviationofthatvariable)toboththeconflictintensity and spending capacity assignment variables in every subdistrict. This slightly changed values forthesevariablesineachsubdistrict.Thenweappliedthetreatmentassignmentruletothis `perturbeddata',whichinturnchangedthesetofsubdistrictsselectedintotreatment.Wedid this10,000timesandtooktheaveragenumberoftimesasubdistrictwasselectedintoBRA KDPof10,000,producingacontinuousmeasureofpropensity.11Inthisway,foreachunit,we estimate the probability that it would have been selected were the underlying data slightly differentfromtherecordeddata. Figure 1.2 shows the actual distribution of treatment and control subdistricts according to continuous propensityscoresgeneratedusingthismethod.Asisclearfromthefigure,there are notable differences between the assignment probabilities for treatment and control sub districts.Asonewouldexpect,thesubdistrictsthatwereactuallytreatedhadhighestimated propensitiesoftreatmentwhilethesubdistrictsthatwerenottreatedhadlowerestimated propensities. Excluded areas had uniformly lower propensities. Nevertheless there was substantial overlap and in particular some nontreated subdistricts had a much higher propensitytobeselectedthanothers,accordingtothemeasure. 11 Toprovideamoreconcreteexample,considerahighconflictaffectedsubdistrictwitha59percentspending capacity.BytheBRA'srule,thatsubdistrictwouldhavebeenineligiblebecauseitdidnotmeetthe60percent spendingcapacity.Nonetheless,thatsubdistrictwas`close'tohavingreceivedtreatment.Saythatafterapplying 10,000smallshockstothatsubdistrict'sscoreonbothvariables,itsconflictintensityscorealmostalwaysremains inthehighcategorybutitsspendingcapacityscorecrossesthe60percentthresholdfortreatment3,500of10,000 times.Thiswouldgiveitapropensityscoreof35percent. 9 These propensities were then used to select 67 (highest propensity) areas to serve as a comparisongrouptothe67treatedsubdistricts.Figure1.3showsthedistributionoftreatment and comparison units over the two main assignment variables, broken down according to whethertheywereselectedintothestudy(rightpanel)ornot(leftpanel)andFigure1.4shows the geographic distribution of treatment and control areas. We see from the figures that control areas are `close' to treated areas both geographically and in terms of the assignment criteria. FIGURE1.3:CHARACTERISTICSOFSUBDISTRICTSSELECTEDINTOSTUDY Not selected for study Selected for study 8 C Conflict Exposure 6 T T T C 4 C C T T T T T T T T T T T T TT T T T T T C C T C T TT C TT T TT TT T T C C T T TT CC C C C C C CT TC 2 C C T T C C T T C C C TT TC C C C CCC C C T T C C C C C C T C CC C C T C T CC C C C C C C C C C C C C CC C C TC T CC C C C C CC C T C C T TC TT C C C C C C CC C C T C C CC T C C C CC T CT C C C CT CC C C C C C C C C C CCC C CC C CC C C C C C C CC CC C T C C C C CC C C C C C T 0 0 50 100 0 50 100 Disbursement record Graphs by select3 Theleftpanelshowsthesetofsubdistrictsnotselectedintothestudyandthevaluesof theseontwooftheassignmentcriteria.Therightpanelshowsthoseselectedinthestudy, markersindicatewhethersubdistrictsweretreated(T)ornot(C). Source:Authors'calculations 10  FIGURE1.4:LOCATIONOFSUBDISTRICTSSELECTEDFORTHESTUDY Source:Conflict&Developmentprogram,WorldBank Datacollection Having identified a sample of 67 BRAKDP and 67 comparison subdistricts, we designed and implemented the Aceh Reintegration Livelihood Surveys (ARLS), a largescale household and villageheadsurveytogathermeasuresthatcouldbeusedtoassesstheimpactoftheprogram. Givenasetofsubdistricts,oneineveryeightvillageswasrandomlyselectedforenumeration. Strata(subdistrictandpopulation)wereusedtoensurebalancetotheextentpossible.Within villages,fivehouseholdswererandomlyselectedtoserveasrespondents.Withinhouseholds, one individual was randomly selected from among all household members aged between 18 and65.Ultimately,thehouseholdsurveywasadministeredto2,315householdsintotal,1,090 of which resided in areas that received BRAKDP. Households were sampled in 67 treatment subdistrictsand68controlsubdistricts,covering17ruraldistrictsand461villagesoverall. The survey instrument asked a set of questions designed to measure outcomes including household and community welfare; individual level behavior and attitudes; communitywide collectiveaction;andperspectivesontheBRAKDPprogramitself.Inaddition,itincludedone behavioralmeasuredesignedtocapturetrustinlocalgovernment.12 A parallel survey of village heads was conducted focusing on communitylevel outcomes includingmaterialwellbeingandcollectiveaction.TheARLSalsosurveyedarandomsampleof excombatants,whichprovidessomedatausedinthisstudy. 12 Thesurveyinstrumentsareavailableatwww.conflictanddevelopment.org 11 Estimationstrategy Our selection of subdistricts into the study helped to ensure that we collected data on the most `similar' comparison subdistricts as well as the treated subdistricts. However simple differencesbetweenoutcomesinthesetwogroupsshouldstillnotbetakenasestimatesofthe causal impact of the BRAKDP program. The reason, as is clear from our earlier discussion, is that even though we selected most similar comparison areas, there are still systematic differences between the treatment areas and the comparison areas that are not due to the impactsoftheprogram,butratherreflecttheassignmentruleoftheprogram. We use two strategies to account for these further differences. As noted above, the vast majority of subdistricts `complied' with their assignment, meaning that those who were selectedtoreceiveBRAKDPdidindeedreceiveit,andthosewhowerenotselectedtoreceive itdidnot.Forthosecasesinwhichthereisthiskindofcompliancetotherule,weareinthe fortunate position in which we know the exact variables that were used for assignment into treatmentandcantakeaccountofthesevariablesinouranalysis.Asdescribedabovethefull assignment rule is complex, but the core substantive criteria make use of only two variables: spendingcapacityandexposuretoconflict.13 Under the twin assumptions that (a) there is a constant treatment effect and (b) treatment assignmentis`unconfounded'conditionalonspendingcapacityandexposuretoconflict(i.e.we have accounted for all variables not related to program impacts that might affect measured outcomes),weestimatetreatmenteffectsbyregressingoutcomesontreatmentalongsidefirst, secondandthirdorderpolynomialsofthesevariablesaswellastheirinteraction.Inessence, thisapproachseekstoaccountforthosefactorsthatdeterminedassignmenttoBRAKDPand thenestimatethecausaleffectofBRAKDPitself,independentoftheassignmentprocess.Both assumptions(a)and(b)arenecessaryforthevalidityofourestimates. There is, however, one further complication. As discussed above, noncompliance presents complications for the analysis by introducing the possibility of unobserved factors that bias outcomes. To account for this, we employ an approach in which we instrument actual treatment (whether a subdistrict participated in BRAKDP) with assignment to treatment (whetherasubdistrictwasassignedtoparticipateinBRAKDP).Inessence,thisenablesusto estimate the effect of BRAKDP for only those areas that `complied' with their treatment assignment and takes into account the potential bias caused by having noncompliers in the study.14Theresultingestimateofthetreatmenteffectis,webelieve,amorereliableestimate ofthecausaleffectoftheprogramthanisthesimpledifferenceinmeans. 13 Indeed,conditionalonoursamplewecancorrectlyclassify87percentofsubdistrictsintoassignment categoriesusinginformationondisbursementandconflictintensityonly;inparticular,byemployingamodelin whichthesetermsraisedtothefirst,secondandthirdpowersareenteredalongwiththeirinteraction. 14 Inaregressiondiscontinuityframeworkthisinstrumentalvariablesapproachcorrespondstoa`fuzzyRD' approach,andinanaveragetreatmenteffectsframeworkitproducesthe`localaveragetreatmenteffect'. 12 In addition to this core strategy we undertake a series of checks of robustness to model specification and report when different approaches yield substantially different outcomes.15 Weillustrateanumberoftheresultsbyshowingtheestimatedintentiontotreateffectsthat result from employing a `regression discontinuity' model in which we create a single running variablethatdeterminesassignmenttotreatmentandestimatetheeffectoftreatmentatthe cutoffpointonthisvariable.Theapproachusedforthesefiguresisdescribedingreaterdetail below. Inthetablesthatfollowthatcomparetreatmentandcontrolgroups,webeginineachcaseby reporting results for the control group (column 1) and the treatment group (column 2). The third column presents the simple difference in means in outcomes across these groups. This difference in means has a very transparent interpretation and reflects well the differences in thesituationsinthedifferentareas.Butitdoesnotnecessarilyprovideagoodestimateofthe causal impacts of the program. The fourth and final column presents the estimated causal effectsaccountingforalltheselectionandnoncomplianceissuesdiscussedabove.Finally,all results reported here take account of the characteristics of our sample (sampling weights, strataandclusters)andclusterstandarderrorsatthesubdistrictlevel,thelevelatwhichthe BRAKDPtreatmentwasassigned. 15 Wefocusonwhatwethinkisthebestestimationstrategygiventhedatastructureexaminedhere.Indifferent robustnesscheckswe:(a)conditiononourestimatedpropensityscores;(b)implementaregressiondiscontinuity designbyconditioningonasingle"runningvariable"thatcombinesthetwokeyvariables;and(c)examinedirect insteadofinstrumentedeffects.Otherapproachestoestimatingcausaleffectsarepossiblehowever.Twostand out.Inone,wecoulduseourestimateofthepropensitytoreceivetreatmentasaconditioningvariableto establish`unconfoundedness'orasbasistocreatematchestoestimateaveragetreatmenteffects.Second,we couldapproximateamorestandardregressiondiscontinuitymodelbyseeking(endogenous)breakpointswithin eachdistrict,andseekingtoestimatethetreatmenteffectatthe(twodimensional)cutoffbetweentreatment andcontrolunits;again,inthiscase,wewouldneedtoinstrumenttoaccountfornoncompliance. 13  2. IMPLEMENTINGBRAKDP 2.1 WhoReceivedAssistancethroughBRAKDP? A key aim of the BRAKDP program was to ensure that BRAKDP funds, while benefiting communitiesingeneral,wereespeciallyeffectiveatreachingconflictvictimsinparticular.The difficultyofachievingthisgoalderivesfromatleastthreechallenges:(a)thelackofconsensus overa clear definition of what constitutes a conflict victim; (b) uncertainty in knowing where conflictvictimsarelocatedandthepracticaldifficultyoftargetingindividualsusingaprogram that, by its nature, works at a higher level of aggregation (BRAKDP is implemented at the villagelevel,buttheprogramisassignedatthesubdistrictlevel,withallvillagesinachosen subdistrictreceivingit);and(c)thefactthattheultimatedecisionsabouthowtotargetfunds restedwithvillagerswhohadfreedomtochoosehowandtowhomfundswouldbeallocated. Weconsidereachofthesethreeaspectsinturn. Whoisaconflictvictim? Havingacleardefinitionofconflictvictimsisrendereddifficultbecauseofthemanydifferent waysanddegreesinwhichindividualscanbeaffectedbyconflict.Theconflictimpactedalmost everyone in Aceh, directly or indirectly. By some accounts everyone was a conflict victim. Moreover,individualperceptionsoftenmatterasmuchaswhatexperiencespeoplemayhave endured. The program opted for subjective assessments of victim status; while guideline categorieswereprovided,itwaslefttocommunitiestodefinewhoisandisnotaconflictvictim. Forthisreason,webaseourcoreanalysesonameasurethatreflectssubjectiveperceptionsof victimstatus.Inaddition,however,wegenerateanobjectivemeasurebasedonselfreported exposure to conflict. In practice these two measures are very closely related, as described below.16 The subjective measure of conflict victim status simply captures whether an individual respondedaffirmativelywhenaskedinthesurveyiftheyconsiderthemselvesaconflictvictim. In order to associate this selfreported status with actual experience of conflict, we also gathereddataonwhyindividuals considerthemselvesvictims.Victimscouldprovidemultiple reasons for victimhood, ranging from death of a family member to internal displacement to mental illness. Using this data we generate a finer category which aims to get at those who weremostaffectedbyconflictamongthosewhodeclarethemselvestobeconflictvictims.We codeanindividualas`mostseverelyconflictaffected'ifeither(a)afamilymemberwaskilledor disappeared,(b)theywerephysicallyinjured,(c)theirhousewasdestroyed,or(d)theywere 16 Thereareanumberofargumentsthatcanbemadeinfavorofusingoneorothermeasure.Thesubjective measurehastheadvantageofcorrespondingmorecloselywiththeprojectapproach;inadditionitcaptures featuressuchasselfreportedmentalillnessesforwhichwehavenoobjectivemeasures.Apotentiallyimportant shortcomingofthesubjectivemeasuresisthatsinceBRAKDPareasweresocializedonwhataconflictvictimis,it ispossiblethatindividualslivinginBRAKDPareasweremorelikelytothinkofthemselvesasvictimsthaninnon BRAKDPareas;thatisthesubjectivemeasureprovidesaposttreatmentmeasureofstratum.However,thefact thatthesetwomeasuresarelargelyconsistentwitheachotherisreassuringonthispoint. 14 displaced by the conflict.17Note that by construction, this `most severely affected' category, thoughtiedto`facts',isnestedwithinthesubjectivecategory. To complement this measure, we also constructed an objective measure that classifies an individualasavictimbasedontheirreportedexposuretoconflict,regardlessofwhetherthey selfidentifyasavictim.Thismeasureexplicitlytakesaccountoftheconflictexperienceofthe individual'shouseholdandnotsimplyoftheirownexperience,usingdataonfamilymember related deaths, disappearances and injuries. We also collect data on homes and workplaces destroyedduetoconflict.Thesubcomponentsofthesubjectiveandobjectivemeasuresrelate broadly to each other although on two items--family member kidnapped and missing body part/physical disability--the two measures record somewhat different ideas 18 and for one item--mentalillness--wehaveasubjectivebutnoobjectivemeasure. Table2.1belowpresentsthemaincategoriesofvictimhood,thelevelatwhicheachvariableis measured, and the correlation between objective and subjective measures of victimhood. Overall,thecorrelationsarestrikinglyhigh,especiallyforthecriteriaformostconflictaffected. Theonlytwoitemsforwhichtherearelowcorrelations(familymemberkidnappedandmissing bodypart/physicaldisability)arethosetwoforwhichthewordingsofthemeasuresdiffer. These high correlations are reassuring and suggest that findings presented here are likely robust to the choice of measure employed. While the main emphasis is on the subjective measurehereandthroughoutthepaper,inthenextsectionwealsoreportboththeobjective measure and a combined measure. The combined measure captures whether either the subjectiveorobjectivecriteriaapply,andcanbeinterpretedastheupperboundonvictimhood. 17 ThisdefinitionofmostconflictaffectedfollowssuggestionsofWorldBankstaff. 18 Formissingbodyparts,thesubjectivemeasureasksrespondentsiftheyhave"missingbodypartsorpermanent physicaldisabilityduetoconflict"whereastheobjectivemeasureasksiftheywere"injuredormaimed(resulting inhospitalizationorinabilitytofunctionnormallyforatleastonemonth)asaresultoftheconflict."Thesubjective measureforkidnappedasksindividualsifa"familymemberdisappeared/waskidnapped/ordetaineddueto conflict".Theobjectivemeasureasks(onlyformembersofthe1998household)whyanindividualinthemain respondent's1998householdisnolongerinthe2008household.Householdmemberswho"disappeared/were takenawaybecauseoftheconflict"werecodedfortheobjectivecriteria. 15  TABLE2.1:CORRELATIONBETWEENSUBJECTIVEANDOBJECTIVEMEASURESOFVICTIMHOOD Levelatwhichsubjective Levelatwhich Correlation itemdefined objectiveitemis between defined measures Nonvictims .67 Conflictvictims(includingmostaffected) .67 Mostconflictaffected .86 Familymemberkilled Household Household .92 Familymemberkidnapped/detained Household Household .12 Missingbodypart/physicaldisability Individual Individual .23 Housedamaged/destroyed Household Household .78 Primarylivelihooddamaged/destroyed Household Household .61 Internallydisplaced Individual Household .93 Mentalillness(selforfamilymember) Individualorhousehold Physicalillness(selforfamilymember) Individualorhousehold Household .29 Source:ARLS AreconflictvictimsconcentratedintheareasthatreceivedBRAKDP? Table 2.2 explores how well BRAKDP reached conflict victims. Focusing on the subjective measure,weseethat42percentofallindividualsinthestudypopulationareconflictvictims; of these 28 percent meet the criteria for mostconflictaffected. (If the combined measure is employed, 45 percent are victims and 30 percent of the population are `mostaffected'). We estimatethatbetween617,000and657,000peopleinareasthatreceivedBRAKDParevictims. The greatest share of conflict victims were internallydisplaced, followed by those suffering frommentalillnesses. 16  TABLE2.2:CONFLICTVICTIMPRIORITIZATION Panel I II III IV Shareofstudy ShareofpopulationinShareof...thatareinEstimatednumber populationthatare... treatmentareasthat treatmentareas(%) of...intreatment (%) are...(%) areas('000) Sub Ob Comb Sub Ob Comb Sub Ob Comb Sub Ob Comb Nonvictims 58 68 55 51 62 48 41 42 40 643 780 603 Conflictvictims 42 32 45 49 38 52 54 55 54 617 480 657 Mostconflictaffected 28 28 30 34 33 37 57 56 56 429 420 462 Shareamongconflict Shareamongconflict Ofall...thatarein Estimatednumber victimsinstudy victimsintreatment treatmentareas(%) ofthosewith...in populationwith...(%) areaswith...(%) treatmentareas ('000) Sub Ob Comb Sub Ob Comb Sub Ob Comb Sub Ob Comb Familymemberkilled 4 7 5 5 8 6 70 66 66 30 38 38 Familymemberdetained 6 1 6 7 1 7 63 56 62 43 5 48 Physicaldisability 4 8 8 5 10 9 68 62 63 28 46 61 Housedamaged/destroyed 19 33 24 19 34 26 55 56 56 117 161 168 Primarylivelihooddamaged 19 18 21 18 20 22 50 61 56 110 95 144 Internallydisplaced 50 65 49 54 68 52 58 56 56 335 337 346 Mentalillness(selforfamily) 32 32 28 28 47 47 175 175 Physicalillness(selforfamily) 19 21 28 17 22 27 45 56 52 102 105 177 Tablereportsestimatedpopulationmeans. Source:ARLS Withintheareasthatreceivedtheprogram,49percentofthepopulationconsiderthemselves conflictvictims;incontrolareas,36percentdoso(notshown).Forthemostconflictaffected, 22 percent meet the criterion in comparison areas, compared to 34 percent in areas that receivedtheprogram.Thesedifferencesarebothsignificantatthe99percentlevel. DifferencesinthedistributionofconflictvictimsatthevillagelevelareillustratedinFigure2.1; thefigureshowsthedistributionoftheshareofthefiverespondentsineachvillagethatwere conflictvictims.Whilethedistributionisskewedtotheleftforcomparisonareasweseethatit is relatively uniform in project areas; there are almost as many villages for example with no victims reporting as there are with all five reporting. The differences between these two distributions is significant: There were approximately twice as many villages in comparison areasthathadzerooffiverespondentsreportingasconflictvictimsandabouttwiceasmany villagesinprojectareaswithallfiverespondingpositivelycomparedtocomparisonareas. 17  FIGURE2.1:NUMBEROFCONFLICTVICTIMSINPROJECTANDCOMPARISON VILLAGES BRA-KDP control group BRA-KDP treatment group 60 Number of villages 40 20 0 0 1 2 3 4 5 0 1 2 3 4 5 Number of conflict victim respondents per village (maximum=5) Note:Figureshowsthedistributionoftheshareofthe5respondentsineachvillagethatwere conflictvictims.Forexample,0of5respondentsclassifiedasconflictvictimsinaboutoneinfour comparisonareavillages,butsuchlowreportswereonlyseeninoneineightprojectarea villages. Source:ARLS Nevertheless,itisclearaswellthattheoverallabilityoftheprogramtoreachconflictvictims was limited, in part because the program was not continued into its planned second round. Althoughconflictcriteriawereusedtoselectprogramareas,amongconflictvictims,onlyabare majority,54percent(seeTable2.2,PanelIII)ofallthoseinthestudyarea,livedinsubdistricts that received BRAKDP; the corresponding numbers for nonconflict victims and the most conflictaffectedare41percentand57percent,respectively.Thesenumbersaremuchhigher fordeathsandphysicaldisability(70percentand68percentoftheseareintreatmentareas) although for some categories (mental and physical illness) the estimated shares are actually higherinthegeneralpopulation. Thissuggestsrealdifficultiesintargetingconflictaffectedindividualsthroughprogramsthatare administered atmoreaggregatelevels(i.e.subdistrict). Thedifficulty reflectsseveralfactors. The conflict in the Aceh was widespread, which will make any effort of this form imperfect. Furthermore, the correlation between the share of victims and the measure of conflict exposureavailabletoBRAinassigningtheprogramisstatisticallystrongbutsubstantivelyweak (0.36).Inotherwords,someareaswithmanyconflictvictimsscoredquitelowonthemeasure ofconflictintensity(examplesincludeBendaharaandSyiahUtama)Italsoreflectsthetradeoff inherentinassigningtreatmentonthebasisofspendingcapacity:someareasthatscoredhigh on conflict measures were not treated because they fell short on the disbursement criterion (PeureulakTimurisonesuchcase).Ofcourse,thesewouldnothavebeenissuesiftheprogram hadbeenextendedintoallremainingsubdistrictsinitssecondyear,aswasoriginallyplanned. 18  Finally, we note that the broad patterns we describe above hold for both the objective and subjectivemeasuresofconflictaffectedness. Doconflictvictimsbenefitmorethanothersintreatmentareas? Conflict victims can benefit from BRAKDP through both private transfers and public investments. As we will see in the next section, the large majority of BRAKDP funding went towards direct private transfers (for example, to support small businesses); moreover, the scope for targeting victims is likely to be greater for private transfers compared to public investments.Inthissection,wefocusonthesetransfersandexaminetheextenttowhichthese weresuccessfullytargetedatconflictvictims. The ARLS data asks respondents whether they benefited directly from BRAKDP.19As seen in Table2.3,thesetransfersreachedwidesegmentsofthepopulation:about530,000individuals live in households that benefited from these transfers, of whom approximately 308,000 are adultsaged1865(confidenceintervalfortheadultsbenefitingdirectlyis230,000