1IntroductionItiswidelyappreciatedthatlaboristhemostabundantresourceof - PDF document

1IntroductionItiswidelyappreciatedthatlaboristhemostabundantresourceofthepoor.Inagriculturaleconomiesthepoormayworkontheirownlandtoproducegoodsforhomeconsumptionormarketsale,andworkforotherpeopleforwages.Paidemploymentoftentakestheformofcasualdaylaborratherthanlonger-termarrangementsgovernedbycontracts,andcanbeanimportantsourceofcashaswellasamechanismforcopingwithnegativeshocksthatreducenon-laborincome.Theimportanceofthistypeoflaborishighlightedbypublicsectoremploymentprogramswithdualgoalsofinfrastructuredevelopmentandincomesupport.Malawi,whichhasinvested$40millioninitsCommunityLivelihoodsSupportFund,isoneof29countriesinsub-SaharanAfricawithapubicsectorworksprogram[18].Asanotherexample,almost45millionhouseholdswereemployedtododaylaborthroughtheNationalRuralEmploymentGuaranteeActinIndiain2008-2009.Despitetheimportanceofdaylabortoruralhouseholdsandthelargescaleinvestmentsinprogramstoemploydaylaborersbygovernmentsindevelopingcountries,littleisknownabouttheelasticityofemploymentfordaylaborers.Infact,thereisscantevidenceontheelasticityoflaborsupplyinanytypeoflabormarketindevelopingcountries.Theusualchallengeinestimatingtheelasticityoflaborsupplyisthatwagesareendogenous;indevelopingcountries,thereistheadditionalchallengeofobtaininghighqualitydata.Iovercometheidenticationproblembyrandomizingwagesforcommunityagriculturaldevelopmentprojectsin10villagesinruralMalawi.Iestimatetheprobabilityofacceptingemploymentinthedaylabormarket,therelevantmarketformillionsofindividualsinpoor,ruralcommunities.Mysampleincludes529adultsfromhouseholdsthathavesupplied\ganyu,"ordaylabor,inthepreviousyear.Theseindividualsareoeredemploymentonedayperweekfor12consecutiveweeks.Wagesvarybyvillage-week,rangingfromMK30($US0.21)toMK140($US1.00)perday,andwagesforeachworkdayareannouncedoneweekinadvance.Iestimatetheelasticityofworkingonagivendayusingadministrativeattendancerecords,andusesurveystostudychangesinlaborsupplyinresponsetohouseholdshocks.Indthatatenpercentincreaseinwagesleadstoa1.5to1.7percentincreaseintheprobabilityofworking,withnodierencesbetweenmenandwomen.Myresultsstandincontrasttothecommonndingindevelopingand2 developedcountries,manyemployeesareconstrainedtoworkeitherfull-timeornotatall,withlittleopportunitytoadjusttheirnumberofweeksperyear.Camereretal.[7],Chou[9],andFarber[10]takeadvantageofanotableexceptiontorigidlabormarketsbystudyingtherelationshipbetweenhoursworkedandtheimpliedhourlywagefortaxicabdrivers.Camereretal.andChoundapuzzlingresult:taxidriverstoworkfewerhoursonmoreprotabledays,implyingadownward-slopingsupplycurve.Theyexplainthisresultthroughso-called\targetearning"behavior:taxidriverssetagoalfordailyearningsandstopworkwhentheyreachtheirgoal.Usingaricherdatasetandadierentapproachtoimputinghourlywages,Farberndsthattaxidriversworklongerhourswhenhourlywagesarehigher,thestandardupward-slopinglaborsupplycurve.Ashenfelteretal.[3]returntothetaxicabdriverpuzzleandstudychangesinhoursworkedinresponsetoexogenouschangesinfares.Theyestimatetheelasticityoflaborsupplyinresponsetoalongrunchangeinwagestobe-0.20.ThoughthesepapersarebasedondatafromtheUnitedStatesandSingapore,theyaresomeoftheonlypapersinthelaborsupplyliteraturetostudyasituationcomparabletothatinMalawi'smarketforganyu,wherelaborsupplycanbefreelyadjustedintheshortrun.Mypaperproceedsasfollows.IdescribetheexperimentinSection2anddescribethedatainSection3.Ipresenttheframeworkandresultsforestimatesofmymainparameter,theelasticityofemployment,inSection4.IhighlightthemethodologicaldierencesbetweenmyresearchandthepreviousliteratureabouttheelasticityoflaborsupplyindevelopingcountriesinSection5,andusesupplementaldatafromMalawi's2004IHStoprovideacontextforthesimilarityofmen'sandwomen'selasticitiesinSection6.IconcludeinSection7.2ExperimentalDesignIrandomizethewagesthat529adultsin10villagesinruralMalawiareoeredfordoingmanuallaboronagriculturaldevelopmentprojects.Projectparticipantsarerecruitedfromhouseholdswhohavedonesimilarpaidworkinthepastyear.Theyareoeredajobonedayperweekfor12consecutiveweeks.Thejobisthesameeachweek,butwageschange.Eachweek,participantscaneitheraccepttheoeredwageandworkforthefullday,orrejectthewageandnotworkatall.Wagesareannouncedoneweekinadvance,andtheMK30to4 Withineachvillage,LHAleadersandextensionworkerschoseaworkactivity.Theseactivitieswerebydesignlaborintensive,unskilled,andhadpublicratherthanprivatebene-ts.Tobeconsistentwithlocalstandards,\oneganyu,"oraday'swork,lastedfourhours.Activitiesincludedclearingandpreparingcommunallandforplanting,diggingshallowwellstobeusedforirrigation,andbuildingcompostheapstobeusedtofertilizecommunalland.Withineachvillage,theactivitywasthesameforall12weeks.Theamountofeortwasheldconstantbyobjectivestandardsfromweektoweek:participantshadtodigthesamenumberofcubicfeetorhoethesamenumberoflinearfeeteachweek.Sinceallanalysesincorporatevillagexedeects,dierencesbetweenactivitiesacrossvillagesdonotaecttheresults.Upto30householdsineachvillagewereinvitedtoparticipateintheproject.Qualifyinghouseholdshadtohaveatleastoneadultmemberwhohadperformedganyuwithinthelastyear.Uptotwoadultsperhousehold{usuallybutnotalwaystheheadofhouseholdandhisspouse{wereinvitedtoparticipate.Whilehavingmultipleparticipantsperhouseholdcomplicatestheanalysisofanindividual'sresponsetoachangeinhisownwagesbecausehouseholdincomeisnotheldconstant,itallowsmetoidentifytheelasticitywithrespecttothechangeinwagesthatisrelevantinthiscontext.Muchoftheliteratureinlaboreconomicsconsiderschangesinwagesforasinglememberofahousehold,holdingconstantincomeforotherhouseholdmembers.Thatistherelevantparameterindevelopedcountriesorurbanareas,wherehouseholdmembersoftenparticipateindierentjobmarkets.However,itisnotrelevantinruralareasindevelopingcountries,whereadultshavehomogenousworkopportunities.InMalawi,menandwomenperformsimilaron-ando-farmlabor.Menandwomenmayparticipateinthegovernment'sPublicWorksProgramme,whichpaysindividualsinpoorhouseholdstoworkoncommunityinfrastructureprojectssuchasroadconstruction.Allowingmultipleadultsperhouseholdtoparticipateinthisprojectisakintostudyingtheeectofatransitorychangeintheprevailingvillagewageforunskilledlabor.Participatinghouseholdsweregiventheopportunitytoworkforpayontheirvillage'sactivityonedayperweekfor12consecutiveweeks.Theworkdaywasthesameeachweekforeachvillage,sothatvillagexedeectsalsocontrolforday-of-weekeects.Participantsweretoldattheoutsetthattheprojectwouldlast12weeks,thattheworkwouldbethe developmentpoliciesandotheractivitiesacrossvillagesunderhisdomain.6 wageswouldbeloweronthemorecomprehensivedenitionofemployment.Thelackofaneectonoutsideemploymentisconsistentwiththenotionthatdemandforlaborisscarceduringthedry,unproductivetimeofyearwhenmyprojecttookplace.Myinterpretationisalsoconsistentwiththelimitedliteratureonemploymentindailywagemarkets:despitesimilargapsbetweenemploymentopportunitiesforstadiumvendors,Oettingerinterpretshisestimatesasintertemporalelasticitiesofsubstitutionoflaborforleisure.TheprojecttookplaceinJune,July,andAugust,monthsthatfallbetweentheharvestandplantingseasonsinMalawiandcomeduringthecountry'sdryseason.Thisisatimeofyearwithlowmarginalproductivityeitheron-oro-farm.Itisthetimeofyearwhenindividualshavethemostfoodandmostcash.Importantly,Icanbecondentthattheopportunitycostoftimewasconstantthroughouttheexperimentalperiod.Laborsupplyelasticitiesmayvaryseasonally,andtheestimatesfromthisexperimentarenotnecessarilyvalidforadierenttimeofyear,whentheopportunitycostoftimeishigher.WagesforthisprojectrangefromMK30/day($US0.21)toMK140/day($US1.00),inincrementsofMK10.3Table1showsthescheduleofwages,whichalternatedhighandlowwagesoverthe12-weekdurationoftheproject,thenshiftedthescheduleforwardinordertohave10separateschedulesthatfollowedthesamepatternofincreasesanddecreases.Using10dierentwageschedulescreatesvillageweekvariationthatallowsmetocontrolforvillageandtimexedeectsseparately.Theshiftedschedule(asopposedtoi.i.d.randomizedwages)meansthateachvillagehasthesametotalearningspotentialandthataveragesacrossvillages,withinweek,areapproximatelyconstant.Sinceitispossiblethatparticipantswillconsiderrelativewages,thescheduleisdesignedsuchthateachvillagefacesthesamenumberofwageincreasesanddecreases.Afterrandomlyallocatingeachvillagetoawage-schedule,IallowedLHAleadersandgovernmentextensionworkerstodeterminethedayoftheweekonwhichvillageswouldbevisited.4 3ThewagesarebasedonoutcomesfromapilotstudyIconductedinMarch2009,where77percentofparticipantsworkedforthelowestoeredwageofMK70,and96percentworkedforthehighestoeredwageofMK120.4ThelistofvillagesgiventoLHAleadersandextensionworkersre ectedtherandomization,i.e.thevillagerandomlyselectedas\villageone"waslistedrst,thevillagerandomlyselectedas\villagetwo"wassecond,etc.TheLHAleadersandextensionworkersretainedthatorderinginmanycaseswhendecidingwhichvillagestovisitonwhichdaysoftheweek.SinceIusevillagexedeects,andsincethewagescheduleisexogenousineachvillage,therelationshipbetweenday-of-weekandwagescheduledoesnotcompromisetheresults.8 Ofthe529individualsincludedintheproject,370respondentsarespouseslivingin185households.Another74arewomeninhouseholdswherebothprojectparticipantsarewomen,and18aremeninhouseholdswherebothprojectparticipantsaremen.Theremaining67areindividualswhoaretheonlyparticipantsintheirhouseholds.Thesurveyteamwasabletointerview495participantstheweekbeforetheprojectbegan.Respondentsinpre-selectedhouseholdswhowerenotavailableduringthesurveyperiodwerenonethelessallowedtoparticipateinthestudy,toavoidcreatingasamplebiasedtowardsthosewithlowopportunitycostoftime.Table2presentsbaselinecharacteristicsforparticipantsinthisproject.Themajorityofthesamplearemarriedwomen.7Participantshaveattendedanaverageoffouryearsofschoolandliveinhouseholdswithapproximatelytwoadultsandthreechildren.Respondentsownanaverageof1.8acresofland;theirhouseshaveanaverageoftworooms;andonly16percentofrespondentshavetinroofsontheirhouses.Theyworkanaverageofonedayintheweekbeforethesurveyor2.7daysinthemonthbeforethesurvey.4ElasticityofemploymentIestimateachangeintheprobabilityofworkingonagivendaywithrespecttoachangeinthatday'swages,aparameterIwillcalltheelasticityofemployment.Thisisareduced-formestimateofanuncompensated,intertemporalparameter,butitdiersfromthefamiliarFrischelasticityortheelasticityoflaborforceparticipationinwaysIexplaininthenextsection.Thechangeintheprobabilityofworkingcapturestherelevantmarginofchoiceinthemarketfordaylaborinpoorruraleconomies,whereindividualsworkeitherafulldayornotatallbutmaychoosetheirnumberofdayswithconsiderablymore exibilitythaniscommonindevelopedcountries.Iestimatethattheelasticityofemploymentisbetween0.15and0.17.Theseestimatesarerobusttoalternativespecicationsusingdierentcombinationsofvillage,week,andindividualxedeects;themarginaleectsfromOLSandprobitspecicationsarevirtuallyidentical.Includingwagesforpreviousorfutureweeksdoesnotchangethepointestimatesoftheelasticitywithrespecttothecurrentweek'swage,andmyinferencesarerobusttoseveralalternativemethodsofcomputingstandarderrors. 7Includingwidowedmenandwomenorthosewhosespousesaredisabledorpermanentlyunavailableforworkwasapreferenceofmypartnerorganization.Allofmyresultsarerobusttolimitingthesampletothe370respondentswhoaremarriedandwhosespousesarealsoparticipatingintheproject.10 whenonlythebinaryparticipationdecisionisobserved,wemayestimatetheextensivemarginelasticityortheelasticityofparticipationfromthederivativeofexpression(4)withrespecttoW:extensive=@Pr(H�0jW;Y) @WW H.Theoretically,themarginaleectofwagesonlaborsupplyattheintensivemarginmaybelargerorsmallerthanthemarginaleectofwagesonlaborsupplyattheextensivemargin.Empirically,\Participation(oremployment)decisionsgenerallymanifestgreaterresponsivenesstowageandincomevariationthandohours-of-workequationsforworkers,"(Heckman[14])basedonempiricalestimatesfordevelopedcountries.Whiletheelasticityoflaborsupplyattheintensivemarginhasreceivedmoreattentionintheempiricalliteratureindevelopedcountries,therearemanyinstanceswheretheextensivemarginelasticityisthepolicyrelevantparameter.Forexample,thechangeinaggregatesupplyoflaborbysinglewomenduetotheexpansionoftheEarnedIncomeTaxCredit(EITC)inthe1990swasdominatedbyanincreaseinlaborforceparticipation(Meyer[19]).UnderstandingtheimpactoftheEITCexpansion,then,requiresanestimateoftheincreaseinlaborforceparticipationduetothepolicychange.Indevelopingcountrieswithlarge-scalepublicworksprograms,includingMalawi's$40millionCommunityLivelihoodsSupportFundandIndia'sNationalRuralEmploymentGuaranteeAct,whichmakesoverabillionpeopleeligibleforupto100daysofworkperyear,understandingthechangeinthefractionofthepopulationwhowouldworkundertheprogramatdierentwagesisofcrucialimportance.Themarketfordaylabor,whereindividualscanworkornotworkfortheprevailingwageeachday,blursthedistinctionbetweentheintensiveandextensivemarginatthesametimeitmakescleartheseparationofparticipationversusemployment.InadailylabormarketthedecisionofH=0orH�0ismadeeachday,andre ectsmovementbetweenemploymentandunemploymentbutnotbetweenlaborforceparticipationandnon-participation.Somepeoplechoosenottoworkonagivendaybecausetheprevailingwageislessthantheiropportunitycost,butwouldhaveworkedhadtheday'swagebeenhigher.Thus,theyareinthemarketfordaylaboreventhoughtheyarenotemployedonagivenday.EmpiricalestimatesoftheprobabilityofworkinginadaylabormarketshouldconditiononadierentparticipationindicatorthanH�0,andestimatealaborsupplyfunctionthatcombineselementsofequations(2)and(4)above:Pr(H�0jW;Y;indailylabormarket)(5)12 4.2PointestimateoftheelasticityofemploymentIndthatoverallemploymentishighandtheelasticityofemploymentislow,preciselyestimated,androbusttomanyalternatespecications.IplotthefractionofthesamplewhoworkateachwageoerinFigure1.AtMK30/day,thelowestwageinthesample,morethanseventypercentofrespondentsworked.Thishighbasehasastrongseasonalcomponent:marginalproductivityathomeoronone'sownfarmislowduringthedryseason,andthereisverylittledemandforo-the-farmlabor.However,employmentatlowwagesischaracteristicofthemarketforganyuinMalawi.ThelowestreportedwagesintheIHSareMK10/day,andaquarterofthosewhodoganyureportreceivingMK40/dayorlessonaverage.Thereisamarginallysignicant(p=0:10)discontinuityintheprobabilityofemploymentatawageofMK100/day.10Despitethisdiscontinuity,IfocusontheelasticityofemploymentacrossthethefullrangeofwagesratherthanthechangeintheprobabilityofworkingatMK100.Muchoftheliteratureaboutlaborsupplyindevelopingcountriesfocusesontheelasticityoflaborsupply,sothischoicefacilitatescomparisonsbetweenmyresultsandpreviousresearch.Furthermore,thedesignofmyexperimentisnotwell-suitedtoidentifyinganon-linearchangeintheprobabilityofworkingatMK100.BecauseofthewagescheduleIuse,everywageofMK100orhigherisanincreasefromthepreviousweek'swage(except,ofcourse,intherstweek),andeverywageofMK90orlowerisadecreasefromthepreviousweek'swage.Therefore,itisnotpossibletodeterminewhetherajumpupintheprobabilityofworkingatMK100isbecauseofareservationwageofMK100,orbecauseofapreferenceforwageincreases.11IfthecorrectmodelisonethatallowsforadiscontinuityatMK100,thenmyestimatesoverstatetheelasticityofemploymentandmyconclusionthattheprobabilityofworkingisinelasticwithrespecttowageswouldbestrengthened.Inordertoestimatetheelasticityofworking,Irunordinaryleastsquaresregressionsoftheformlaboritv=+ln(wagetv)+itv.Thecoecientisthemarginaleectofaonelog-point,orapproximatelyone-percent,changeinwagesontheprobabilitythatanindividualworks.Themarginaleectisnotanelasticity,butitiseasilytransformedintoone 10Thegovernment'sratefordaylaboriscurrentlysetatMK200,butwaspreviouslyMK110.Thediscontinuitydoesnotsuggestareferencepointcorrespondingtothegovernment'swagerate.11Usingdatafromtherstweekonlyandrelyingoncross-villageidenticationforvariation,theprobabilityofworkingforwagesofMK90andlowerisnotstatisticallydierentfromtheprobabilityofworkingforwagesofMK100andhigher.14 specication.Toincludefuturewages,Ihavetolimitthesampleaccordingly.ThelefthandpanelofTableA1includesweeksoneto11.Irstpresentabaselinespecicationforthesubsample,thenshowspecicationswithfuturewagesandwithxedeects.Column(1),includedforreference,isthesamespecicationasTable3column(1).Theestimatedelasticitywhenusingtherst11weeksofdatabarelydiersfromthatforthefullsample.Addingameasureofwagesoneweekinthefuturedoesnotchangetheestimatedelasticity,andthecoecientonfuturewagesisverysmallandnotstatisticallydierentfromzeroinbothcolumn(2),whichdoesnotincludexedeects,andcolumn(3),whichincludesindividualandweekxedeects.IntherighthandpanelofTableA1,Ifurtherlimitthesampleinordertoincludemoreweeksoffuturewages.Noneofthecoecientsonthemeasuresoffuturewagesaresignicant,andIalsorejectjointsignicanceofthecoecientsonfuturewages.Iinterpretthistableasevidencethatparticipantsdidnotdetectthenegativeserialcorrelationinthewages,andthattheirlaborsupplydecisionwasbasedoncurrentwagesratherthananticipationoffuturewages.Anotherchallengetotheinterpretationofmyestimatesasintertemporalparametersisthattheunderlyingexpectationsaboutwagescouldhavechangedoverthecourseoftheexperiment.ThoughIdesigntheexperimenttoreplicatetypicalmarketemploymentasmuchaspossiblebyhavingregularemployerssupervisetheworkanddistributewages,andbyusingataskforwhichawagemarketdoesexist,participantswereawarethattheywereworkingfora\project"withtheverynon-standardfeatureofhigh-variancewages.Atthebeginningoftheproject,itisreasonabletoassumethattheyexpectedawageofMK110{theusualwagerateongovernmentprojects.Theassumptionisthatthetemporary,announcedchangesinwagesfortheprojectdidnotalterparticipants'underlyingexpectations.If,however,expectationsevolvedinresponsetorealizedwageshocks,thentheestimatedelasticitywouldnotbeintertemporalinthestandardsenseofachangeinlaborsupplyinresponsetoananticipatedtemporarychangefromthelongrunexpectationofwages.Therobustnessofmyestimatestoweekxedeectsprovidessomeindicationthatchangesinexpectations{whichwouldbecorrelatedwithtimeintheproject{arenotamajorfactor.Foramoredirecttest,Iincludewagesinpastweeks,usingspecicationsanalogoustothoseforfutureweeksinTableA1.Thatpastwagesdonotaecttheprobabilityofworkingandthatthecoecientoncurrentwagesdoesnotchangewhenpastwagesareaddedtothe16 someotheraspectoftheexperimentalsetting.Aspecicpitfallwouldbeiflaborsupplywasinelasticbecauserespondentsfeltpressuredtoworkdespitethewage,orthoughttheywouldbeeligibleforsomeotherbenetiftheywereperceivedas\cooperative"or\hard-working."Ihaveevidencethatthisisnotthecase.Respondentslisteduptothreereasonsforworkinginweeksthattheyworked,orthreereasonsfornotworkinginweekstheydidnotwork.Wagesdonotappeartobeamajorfactorinthedecisioneithertoworkornottowork.Reasonsforworkingweregroupedintofourcategories:becauseofthewage(usedonlywhentherespondent'sliteralanswerwas\becauseofthewage"or\becausethewagewasgood"),togetmoneytospendimmediately,togetmoneytosave,orbecauseofsocialpressureorperceivedbenetsbesidesthewage.Figure2showsthefractionofindividualswhomentionedeachreason,aggregatedacrossweeksforindividualswhoworkedateachwage.Earningmoneytospendimmediatelyisthedominantfactoratallwagelevelsandismentionedbyover70percentofrespondents,nomatterwhatthewage.Socialpressuretowork,whichincludesbeingtoldtoworkbyalocalleaderorgovernmentextensionworkeroranticipatingsomerewardforcooperation,seemsrelevantonlyatthelowestwage,MK30.ThewageitselfismentionedbyfewerthantwopercentofrespondentsforallwageslessthanMK100,butby30percentormoreofrespondentsatwagesofMK100orhigher.Reasonsfornotworkingweregroupedintosixcategories:becauseofthewage(again,usedonlywhenrespondentsspecicallyreferencedbadwages),becausetherespondentwasoccupiedwithotherwork,becausemoneywasnotneeded,becauseofafuneral,becauseofillness(totherespondentorsomeonehe/shewascaringfor),andbecauseofsocialpressurenottowork.Figure3showsthereasonsfornotworkingateachwage.Illnessesandfuneralswerethedominantcausesofnotworking,whichisconsistentwiththestrongnegativeeectoffuneralsonlaborsupplyintheadministrativedata.Wageswerementionedbyfewerthan20percentofrespondentsatallwagelevelsexceptforthelowesttwo,MK30andMK40,andanunexplainedspikeatMK80.Theseself-reporteddataareconsistentwiththehighlyinelasticlaborsupplyestimatedintheprevioussection.Otherfactorsdominatewagesinthedecisiontoworkornottowork,evenatveryhighorverylowwagelevels.18 estimatestowhichanalystsarelimitedwhenusingcrosssectionaldatawithoutexogenousvariationinwages.Therstdierenceiscontext:theremaybeinherentdierencesbetweenthelabormarketsinruralMalawi,WestBengal,andGhana.Theseconddierenceistheparameterbeingestimated.Iestimateanextensivemarginelasticity,thechangeintheprobabilityofworkingonagivendayforpeoplewhohavealreadyselectedintothemarketfordaylabor.Mostoftheliteraturefocusesonanintensivemargin,thechangeinhours(ordays)worked.Thepointestimatesoftheelasticitiesatthesetwomarginswouldbedierentevenifestimatedfromthesamedataset.Thethirddierenceisinthedistributionofwages:Iobservethefulldistributionofwageoers,whilemostestimateshavedataoncensoredwages.Thefourthdierenceisinthesourceofvariationinwages.Wagesareexogenousbydesigninmyproject,butendogenousinnon-experimentalcrosssectionaldata.Usingthetime-aggregatedcrosssectionallowsmetoholdconstantthemethodologicalissues,whicharethesecond,third,andfourthdierencesdiscussedabove.IcanthenassesswhetherlackofexternalvalidityexplainswhytheelasticitiesIpresentinsection4arelowerthanthoseinthepreviousliteratureaboutdevelopingcountries.Toconstructthedependentvariable,Iaddupthetotalnumberofdaysworked(whichrangesfrom0to12).ThisistheconceptthatBardhanusesbytakingthetotalnumberofdaysworkedintheseven-dayperiodcoveredbythesurveyofhouseholdsinWestBengalthatheanalyzes.Notethatthismeasureinmycrosssectionisalreadymoreprecisethannormalinsurveydata,becauseitcomesfromadministrativerecordsratherthanself-reports.Everyindividualinthesampleworkedatleasttwodays,and,onaverage,individualsworked10days.Sinceeveryindividualworkedatleastonce,itisnotpossibletoestimatetheelasticityoflaborforceparticipationusingthecrosssectionaldataforthissample.Iconstructthreedierentmeasuresofwages.First,Iusethecommon\averagewage"measurebytakingthewithin-personacross-weekaverageacceptedwage.Thismeasuredoesnotcorrectforendogenouswagesorselectionintoemploymentatall.Also,becauseallwagesthatwereoeredinthisexperimentwereacceptedbyatleastsomeparticipants(andinpractice,eventhelowestwagewasaccepted73percentofthetimeitwasoered)andallparticipantshadthesamedistributionofwageoers,theindividualaveragewagemeasuresinthesimulatedcrosssectionareendogenousbutnotcensoredonthedependentvariable.Second,followingBardhan,Icomputethe\villageaveragewage"asthewithin-villageacross-20 workedveofsixpossibledays.Theelasticityoflaborsupplywithrespecttothisbetter-measuredconceptofaveragevillagewagesisbetween0.33(withoutcovariates)and0.30(withindividualcovariates).13WhenIuseacomparabledatasettoidentifytheintensivemarginelasticity,myestimatesaresimilartoorlargerthanelasticitiesestimatedbyBardhan[4](0.20to0.29)andAbdulaiandDelgado[1](0.32formenand0.66forwomen).ThissuggeststhatthehighlyinelasticestimatesinmypreferredspecicationsthattakeadvantageoftheexperimentaldesignandestimatethechangeintheprobabilityofworkingonagivendayarenotexplainedbyinherentdierencesbetweenthelabormarketsinruralMalawiandtheseothercountries.Instead,acombinationofthethreetypesofmethodologicaldierencesIdiscussedatthebeginningofthissectionleadstomuchlowerestimatesthanfoundinpreviousresearch.Iwouldndhigherelasticitiesusingdatafromthesamelabormarketifmydataweresubjecttothebiasesinstandardanalysisofanon-experimentalcrosssection.6GenderAlongliteraturesuggeststhatwomensupplylabormoreelasticallythanmenindevelopedcountries(e.g.Killingsworth[15],Heckman[14]).PreviousworkindevelopingcountriesisalsoconsistentwithwomensupplyinglabormoreelasticallythanmeninIndia[22]andGhana[1].InTables5and6,Ilookatmyexperimentalsamplesofmenandwomenseparately.Onaverage,81percentofmenworkwhenoeredemployment.Theestimatedelasticityformenrangesbetween0.16and0.19,withxedeectsaddedacrosscolumnsinTable5asinTable3.Resultsforwomenarestrikinglysimilar.Some86percentofwomenworkacrosstheentiresample.Theirelasticitywithrespecttowagesfallsbetween0.14and0.15,estimatesthatarenotstatisticallydierentfromtheestimatedelasticitiesformen.Inthissection,IdemonstratethatsimilarelasticitiesformenandwomenisacharacteristicofthemarketforganyuduringMalawi'sdryseasonratherthananartifactofmyexperimentaldesign.Justastherearemanyreasonsthatmypointestimatesoftheelasticityofemployment 13Ialsoestimatetheelasticityoflaborsupplybydrawingsixconsecutiveweeksofdataforeachvillage,becauseconsecutiveweeksismorecloselyanalogoustotheconceptmeasuredincrosssectionaldata.Theelasticitiesfromestimatesusingcrosssectionaldataare0.39(withoutcovariates)and0.37(withcovariates).Mypreferredspecicationistheoneusingnon-consecutiveweeks,becausethewageschedulemechanicallyreducestheacross-villagevariationwhenusingconsecutiveweeks.22 datatomatchthepointestimatesfrommyexperiment:wagesintheIHSareendogenousandestimatesusingtheIHSarelikelybiased.Iaminterestedincomparingthepatternofelasticitiesbygender,notthepointestimates.Table7showsresultsfromthisexercise.PanelAcontainsresultsforthedryseason,whichincludesJune-November.PanelBcontainsresultsforthewetseason,December-May.Thesampleislimitedtotheheadofhouseholdandhisorherspouse,ifpresent,tomatchtheselectioncriteriaformyexperiment.Allregressionscontrolforgender,age,householditemsscore,housingqualityscore,landareafarmedduringthedryseason,landareafarmedduringthewetseason,amountoffertilizerusedduringtherainyseason,education,anddistrictofresidence.Columns(1)to(3)capturetheintensivemarginelasticityfromtheregressionofloghoursonlogwages.Theelasticityduringthedryseasonis0.475,marginallydierentfromzero.Estimatesformenandwomenareimprecisebutnotsignicantlydierentfromeachother.Duringthewetseason,however,theintensivemarginelasticityfallsbyhalfandisnotstatisticallydierentfromzero.However,theseparateestimatesformenandwomentelladierentstory.Formen,thepointestimateis-0.256,which,whilenotstatisticallydierentfromzero,isconsistentwithpreviousndingsthatmen'slaborsupplyiseitherinelasticorinthebackward-bendingportionofthelaborsupplycurve.Womenhaveanelasticityof0.639,signicantlyhigherthanmen.Duringtherainyorhigh-productivityseason,then,thepatternofmen'sandwomen'sintensivemarginelasticitiesinruralMalawiareconsistentwithevidencefromotherdevelopingcountries.Duringthedryseason,though,genderdierencesaremuchhardertodetect.Theextensivemarginestimatesincolumns(4)to(6)aremorecomparabletoestimatesfrommyexperiment.Duringthedryseason,theelasticityofworkinginthepastweekformenandwomencombinedis0.27.Womenhavesomewhatlargerelasticitiesthanmen,butthedierencebetweenmenandwomenisnotstatisticallysignicant.Inthewetseason,though,theelasticityforwomenis0.45,signicantlydierentfromzero,whiletheelasticityformenis-0.11andnotstatisticallysignicant.Inotherwords,ndingpositiveelasticitiesofworkingthataresimilarformenandwomendoesnotappeartobeanartifactofmyexperimentaldesign.Thesamepatternispresentinnationally-representativesurveydatawhenlookingatdatafromthesamepartoftheagriculturalseason,thoughtheestimatesarelessprecise.24 ofwagesandworkhistoryconrmthatrespondentsareaccurateintheirmemoryoftheevents,reportingbothwagesandpastworkaccuratelyin83percentofthecases.Ithenuseinformationfromweeksinwhichrespondentsrememberedthewageandwhethertheyworkedtoexamineself-reportedreasonsforworking.Atallwagelevels,earningmoneytospendimmediatelyisthemostfrequentlyreportedreasonforworking,andfuneralsandillnessesarethedominantreasonsfornotworking.Wagesarecitedbymorethan20percentofrespondentsasareasonfornotworkingpredominantlyatverylowwages(MK30andMK40),andasareasonforworkingonlyathighwagesofMK100orhigher.Thesesurveyresponsesareconsistentwiththeinelasticsupplyoflaborobservedintheadministrativedata.UnderstandingthelaborsupplybehaviorofpoorindividualsiscrucialforthedesignofpublicemploymentprojectsinMalawiandotherdevelopingcountries.TheGovernmentofMalawiandtheWorldBankarespending$40milliononaCommunityLivelihoodsSupportfundthatusespublicsectoremploymenttomeetdualgoals:providingasafetynetforpoorindividualsbyoeringemployment,andimprovinginfrastructureinthecommunitieswherethoseindividualslive.Inelasticlaborforceparticipationmakesitclearthattherearestarktradeosbetweenthesegoalswhendeterminingwagelevelsfortheprogram.Malawiisnottheonlydevelopingcountrywithaninterestinpublicemploymentprograms:29countriesinsub-SaharanAfricaalonehavesuchprograms.TheestimatesIobtainfrommyexperimentinMalawinotonlycontributetothelongandevolvingliteratureaboutlaborsupplyindevelopingcountries,butalsoprovideimportantparametersforunderstandingtheimpactofgovernmentandNGOprogramsthatarealreadyreachingmillionsofpeople.26 Table3:Elasticityofemploymentw.r.t.wages (1)(2)(3)(4)(5)(6)Dependentvariable:Individual*dayindicatorforworking Ln(wage)0.127***0.127***0.140***0.140***0.127***0.140***(0.033)(0.033)(0.032)(0.032)(0.033)(0.032) VillageeectsxxWeekeectsxxxIndividualeectsxx Observations633363336333633363336333Meanofdependentvariable0.840.840.840.840.840.84Elasticity0.150.150.170.170.150.17(0.040)(0.040)(0.040)(0.040)(0.040)(0.040) OLSestimates.Clusterbootstrappedstandarderrors(clusteredatthevillagelevel).Unitofobservationisindividual*week,sampleisallindividuals.*p0.10,**p0.05,***p0.001 28 Table5:Elasticityofmen'semploymentw.r.t.wages (1)(2)(3)(4)(5)(6)Dependentvariable:Individual*dayindicatorforworking Ln(wage)0.139***0.139***0.157***0.157***0.139***0.157***(0.032)(0.032)(0.035)(0.035)(0.032)(0.035) VillageeectsxxWeekeectsxxxIndividualeectsxx Observations253225322532253225322532Meanofdependentvariable0.810.810.810.810.810.81Elasticity0.170.170.190.190.170.19(0.043)(0.043)(0.047)(0.047)(0.043)(0.047) OLSestimates.Clusterbootstrappedstandarderrors(clusteredatthevillagelevel).Unitofobservationisindividual*week,sampleisallmen.*p0.10,**p0.05,***p0.001 Table6:Elasticityofwomen'semploymentw.r.t.wages (1)(2)(3)(4)(5)(6)Dependentvariable:Individual*dayindicatorforworking Ln(wage)0.119**0.119**0.129***0.129***0.119**0.129***(0.035)(0.035)(0.032)(0.032)(0.035)(0.032) VillageeectsxxWeekeectsxxxIndividualeectsxx Observations380138013801380138013801Meanofdependentvariable0.860.860.860.860.860.86Elasticity0.140.140.150.150.140.15(0.041)(0.041)(0.038)(0.039)(0.041)(0.038) OLSestimates.Clusterbootstrappedstandarderrors(clusteredatthevillagelevel).Unitofobservationisindividual*week,sampleisallwomen.*p0.10,**p0.05,***p0.001 30 Table8:Eectofsavingsaccountsonelasticityofemploymentw.r.t.wages (1)(2)(3)(4)Dependentvariable:IndividualIndividualHouseholdHousehold Ln(wage)0.141***0.140***0.287***0.269***(0.033)(0.027)(0.044)(0.051)Account-0.017-0.026-0.041-0.195(0.013)(0.097)(0.030)(0.191)Account*Ln(wage)0.0020.035(0.020)(0.041) VillageeectsxxxxWeekeectsxxxx Observations6285628527482748Meanofdependentvariable0.840.840.740.74Elasticity0.170.17(0.041)(0.042)Elasticity(noaccount)0.170.16(0.034)(0.035)Elasticity(account)0.170.18(0.049)(0.051) OLSestimates.Clusterbootstrappedstandarderrors(clusteredatthevillagelevel).Incolumns(1)and(2),unitofobservationisindividual*week,sampleisallindividuals.Incolumns(3)and(4),unitofobservationisHH*week,sampleisHHswithtwoparticipants.*p0.10,**p0.05,***p0.001 32 FiguresFigure1:Fractionworkingateachwage(wagesinMK) 34 Figure3:Self-ReportedReasonsforNotWorking 36 [11]ErnestFehrandLorenzGoette.Doworkersworkmoreifwagesarehigh?evidencefromarandomizedeldexperiment.AmericanEconomicReview,97(1):298{317,Apr2007.[12]JohnHam.Testingwhetherunemploymentrepresentslife-cyclelaborsupply.ReviewofEconomicsStudies,54(4):559{578,October1986.[13]JohnHamandKevinReilly.Testingintertemporalsubstitution,implicitcontract,andhoursrestrictionmodelsofthelabormarketusingmicrodata.AmericanEconomicReview,pages905{927,2002.[14]JamesJHeckman.Whathasbeenlearnedaboutlaborsupplyinthepasttwentyyears?AmericanEconomicReview,83(2):116{121,Sep1993.[15]MarkR.Killingsworth.LaborSupply.CambridgeSurveysofEconomicLiterature.PressSyndicateoftheUniversityofCambridge,1983.[16]AnjiniKochar.Smoothingconsumptionbysmoothingincome:hours-of-workresponsestoidiosyncraticagriculturalshocksinruralindia.TheReviewofEconomicsandStatis-tics,81(1):50{61,Sep1999.[17]WArthurLewis.Economicdevelopmentwithunlimitedsuppliesoflabour.TheManch-esterSchoolofEconomicandSocialStudies,XXII(2):139{191,1954.[18]AnnaMcCordandRachelSlater.Overviewofpublicworksprogrammesinsub-saharanafrica.Technicalreport,OverseasDevelopmentInstitute,2009.[19]BMeyer.Laborsupplyattheextensiveandintensivemargins:theeitc,welfare,andhoursworked.AmericanEconomicReview,Jan2002.[20]GeraldSOettinger.Anempiricalanalysisofthedailylaborsupplyofstadiumvendors.JournalofPoliticalEconomy,107(2):360{392,Dec1999.[21]ElainaRose.Exanteandexpostlaborsupplyresponsetoriskinalow-incomearea.JournalofDevelopmentEconomics,64:371{388,Jan2001.[22]MarkRosenzweig.Ruralwages,laborsupply,andlandreform:Atheoreticalandem-piricalanalysis.TheAmericanEconomicReview,pages847{861,Dec1978.38 Asecondconcernisthattherecouldbevillage-weekcorrelationinoutcomes.Thiscouldtaketheformofvillage-weekspecicshocks,suchasanillnessthataectsonevillageinasingleweek.Inthiscase,theresidualshavethestructure=tv+itvandvillage-weekclusteredstandarderrorsareappropriate.Ireportthesestandarderrorsincolumn(3).Thestandarderrorforthecoecientonlogwagesincreasesto0.029,forat-statisticof4.306.Analternativeapproachforaddressingvillage-weekcorrelationistoaggregateto120village-weekobservations.AngristandPischke[2]suggestshowingthatresultsarerobusttoanalysisatthegrouplevelwhenthenumberofclustersissmall.Sincetreatmentisatthevillagelevel,thisapproachalsomakesclearthesourceofvariation.Incolumns(4)and(5),thedependentvariableisthefractionofparticipantsineachvillagevwhoworkinweekt.IuseStata'saweightstoweightbythesquarerootofthenumberofparticipantspervillage.Thestandarderrorincolumn(4)isunadjusted,andthestandarderrorincolumn(5)isrobusttoheteroskedasticity.Asexpected,thestandarderrorsobtainedfromusingvillageaveragesarenotmuchdierentthantheclusteredstandarderrors,andconclusionsaboutthemagnitudeoftheelasticityofemploymentarerobusttogroup-levelanalysis.Athirdconcernisthattherecouldbevillage-levelcorrelationintheoutcomes.Villagelevelcorrelationcouldcomefrompersistentvillage-levelshocks,suchasanillnessthatstrikesinoneweekandlingersorhaseectsinsubsequentweeks,orcouldsimplybethatoutcomesinvillagesarecorrelatedbecausethepeoplewholiveinthesamevillagehavemanyunob-served(butnottime-invariant)characteristicsthataecttheiremploymentprobabilitiesincommon.Ineithercase,theresidualswouldhavethestructure=v+itv.Inthiscase,standarderrorsshouldbeclusteredatthevillagelevel.Thevillagelevelisalsothelevelofrandomization,andsincetheregressorofinterestvariesonlyatthegroupleveltheimpactofclusteringispotentiallylarge.Thestandarderrorsincolumn(6)areclusteredatthevillagelevel.Thestandarderrorofis0.035;thet-statisticforthetestthat=0is3.600,andthep-valueforthattestis0.006.Asexpected,clusteringincreasesthemagnitudeofthestandarderrors.However,thepointestimateofremainssignicantlydierentfromzerowhenusingclusteredstandarderrors.Therelativelysmallnumberofvillagesinmysamplemaybeproblematiciftherearepersistentvillage-levelshocks.Incolumn(7),Iallowforpersistentvillage-levelshocksandaddressthesmallnumberofvillagesbycalculatingthestandarderrorsfrom500block-40 Foreachvectorofresiduals^rh,IfollowCameronetal.'smethodofrandomlyswappingthesignofhalfoftheresidualsrhi,thencomputinganewpredictedoutcome^laboritvbyaddingtheresidualtotheobservedoutcomeforeachobservation.Ithenestimate^laboritv=~+~ln(wagetv)+uandtakethet-statisticforthetestthat~=H0h.Iobtain500t-statisticsforeachofthe101nullhypotheses.Ireportthe95percentcondenceintervalsfort-statisticsoftheteststhat=0ande=0incolumn(8).Reportingthestatisticforthetestof=0isthestandardconventioninregressionoutputandcorrespondstothesignicancelevelsfromblock-bootstrappedstandarderrorsthatIreportthroughoutthispaper.However,asdiscussedabove,theteststhatandespeciallyearezeroareperhapsnotthemostrelevantwhenestimatingtheelasticityofemployment.Insteadoftakingastandonthemostappropriatenullhypothesis,inFiguresA1andA2Iplottherejectionrate(t-statisticsbelow-1.96orabove+1.96)againsteachpossiblevalueof0ebetween0and1.Rejectionratesfromthewildbootstrapprocedurearelowestfornullhypothesesofandethatapproximatethecondenceintervalsfromtheclusteredorblock-bootstrappedstandarderrors.Mymainresultsarerobusttoadjustingstandarderrorstoallowforgenericheteroskedas-ticity,village-weekcorrelationinoutcomes,andvillagelevelcorrelationinoutcomes.Theresultsalsostanduptobootstrappingmethodsthattakeaccountofthesmallnumberofclustersinmydata.Theblock-bootstrappedstandarderrorsthatIusethroughoutthepa-perareconservativeintheirmagnitudeandaddressbothvillagelevelcorrelationinstandarderrorsandthesmallnumberofvillagesinthesample.42 TableA2:Elasticityofemploymentw.r.t.pastwages (1)(2)(3) (4)(5)(6)Weeks2to12Weeks5to12Dependentvariable:Individual*dayindicatorforworking Ln(wage)0.141***0.123***0.143*** 0.166***0.176***0.175***(0.036)(0.034)(0.033) (0.027)(0.031)(0.028)Ln(waget�1)-0.057**-0.029 0.0110.014(0.025)(0.026) (0.011)(0.016)Ln(waget�2) -0.005-0.004 (0.019)(0.017)Ln(waget�3) 0.0090.009 (0.018)(0.011)Ln(waget�4) -0.002-0.002 (0.018)(0.017) Weekeectsx xIndividualeectsx x Observations581358135813 422642264226Meanofdependentvariable0.840.840.84 0.890.890.89Elasticity0.170.150.17 0.190.200.20(0.044)(0.042)(0.041) (0.034)(0.038)(0.034) OLSestimates.Clusterbootstrappedstandarderrors(clusteredatthevillagelevel).Unitofobservationisindividual*week,sampleisallindividuals.*p0.10,**p0.05,***p0.001 TableA3:Elasticityofemploymentw.r.t.theaverageofpastwages (1)(2)(3)Sample:Weeks2to12Dependentvariable:Individual*dayindicatorforworking Ln(wage)0.141***0.152***0.140***(0.036)(0.035)(0.034)Ln( waget�1)0.075-0.143(0.065)(0.087) WeekeectsxIndividualeectsx Observations581358135813Meanofdependentvariable0.840.840.84Elasticity0.170.180.17(0.044)(0.043)(0.042) OLSestimates.Clusterbootstrappedstandarderrors(clusteredatthevillagelevel).Unitofobservationisindividual*week,sampleisallindividuals.*p0.10,**p0.05,***p0.001 44 TableA5:Dierentmethodsforcomputingstandarderrors (1)(2)(3)(4)(5)(6)(7)(8)Dependentvariable:Individual*dayVillage*dayIndividual*dayindicatorforworkingaverageemploymentindicatorforworking Ln(wage)0.127***0.127***0.127***0.128***0.126***0.127**0.127***0.127(0.010)(0.010)(0.029)(0.031)(0.030)(0.035)(0.033)t-statistic13.14912.5664.3064.0624.1523.6003.848[0.781,3.128]p-value0.0000.0000.0000.0000.0000.006Elasticity0.150.150.150.150.150.150.150.15(0.040)3.75[0.826,3.101] Observations633363336333120120633363336333Meanofdependentvariable0.840.840.840.850.850.840.840.84SEmethodnorobusttoclusteredcollapsedcollapsedclusteredblockwildadjustmentheteroskedasticityatvillage-weekleveltovillage-weekandrobustatvillagelevelbootstrappedbootstrapped OLSestimates.Unitofobservationisindividual*weekincolumns(1)to(3)and(6)to(8),andvillage*weekincolumns(4)and(5).Sampleisallindividuals.*p0.10,**p0.05,***p0.001 46 FigureA2:Rejectionratefornullhypothesesaboutefrombootstrap-tprocedure 48