Because political parties operate over large areas and party agents may relocate away from observed registration centers observers may displace rather than deter irregularities We design and implement a largescale twolevel randomized eld experiment ID: 78699
Download Pdf The PPT/PDF document "Deterring or Displacing Electoral Irregu..." 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.
TheJournalofPolitics,Vol.74,No.1,January2012,Pp.292Ð307doi:10.1017/S0022381611001368SouthernPoliticalScienceAssociation,2012ISSN0022-3816 partialdemocracies,Ghanaisarapidlyurbanizingdevelopingcountrywithalarge,poor,ruralpopula-tion,whereresourcesareconcentratedinthestateÑconditionsthatarefrequentlyassociatedwithvotebuyingandotherelectoralmalpractices(KitscheltandWilkinson2007;Stokes2005).Furthermore,GhanahassigniÞcantexperiencewithbothinternationalanddomesticelectionob-serverswho,asinothercountries,havehistoricallyfocusedonelectiondayactivitiesandmayhavepushedmalpracticesbypoliticalpartiestothepreelectionstage.Indeed,thereisprimafaciedenceofinßationofthevotersregister.In2008,theElectoralCommissionexpectedtoregister800,000voters,theestimatednumberofcitizenswhohadattainedvotingagesincethepreviousregistration.However,theElectoralCommissionregisterednearly2millionnewvoters,aÞguresigniÞcantlygreaterthanthevotemargininthepreviouspresidentialelectionandforaprovisionaltotalnumberofregis-teredvotersgreaterthantheestimatedadultpopula-tionofGhana.Somebutnotalloftheseunexpected1.2millionregistrationswerepeoplemistakenlyre-registeringinsteadofrequestingareplacementforalostvoterIDcard.However,inseveralconstituencies,thetwomainpoliticalpartiestradedaccusationsoftransportingsupporterstohavethemillegallyregistertovote(Boateng2008).Ourcontributionisthreefold.Firstandmostimmediately,weshiftthefocusinthisliteratureonelectoralmisconducttothestage.WedemonstrateandquantifysizableirregularitiesinvoterregistrationinacountrythatisconsideredaÔÔmodelÕÕnewdemocracy16yearsremovedfromitstransitionfromautocraticrule.Toourknowledge,ourworkistheÞrstlarge-scaleexperimentalstudytoexaminepreelectionirregularitiesandtoworkwithdomesticelectionobservers.ItcomplementsrelatedworkbyHyde(2007)withelectiondayinternationalmonitorsandtheextensivequalitativereportingonelectionfraudbydomesticandinternationalelectionobserverssupportedbyorganizationssuchastheCarterCenter,NationalDemocraticInstitute,andtheEuropeanUnion.Ourworkalsoaddstotheliter-atureonstatisticalmethodstodetectproblemswithelectionresults(Mebane2006;Myagkov,Ordeshook,andShakin2009).Second,weconsiderexplicitlyhowpoliticalpartiesareorganizationsthatcoverawidegeograph-icalareaandcreateconnectionsacrosspoliticalunits,andwestudytheimplicationthatinterventionsonillicitpoliticalactivitieswillhavespillovereffects.WeÞndevidenceforsuchinterferenceacrossspatialunits,butalsoconcludethatthespilloversarelimitedbygeographicaldistance.Ourworkaddstothehandfulofrecentstudiesinpoliticalsciencethathaveexplicitlyconsideredspillovers,suchasNickersonÕs(2008)experimentalstudyofpeereffectsonturnoutinanAmericanelection.Third,thisresearchcontributestoagrowingbodyofscholarshipthatusesrandomizedÞeldexperimentstostudyhowdemocracyworksinpracticeinspeciÞcdevelopingcountriesandclarifydebatesandgeneralizationsfromobservationalstud-ies(CollierandVicente2010;Humphreys,Masters,andSandbu2006;Olken2010;Wantchekon2003).OurexperimentalstudyreinforcesBirch(2007)ÕsconcernsabouttherobustnessofÞndingsfromempiricalstudiesthatmeasurefraudusingobserverreports.Theseincludemostobviouslystud-iesonthecausesandconsequencesofelectoralfraud,butalsothoseontherelationshipbetweenelectionqualityanddemocraticdevelopment(Lindberg2006).OurworkalsosuggestsgreatercircumspectionabouttheefÞcacyofobserversinreducingelectoralirregularities(Hyde2007;Kelley2012)andpointstofutureresearchonhowfraudisorganizedandontherelationshipbetweenpreelectionfraud,electiondayproblems,andpostelectionrigging.WeproceedbyÞrstpresentingourhypothesesonobservers,partyagents,andvoterregistration.Wedescribethevoterregistrationprocessandthe2008generalelectionsinGhana,thenpresenttheexper-imentaldesign,data,andanalysis.ObserversandPartyAgentsWhatistheeffectofdomesticobserversonthebehaviorofpoliticalpartyagentsandtheextentofirregularitiesinvoterregistration?Thebasicpremiseisthatpoliticalpartieswanttowintheelectionandwanttoappeartodosocleanly.Ifthepartiesintendtoinßatethevotetotalintheirfavoronelectionday,theymaywishtoinßatetheregisterwiththeirsupport-ersandatthesametimedonotwanttheiragentstobecaughtdoingso.Therefore,observersshould,infact,deterpartyagentsfromorganizinglogisticsforfrau-dulentregistrationsattheregistrationcentersthattheTheNationalDemocraticInstitutehasmonitoredmorethan270elections(http://www.ndi.org/content/elections),andtheCarterCenterandEUhavebothobservedmorethan60elections(http://cartercenter.org/peace/democracy/observed.html;http://ec.europa.eu/external_relations/human_rightshtm).NAHOMIICHINOANDMATTHIASSCHU withregionalandethnicbases(LindbergandMorrison2005;Nugent2001).InordertovoteinGhana,avotermustgototheparticularpollingstationassociatedwithhisresidenceandpresenthisvoterIDcardwhichshouldhaveaphotographtakenatthetimeofregistrationifacamerawasavailable;nootherformofidentiÞcationisrequired.TheelectoralofÞcialcomparesthiscardwiththeinformationandphotographprintedonthevotersregisterbeforeallowingthepersontovote.WhileinternationalbordersareclosedaroundelectiondaytopreventTogolese,Burkinabe,andotherforeignersfromenteringthecountrytovote,internalroadsareleftopen,andsomeonewhowishestovoteinmultiplelocationsonelectiondaycouldeasilydosoaslongasheisregisteredatthosemultiplelocations.CitizensofGhanamayregistertovoteonlyduringdesignatedregistrationperiods,onlyinper-son,andonlyatparticularregistrationcentersasso-ciatedwiththepollingstationandelectoralareafortheirresidence.Someonewhowishestoregistertovotemustdeclarehisorhername,address,parentsÕnames,andhomearea,andtheelectoralofÞcialwillÞlloutthisinformationonaregistrationform.Theregistrantisphotographedifacameraisavailable,andthephotographisattachedtotheform,coveredbyastickyplasticsleeveandbecomestheofÞcialvoterIDcard.LiketheUnitedStatesandseveralotherformerBritishcolonies,GhanadoesnothaveanationalIDcardsystemandelectoralofÞcialshavenomeanstocheckaregistrantÕsidentity,sothatitisfairlyeasytodeclarefalseinformation.ElectoralofÞcialsmayremindthepersonregisteringthatthepenaltyforgivingfalseinformationorregisteringmultipletimesisuptoayearinprison,butalmostnooneiseverprosecutedforfalseregistration.Voterregistrationwasdelayedseveraltimesduetoacontroversyaroundasummaryofthe2006votersregister,aswellasduetosupposeddelaysinprocuringequipment,releaseoffundsfromthegovernment,andhiringofqualiÞedtemporarystaff.VoterregistrationÞnallybeganon31July2008,withonlyonedayÕsadvancenotice.Althougheachoftheapproximately4,800electoralareasinthecountrywasexpectedtohavearegistrationworkstation,onlyabout2,500workstationswereavailable.TheregionalElectoralCommissionersdistributedequipmentandregistrationmaterialsateachregionalheadquarterstodistrict-levelElectoralCommissionofÞcials,whotransportedthemanddistributedthematthedistrictofÞcestotemporarystaffhiredbythedistrictofÞce.Thedistrict-levelElectoralCommissionofÞcialsthendrewuptheplansforwhichelectoralareaswouldhaveregistrationcentersonwhichdays.Conse-quently,noadvanceinformationwasavailablecen-trallyonwherethemobileregistrationworkstationswouldbelocatedonwhichdates.Inatleastoneregion,thedistributionofequipmentandmaterialsamongthenumerousdistrictswashaphazardanddidnotfollowanyformulathatconsideredthesizeoftheAsinpreviousvoterregistrationexercises,thepoliticalpartiesactivelyferriedpeopletoregistra-tioncenters.Onthelastdayofthescheduled11-dayregistra-tionperiod,theElectoralCommissionextendedregistrationbytwodaysduetowidespreadreportsofshortagesofmaterialsandequipment.TheElec-toralCommissionofGhanathenprocessedalltheregistrationformsatitsheadquartersinAccraandproducedaprovisionalvotersregister.Bylaw,thisprovisionalregistermustbemadeavailableduringanexhibitionperiod,duringwhichanofÞcialfromtheElectoralCommissionsitswiththeprovisionalvotersregisteratparticularlocations(usuallyonecentrallocationineachelectoralarea)sothatvoterscancheckfortheirnames.ObjectionstoanynamesontheregistermaybelodgedwiththeElectoralCom-missionatthistime.Approximately0.4%ofnewregistrationswerechallengedin2008,whichis10timestherateofchallengesagainstnewregistrationsin2004,andthisprovisionallistwascutdowntoaÞnallistofapproximately12.5millionvoters.InGhana,thevotersregisterisvettedfordeceasedvotersorotherswhoshouldnotbeontheregisteronlyduringthisperiodbetweenproductionoftheprovisionalandÞnalvotersregisters.TheElectoralCommissiondescribesthisprocessasÔÔlamina-tionÕÕbuttheplasticsleeveisnotheatedormelted;itisonlyself-AprominentexceptionisthecaseofPiusOpokuBoateng,whocameunderheightenedscrutinyastheNDCparliamentarycandidateforKwabreWestconstituencyandwassentencedto12monthsinprisonfordoubleregistration(Alhassan2008).MeetingwithDeputyElectoralCommissionerDavidKangah,inAccra,Ghana,July2008.ObservationbyresearchassistantatregionalElectoralCommis-sionheadquarters,30July2008.AnNDCagentandataxidriverindependentlyreportedtoadomesticregistrationobserverinTrobu-AmasamanconstituencyinGreaterAccraRegionthat,priortotheobserverÕsarrival,NPPpick-uptrucksconveyedpeoplefromnearbyvillagestotheregistrationcenter.SimilarlyinNingo-PrampramconstituencyinGreaterAccraRegion,adomesticregistrationobserverreportedthatbothNDCandNPPwerebussingpeopletoregistrationcenters.NAHOMIICHINOANDMATTHIASSCHU registrationworkstationsthanelectoralareasandthatsomeelectoralareaswouldshareworkstations,weconductedourrandomizationoverthelistofelec-toralareasfromthe2006electionbecausethelocationoftheregistrationworkstationsweretobedeterminedbylocalElectoralCommissionofÞcialsandunavailableaheadoftime.OurrandomizationprocedureclassiÞeselectoralareasintooneofthreegroups:controlelectoralareasin26controlconstituencies,controlelectoralareasin13treatmentconstituencies,andtreatmentelectoralareasinthose13treatmentconstituencies.IntheestimationwetakeintoaccountthisdesignthroughtheinclusionofthefullsetofblockÞxedeffectsandcorrectionofstandarderrorsforclusteringattheconstituencylevel(BruhnandMcKenzie2009;Dußo,Glennerster,andKremer2007).RegistrationobserverswererecruitedfromCODEOmemberorganizationsthatwouldordinarilyÞeldregistrationobservers,weretrainedtogetherbyoneoftheauthorsandCODEOleaders,accreditedbytheElectoralCommissionasregularobservers,anddeployedatthesametime.Theregistrationobserverswereinstructedtogototheirassignedconstituencies,Þndoutwhereandwhentheregistra-tioncenterswouldbeopen,andthenvisitunan-nouncedonlyregistrationworkstationsintheelectoralareasontheirlist.Theywereinstructedtovisitalltheelectoralareasontheirlistatthebeginningoftheregistrationperiodbeforerevisitingtheregistrationcentersonce(mostlyinruralareas)ortwice(mostlyinurbanareas)moreduringtheregistrationperiod.Registrationobserversstayedateachregistrationcenterforabout1Ð2hoursoneachÞrstvisitanduptoafulldayinthelatervisits.InGhana,observersarenotpermittedtoassistorinterferewiththeproceed-ings,althoughtheymayinteractwiththepartyagentsandelectoralofÞcialsattheregistrationcenter.TheregistrationobserverswereaskedtoÞlloutaone-pagechecklistwithquestionssuchaswhethertheregistrationcenterwasopenupontheobserverÕsarrival,whetherithadaworkstation,whetheritwaswell-markedandeasytoreach,whethertherehadbeenanyviolence,whethertheregistrationcenterhadbeenforcedtocloseatanytime,whetheranypeopletheobserverthoughtwereineligible(e.g.,underage)hadbeenregistered,andwhetheranypeopletheobserverthoughtwereeligiblehadnotbeenpermittedtoregister.RegistrationobserversweredirectedtofaxbackthesecheckliststotheCODEOsecretariateverycoupleofdaysorwhenevertheywereinanurbanareaandhadaccesstotelephones/faxlines.CODEOofÞcialsreadthesereportsandreleasedonegeneralpressstatementduringtheregistrationperiodandoneatitsWeuseacombinationofdatafromourexperimentandofÞcialsourcesforouranalysis.WegainedaccesstotheofÞcialnumberofregisteredvotersatthepollingstationlevelin2004and2008andcompiledtheseintoelectoralarea-levelÞgures.WeusewhetheranobserverÞledareportfortheregistrationcenterforaparticularelectoralareaasourmeasureofwhetherthatelectoralareawasvisitedbyaregistrationobserver.WedigitizedtheElectoralCommissionÕsmapofconstituencyboundaries.Wealsogeocodedthe868electoralareasinourexperimentaspointsbycomparingthenamesofthepollingstationslocatedwithinthoseelectoralareaswithpubliclyavailableprintedanddigitalmaps,datafromthe2000populationandhousingcensus,coordinatesfromGPSwedeployedwithsomeoftheregistrationobservers,andonoccasion,informationgivenbylocalelectoralofÞcials(Figure1).WeusedArcGIStocalculatethedistancebetweenallpairsofelectoralareasandconstructvariablesthatindicatethenum-berofelectoralareasina5kmradiusinthesameconstituency,numberofelectoralareasina10kmradiusinthesameconstituency,distancetonearestelectoralareainthesameconstituency(km),numberofelectoralareasassignedregistrationobserversina5kmradiusinthesameconstituency,numberofelectoralareasassignedregistrationobserversina10kmradiusinthesameconstituency,anddistancetonearestelectoralareaassignedaregistrationobserver(km).ThesevariablesaresummarizedinTable1.WealsousedthesegeocodedelectoralareastomakesmalladjustmentstotheconstituencyboundariessothatelectoralareasidentiÞedandlocatedbymatchingpollingstationnameswithothersourcesactuallyfellwithintheboundaries.NeitherpopulationÞguresnorpreviouselectionsresultswereavailableatthislevelofdisaggregation.Forpreviouselections,CODEOtrainedtheirregistrationob-serverstoaddresstheseissuesinaone-to-two-pagewrittenreportratherthanonapreprintedchecklistwithspacefordescriptionsofanyincidents.Theseobserverswerefreetoselectwhichelectoralareastovisit.WecannotuseCODEOdatatoinvestigateeffectsofthetreatment,becausethesehavenoinformationforthecontrolelectoralareas.Replicationdataareavailableat:http://dvn.iq.harvard.edu/dvn/NAHOMIICHINOANDMATTHIASSCHU Toinvestigatethefullmodel,includingthelocalizedspillovereffectsofobservers,weestimatetheisthepercentagechangeinthenumberofregisteredvotersfrom2004to2008inelectoralareainconstituencyisanindicatorforwhetheranobserverwasrandomlyassignedtoelectoralareainconstituencyduringtheregistrationperiodinAugust2008,isanindicatorforwhetheranobserverwasassignedtoanyoftheelectoralareasinconstituencyduringregistration,andblockÞxedeffects.Ourvariableforcapturinglocal-izedspilloversis,thenumberofelectoralareasinassignedtotreatmentwithindistanceofelectoralareainconstituencyisthetotalnumberofelectoralareaswithindistanceelectoralareainconstituencyAsnotedearlier,thestableunittreatmentvalueassumption(SUTVA)isviolatedinaworldwithspillovers,andignoringthesespilloverswillresultinbiasedestimatesoftheprimaryeffect.SinceanelectoralareathathasneighboringELAsthataretreatedisalsoalwaysinatreatmentconstituency,theneighborhoodtreatmentvariables()arepos-itivelycorrelatedwiththeindicatorvariableforthetreatmentconstituency().Thereforeiftheeffectoftheneighborhoodtreatmentvariables(isnonzeroandtheyareomittedfromthemodel,theisendogenous.Ifweassumethatthereisnointeractioneffectbetweenandtheneighbor-hoodtreatmentvariablessothatthecoefÞcientsarezero,thenwecandeterminethedirectionofthebiasresultingfromtheomission.Sincethecorrelationbetweentheneighborhoodtreatmentvariablesispositive,thesignwillbedeterminedbythetruecoefÞcientsoftheneighborhoodtreatmentvariables.Underourhy-pothesisofpositivelocalizedspillovereffects,therewillbeanupwardbias,andifthetrueprimaryeffectisnegative,aspeciÞcationthatexcludestheneigh-borhoodtreatmentvariables()willleadtoacoefÞcientthatisclosertozero(andpotentiallyinsigniÞcant)becauseofthisupwardbias.InthemorecomplexworldinwhichthereisfeedbackamongtreatmentELAs,sothatthecoefÞcientsonarenotzero,theneighborhoodtreatmentvariablesarestillcorrelatedwiththetreatmentELAevenaftertheeffectoftheconstituency-leveltreatmenthasbeenpartialledout.Establish-ingthedirectionofthebiasinthiscaseismoredifÞcultasitwouldrequireknowledgeofthecova-riancesofthevariableswitheachother.Ineithercase,excludingtheneighborhoodvariablesfromthemodelspeciÞcationwillleadtobiasedestimatesbecauseoftheSUTVAviolation.Whileourrandomizationprocedureguaranteesthatallelectoralareaswithinaconstituencyhaveanequalprobabilityofbeingselectedfortreatment,electoralareasdonotallhavethesameprobabilitiesexanteofbeingassignedaparticularnumberoftreatedneighborsbecausesomeelectoralareasaremorecentrallylocatedthanothers.Inpractice,treat-mentandcontrolelectoralareaswithintreatmentconstituencieshavethesamenumberoftreatedelectoralareasintheneighborhoodonaverage(Table1),buttherandomizationproceduredoesnotguaranteethatthedensityoftreatmentintheneighborhoodofanelectoralareawillbeuncorre-latedwithothercharacteristicsofelectoralareas,suchaspopulationdensity,distancetoroads,andotherlocalcharacteristics,thatmayaffectvoterregistra-tion.Therefore,ourmodelalsoincludes,thetotalnumberofelectoralareasinconstituencyofelectoralareainconstituency,whichwillcapturealltheseaspectsthatareunrelatedtotreatment,forwhichmightproxy.Asnotedearlier,electoralareasaregeographicallycodedaspoints,andbotharecomputedascountsofpointsthatfallwithinaparticulardistanceofthegivenpoint.Ifregistrationobserversdeterregistrationintheelectoralareastheyvisitbutthesedeterredregistra-tionsaredisplacedtonearbyelectoralareas,then0and0.Weuse0005todenoteelectoralareaswithina5kmradiusand0510todenoteelectoralareaslocatedbetween5kmand10kmfromaparticularelectoralarea.WealsoaddinteractiontermsbetweenthetreatmentindicatorandbetweeninsomeTable3presentsourresults,withColumn3reportingresultsforthemainOLSspeciÞcation.AllowingtheeffectofaregistrationobserverinanearbyelectoralareatovarywithtreatmentstatusanddeÞningÔÔnearbyÕÕasa5kmradiusoftheelectoralarea,weÞndthatwithintreatmentconstituencies,electoralareaswithregistrationobservershaveandeterringordisplacingelectoralirregularities oftheseregistrationobserversattheconstituencylevel.ThecoefÞcientsontheconstituency-levelandelectoralarea-leveltreatmentindicatorshavethesamesign,andtheestimateoftheformer()is0.041,witha-valueof0.086intheOLSspeciÞca-tions.Theseresultsimplythatanelectoralareaassignedaregistrationobserver,butwithnoelectoralareasassignedregistrationobserversina5kmradius0;henceforthweomittheforsimplicity),hasonaverageanapproximately7.6percentagepointsmallerincreaseinregistrationthananelectoralareainacontrolconstituency(0).Thisaveragedifferenceshrinksto5.9percentagepointsifanelectoralareainthe5kmneighborhoodisassignedaregistrationobserver(1)andto1.4percentagepointsforanelectoralareawithoutaregistrationobserverbutwithanelectoralareainthe5kmneighborhoodassignedaregistrationobserver00051).ForthespeciÞcationinColumn3,werejectthenullhypothesisofnotreat-menteffect,0,inatwo-tailedtestwith8.20anda-valueoflessthan0.0001.Wealsorejectthenullhypothesisofnoprimaryorgeneralspillovereffect,0,inatwo-tailedtestwith4.82anda-valueof0.014.Weuseourrandomizedassignmenttotreatmentasaninstrumentforwhetheranelectoralareawasvisitedbyaregistrationobserver()inColumns4Ð6ofTable3.EarlyAugustistheendoftherainyseasoninsouthernGhanaandstillthemiddleoftherainyseasoninnorthernGhana,andtheregistrationobserversnoteddifÞcultytravelingonmanyruralroadsandcrossingrivers.Moreover,therewassomeconfusionsurroundingthescheduleofwhichregis-trationcenterswouldbeopenonagivendate.However,compliancewasgenerallyverygood(Table4)andconsequentlythereareonlysmalldifferencesbetweenourOLSandIVestimates.Wecheckforrobustnessbyestimatingequation(1)atotherdistances(4and8km,6and12km),includingthelognumberofregisteredvotersin2004asacontrolandusingthelognumberofregisteredvotersin2008astheoutcomewiththe2004Þguresasacontrol,andtheresultsremainsubstantivelythesame.Ingeneral,theestimatedprimaryITTeffectofregistrationobservers)isabout-4%andtheestimatedlocalizedspillovereffect()isgreateratshorterradii,whichisconsistentwithadisplacementofpotentialregistrantsawayfromanobservedregistrationcentertocloseralternativeregistrationcenters.TheresultsalsodonotchangesubstantivelywhenweincludeanadditionalÔÔringÕÕ(1020,forexample)toconsidertheeffectofregistrationobserversinelectoralareasfurtheraway.Wealsoestimateequation(1)withoutbutwithconstituencyÞxedeffectsinsteadofblockÞxedeffects,andtheresultsfortheprimaryandspillovereffectsofobserversremainsubstantivelythesame(notshown).Asanadditionalrobustnesscheck,weusethedistancetothenearestneighboringtreatmentelectoralareainsteadofthenumberoftreatmentelectoralareasinacertainradius.Weregressthepercentagechangeinregistrationfrom2004to2008ontheconstituencyandelectoralarealeveltreatmentindicators,theinverseofthedistancetothenearestneighboringtreatmentelectoralarea(),itsinteractionwiththeconstituency-leveltreatmentindicatorvariable,andthefullsetofblockÞxedeffects:Weexpecttheeffecttobelargerthecloserthenearesttreatmentelectoralarea,sincedistanceraisesthecostofvisitinganotherregistrationcenter.Weusethistransformationofthedistancemeasuretoallowforthiseffecttodiminishmorerapidlyatcloserdis-tancesthanatgreaterdistances.Wealsorestrictthesampletoelectoralareaswhosenearestneighbortreatedliesatlessthanthemaximumdistanceforelectoralareasintreatmentconstituencies(54km)forthisanalysis.Withthissetup,theestimatedcoefÞcientontheinverseofthedistancetothenearestneighboringtreatmentelectoralareaforelectoralareasintreatmentconstituenciesis042witha-valueof0.028,whilethecoefÞcientondistanceforelectoralareasincontrolconstituencies(148)isstatisticallyindistinguishablefromzerowitha-valueof0.490.Theestimateforis-0.10(withastandarderrorof0.012)andthecoefÞcientontheconstituency-leveltreatmentindicator()isestimatedtobe-0.016(withastandarderrorof0.018).TheestimatedcoefÞcientontheconstituency-leveltreatmentindica-tor()issmallerthanourestimatesforequation(1)reportedinTable3becauseiteffectivelyincorporatesWealsouserandomizationinference(Fisher1935;Rosenbaum2002)totesttheexactnullofnotreatmenteffect,0,fortheintent-to-treatanalysisusingradiiof5and10km.Usingthe-statisticfromtheactualexperimentandtherandomizationprocedurefromtheexperimenttogeneratethenulldistributionbasedon10,000randomizations,werejectthenullinatwo-tailedtestwithanexact-valueofdeterringordisplacingelectoralirregularities wereexhibitionperiodobservers,sothatnoregistra-tionobserverswereinvolvedinevaluatingtheirownTheseexhibitionperiodobserversconductedasurveyofelectoralofÞcersandanypartyagentspresentaboutvoterregistrationinthatareaandaskedforprovisionalregistrationnumbers.TheseobserverscompletedsurveysofelectoralofÞcersin304electoralareas(ofwhichsixaremissinginfor-mationneededtoidentifytheconstituencyorelectoralarea).Unfortunately,thedistrict-levelElec-toralCommissionofÞcershirealargenumberoftemporarystaffforthesenational-scaleexercises,andmanyoftheofÞcialspostedduringtheexhibitionperiodwerenotthesameasthosepostedduringtheregistrationexercise.AlthoughofÞciallyaregistrantdoesnotneedtopresentIDinordertoberegistered,approximately60%ofexhibition-periodelectoralofÞcialswhoreportedthattherewereobserversattheregistrationcenterintheirelectoralareainAugustaskedforID.Anearlyidenticalproportion(56%)oftheseelectoralofÞcialswhorespondedthatnoobserversvisitedduringregistrationalsorespondedthatidentiÞcationwasrequestedofregistrants.LocalelectoralofÞcialsfrequentlyexpoundedupontheircreativesolutionstoshortagesofregistrationformsandmalfunctioningregistrationequipmentinordertoaccommodatetheunexpectedlylargenumberofpeoplewhoturnedouttoregister.RegistrationobserversthereforelikelyhadverylittleeffectonthebehaviorofelectoralofÞcialsinwaysthatdepressedregistration.Aswiththecitizensmechanism,itisalsodifÞculttoimaginethatofÞciousnesswouldaccountforsuchalargepositivelocalizedspillovereffect.RegistrationobserversmightalterthebehaviorofelectoralofÞcialsinotherways,however.ElectoralofÞcialswhoseeregistrationobserversataregistra-tioncentercouldreportthisupthechainofcommand,affectingthebehavioroftheircounter-partsatotherregistrationcenters.ThisisinaccordwiththeÞndingofnodifferenceinregistrationincreasebetweenelectoralareaswithandwithoutregistrationobserversinthemodelwithoutthespillovervariables(Column1)andwiththegeneralspillovereffectfoundinthefullspeciÞcation(Column3)inTable3.However,wehavenodirectevidencetosupportthiscontention,andbyitself,thiscannotaccountfortheÞndingofpositivelocalizedspillovers.Theestimatedprimaryeffectofregistrationob-serversonregistration(),takingintoaccountlocalizedspillovereffects,maythenbeinterpretedastheeffectofregistrationobserversthroughtheirinßuenceonpartyagentsactiveduringregistration.Wedonotarguethatallregistrationirregularitiesweredeterredwhereregistrationobserverswerepresent,andwewouldunderestimatetheextentofregistrationinßationiftheregistrationofeligiblevoterswasalsosuppressed.Hence,thisisanestimateofthelowerboundontheextentoftheregistrationinßationformofirregularitiesinGhanain2008.ConclusionThisarticleextendstheempiricalscholarshiponelectoralfraudtothestudyofmisconductatthepreelectionstage.ItpresentsÞndingsfromarandom-izedÞeldexperimentontheeffectofdomesticobserversontheextentofirregularitiesinvoterregistrationinGhanain2008.Ourresearchdesignandanalysisexplicitlytakeintoaccountthespilloversthatmayresultfromtheorganizationofpoliticalpartiesacrossmultipleelectoralareasandtheca-pacityofpartyagentstotransportsupportersfromoneelectoralareatoanother.WeÞndageneralspilloverorconstituency-leveleffect;theincreaseinthenumberofregisteredvotersfrom2004to2008wasonaverage4.1percentagepointssmallerforelectoralareasinconstituencieswithsomeregistrationobserversthanelectoralareasinconstituencieswithnoregistrationobservers.Furthermore,withinconstituencieswithregistrationobservers,theincreaseinregistrationwasonaverageapproximately3.5percentagepointssmallerinelec-toralareaswithobserversthanwithout(primaryeffect).However,anelectoralareawitharegistrationobserverlocatedwithin5kmledto,onaverage,a2.7percentagepointgreaterincreaseinregistration.Thiscombinationofapositivelocalizedspillovereffectfromnearbyelectoralareaswithanegativeprimaryeffectisstrongevidencethatdeterredextraregistra-tionsarebeingdisplaced.Baseduponthedesignoftheexperiment,weattributethiseffecttotheregis-trationobserversÕinßuenceontheactivityofpartyagents.Therefore,weinterpretthenegativeprimarytreatmenteffectasalowerboundontheextentofThisresearchonirregularitiesinvoterregistra-tionhasimplicationsforbothprodemocracyactorsandscholarsofdemocratizationandelectoralfraudManyschoolteachersanduniversitystudentswerehiredforregistrationsincetheyareliterateandregistrationtookplaceduringtheschoolholidays;theywereatschoolinOctober.deterringordisplacingelectoralirregularities Epstein,DavidL.,RobertBates,JackGoldstone,IdaKristensen,andSharynOÕHalloran.2006.ÔÔDemocraticTransitions.ÕÕAmericanJournalofPoliticalScience50(3):551Ð69.Fisher,RonaldAylmer.1935.TheDesignofExperiments.London:OliverandBoyd.Gaige,Fred,andJohnScholz.1991.ÔÔThe1991ParliamentaryElectionsinNepal:PoliticalFreedomandStability.ÕÕ31(11):1040Ð60.Hartlyn,Jonathan,JenniferMcCoy,andThomasM.Mustillo.2008.ÔÔElectoralGovernanceMatters:ExplainingtheQualityofElectionsinContemporaryLatinAmerica.ÕÕPoliticalStudies41(1):73Ð98.Humphreys,Macartan,WilliamA.Masters,andMartinE.Sandbu.2006.ÔÔTheRoleofLeadersinDemocraticDeliber-ation:ResultsfromaFieldExperimentinSaoTomeWorldPolitics58(4):583Ð622.Hyde,Susan.2007.ÔÔTheObserverEffectinInternationalPolitics:EvidencefromaNaturalExperiment.ÕÕWorldPolitics60(1):Kelley,Judith.2012.MonitoringDemocracy:WhenInternationalElectionObservationWorks,andWhyitOftenFails.Princeton,NJ:PrincetonUniversityPress.Kitschelt,Herbert,andStevenI.Wilkinson,eds.2007.Patrons,Clients,andPolicies:PatternsofDemocraticAccount-abilityandPoliticalCompetition.NewYork:CambridgeUniversityPress.Lehoucq,Fabrice.2003.ÔÔElectoralFraud:Cause,Types,andAnnualReviewofPoliticalScienceLehoucq,Fabrice,andIvanMolina.2002.Stuf ngtheBallotBox:Fraud,ElectoralReform,andDemocratizationinCostaRicaNewYork:CambridgeUniversityPress.Lindberg,StaffanI.2006.DemocracyandElectionsinAfricaBaltimore:JohnsHopkinsUniversityPress.Lindberg,StaffanI.,andMinionK.C.Morrison.2005.ÔÔExploringVoterAlignmentsinAfrica:CoreandSwingVotersinGhana.ÕÕJournalofModernAfricanStudies43(4):McCann,JamesA.,andJorgeI.Domõnguez.1998.ÔÔMexicansReacttoPoliticalFraudandCorruption:AnAssessmentofPublicOpinionandVotingBehavior.ÕÕElectoralStudies(4):483Ð503.Mebane,WalterR.,Jr..2006.ÔÔElectionForensics:VoteCountsandBenfordÕsLaw.ÕÕPresentedattheannualmeetingofthePoliticalMethodologySociety.Miguel,Edward,andMichaelKremer.2004.ÔÔWorms:Identify-ingImpactsonEducationandHealthinthePresenceofTreatmentExternalities.ÕÕ72(1):159Ð217.Myagkov,Mikhail,PeterC.Ordeshook,andDimitriShakin.TheForensicsofElectionFraud:RussiaandUkraine.NewYork:CambridgeUniversityPress.Nickerson,DavidW.2008.ÔÔIsVotingContagious?EvidencefromTwoFieldExperiments.ÕÕAmericanPoliticalScience102(1):49Ð57.Nugent,Paul.2001.ÔÔEthnicityasanExplanatoryFactorintheGhana2000Elections.ÕÕAfricanIssues29(1/2):2Ð27.Olken,BenjaminA.2010.ÔÔDirectDemocracyandLocalPublicGoods:EvidencefromaFieldExperimentinIndonesia.ÕÕAmericanPoliticalScienceReview104(2):243Ð67.Pastor,R.A.1999.ÔÔTheRoleofElectoralAdministrationinDemocraticTransitions:ImplicationsforPolicyand6(4):1Ð27.RadioFranceInternational(RFI).2010.ÔÔGbagboDissolvesGovernmentandElectoralCommission.ÕÕFebruary13,http://www.rÞ.fr/actuen/articles/122/article_6843.asp(May17,2010).Rose,Richard,andWilliamMishler.2009.ÔÔHowDoElectorsRespondtoanÔÔUnfairÕÕElection?TheExperienceofRus-Post-SovietAffairs25(2):118Ð36.Rosenbaum,PaulR.2002.ObservationalStudies.NewYork:Rubin,DonaldB.1978.ÔÔBayesianInferenceforCausalEffects:TheRoleofRandomization.ÕÕAnnalsofStatistics6:34Ð48.Schedler,Andreas.2002.ÔÔTheMenuofManipulation.ÕÕofDemocracy13(2):36Ð50.Simpser,Alberto.2010.ÔÔMorethanWinning:TheStrategyofElectoralManipulation.ÕÕUnpublishedmanuscript.UniversityofChicago.Stokes,SusanC.2005.ÔÔPerverseAccountability:AFormalModelofMachinePoliticswithEvidencefromArgentina.ÕÕPoliticalScienceReview99(3):315Ð25.Villalon,LeonardoA.1994.ÔÔDemocratizinga(Quasi)Democ-racy-theSenegaleseElectionsof1993.ÕÕAfricanAffairs(371):163Ð93.Wantchekon,Leonard.2003.ÔÔClientelismandVotingBehavior:EvidencefromaFieldExperimentinBenin.ÕÕWorldPolitics55(3):399Ð422.Yemen,Times.2009.ÔÔTheRootsofProtest:PriorElectionsImpactFuturePolls.ÕÕJuly5,DocumentYEM-(May15,2010).,Fulgence.2010.ÔÔCotedÕIvoire:CrisiswithinaCrisisDelaysElectionsAgain.ÕÕFebruary20,http://allafrica.com/stories/201002210001.html(May17,2010).Ziblatt,Daniel.2006.ÔÔHowDidEuropeDemocratize?ÕÕ58(2):311Ð38.Ziblatt,Daniel.2009.ÔÔShapingDemocraticPracticeandtheCausesofElectoralFraud:TheoryandEvidencefromPre-1914Germany.ÕÕAmericanPoliticalScienceReview(1):1Ð22.NahomiIchinoisAssistantProfessorofGovernmentatHarvardUniversity,Cambridge,MAMatthiasSchundelnisProfessorofEconomicsatGoetheUniversityFrankfurt,60323FrankfurtamMain,Germany.deterringordisplacingelectoralirregularities