【世界銀行】坦桑尼亞的貧困衡量：日記和回憶方法在食品消費(fèi)數(shù)據(jù)收集中的比較（英）

PolicyResearchWorkingPaper10871

MeasuringPovertyinTanzania

ComparisonofDiaryandRecallApproachestoFoodConsumptionDataCollection

AkuffoAmankwah

DarceyJeanneGenouJohnson

JosephineOforiAdofo

MaryamGul

AmparoPalacios-Lopez

WORLDBANKGROUP

DevelopmentEconomicsDevelopmentDataGroupAugust2024

ReproducibleResearchRepository

Averifiedreproducibilitypackageforthispaperisavailableat

,clickherefordirectaccess.

PolicyResearchWorkingPaper10871

Abstract

Consumptiondatafromhouseholdsurveyscontinuetobethemainsourceforpovertyandinequalitystatisticsinlow-andmiddle-incomecountries.Althoughrecentresearchhasdemonstratedthatthechoiceofdiary-versusrecall-basedmethodsforconsumptiondatacollectioncandirectlyimpactpovertymeasurement,theavailableevidencestemsfromsmall-scale,subnationalsurveyexperiments.ThisstudyusesdatafromanationallyrepresentativerandomizedsurveyexperimentinTanzaniatoprovideacomparativeassessmentofhowhouseholdconsumptionandpov-ertymeasuresmaybeimpactedbyrelyingona14-dayfoodconsumptiondiaryversustwodifferentvariantsof

7-day-recall-basedfoodconsumptiondatacollection.Theanalysisrevealssignificantdifferencesinfoodconsump-tionexpendituresacrossthediaryandrecallarms,andthesedifferencesresultindifferencesintotalconsumptionexpendituresaswell.Theresultsfurthershowthatthediarymethodcapturesmorediversefoodconsumptionitems,buttheoverallconsumptionexpenditureappearssignificantlylowerthanintherecallarms,evenatdifferentpercentiles.Despitethesedisparities,thepaperfindslittlestatisticallysignificantdifferenceinpovertyratesbetweenthediaryandrecallarms,evenatdifferentthresholds.

ThispaperisaproductoftheDevelopmentDataGroup,DevelopmentEconomics.ItispartofalargereffortbytheWorldBanktoprovideopenaccesstoitsresearchandmakeacontributiontodevelopmentpolicydiscussionsaroundtheworld.PolicyResearchWorkingPapersarealsopostedontheWebat

/prwp

.Theauthorsmaybecontactedataamankwah@.Averifiedreproducibilitypackageforthispaperisavailableathttp://,clickherefordirectaccess.

R

Y

C

I

L

R

A

E

S

E

O

P

H

C

S

TRANSPARENT

P

E

R

W

O

R

K

I

ANALYSIS

A

NGP

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

ProducedbytheResearchSupportTeam

MeasuringPovertyinTanzania:ComparisonofDiaryandRecallApproachestoFoodConsumptionDataCollection

AkuffoAmankwah1,DarceyJeanneGenouJohnson1,JosephineOforiAdofo2,MaryamGul1,AmparoPalacios-Lopez1

1LivingStandardsMeasurementStudy,WorldBankGroup,WashingtonDC,USA.

2AfricaChiefEconomistOffice,WorldBankGroup,WashingtonDC,USA.

KeyWords:Foodconsumptionexpenditure;Recalldesign;Diarydesign;Poverty;Inequality;Tanzania.

WBThematicAreas:Poverty;InequalityandGrowth.JELCodes:C83;D12

2

1.Introduction

Inmostlow-andmiddle-incomecountries,householdsurveyscontinuetobethemaindatasourceformeasuringandmonitoringpovertyandinequality.Atthecenterofpovertyandinequalitymeasurementsareincomeandconsumptionexpenditureapproaches(DeatonandZaidi2002;ManciniandVecchi2022).Giventhechallengesofaccuratelymeasuringincomeinhouseholdsurveysinlow-incomesettingswithahighproportionofthepopulationworkingininformalandseasonaljobs,theconsumptionexpenditureapproachhasdominatedpovertyandinequalitymeasurementsovertheyearsandcontinuestogarnerattentioninthedevelopmenteconomicsliterature.Acriticalcomponentofhouseholdconsumptionexpenditureisfood,whichisestimatedtocontributeabout40%oftotalhouseholdspendinginSub-SaharanAfrica(SSA)and17%inadvancedeconomies(Bogmansetal.2022).Thereisgrowinginterestinimprovingtheaccuracyandqualityoffoodconsumptionexpendituremeasurementinhouseholdsurveys(Beegleetal.2012;Christiaensenetal.2022;Battistinetal.2023;Brzozowskietal.2017),includingunderstandingthesourcesofmeasurementerrorsinthesevariables(Friedmanetal.2017).

Twomethodsaregloballyusedtogeneratehouseholdfoodconsumptiondata–diaryandrecall–thoughthedurationandreferenceperiodsvarybythecountriesadoptingeachapproach.Generally,themethodusedbyacountryisbasedontheirowndiscretionandisafunctionofwhattheyhavedoneinthepastandhavethehumancapitaltocontinuewithsuch.Whilethismayposenoproblemwithnationalpovertyandinequalityanalysisinrespectivecountriesacrosstime,internationalcomparisonbecomesdifficult,bothcross-sectionalandtemporal.Althoughcross-countrydifferencesintheseindicatorscouldbeattributedtoactualchangesinthestandardoflivingofthecountriesorvariationsinmethodology,orboth,internationalcomparisonsgenerallydonotconsidermethodologicaldisparities.

3

Thatnotwithstanding,thereisagrowingpushforharmonizingthemethodologyforproducingpovertystatisticstoensureregionalandglobalcomparability(ManciniandVecchi2022;Zezzaetal.2017),includingthetransitionfromdiarytorecallmethods,dueinparttothehighcostandhumancapitalrequirements(intheformofliteracyandsupervision)indiaryimplementation(Beegleetal.2012).TheEastAfricanCommunity(EAC)isonesuchregionalbodythatisrecommendingtomembercountriestousethe7-dayrecallmethodologyforfoodconsumptiondataaspartofabroaderefforttoharmonizestatisticsintheregioningeneral.ManciniandVecchi(2022)recommendthatanymethodologicalchange(e.g.,fromdiarytorecallorviceversa)shouldbeprecededbyanexperimenttoensurethatthereisnobreakinthepovertyseries.

TheseharmonizationeffortsandtheexperimentalrecommendationbyManciniandVecchiprovideaninterestingavenueforthisstudyinTanzania.SincetheworkofBeegleetal.(2012)onunderstandinghowdifferentmethodsofcollectingfoodconsumptiondatacouldimpactconsumptionaggregatesandultimately,livingstandardsmeasurements,severalotherstudieshavebeenconductedinthisarea(seefore.g.,Christiaensenetal.2022;Battistinetal.2023;Brzozowskietal2017).Thesestudiesgenerallyfocusonunderstandingmeasurementerrorsinfoodconsumptiondata,anddifferentapproachesforimprovingthequalityofmeasurement.Thereareotherstudiesthathavefieldedcountry-specificfoodconsumptiondatacollectionmethodstocomparethemwithpotentiallynewmethodstobeadoptedbytherespectivecountry(Durazoetal.2017;Backiny-Yetnaetal.2017).

ThisstudyaddstothegrowingbodyofliteratureonmeasuringhouseholdfoodconsumptionexpenditurebyusingnationallyrepresentativerandomizedhouseholdsurveydatafromTanzaniatocomparetwomethodsofdatacollection–recallanddiary.Thestudycompares

4

the14-daydiaryapproachusedintheHouseholdBudgetSurvey(HBS)andthe7-dayrecallapproachusedintheNationalPanelSurvey(NPS),bothsurveysimplementedbytheTanzaniaNationalBureauofStatistics(NBS)andtheOfficeoftheChiefGovernmentStatistician(OCGS)Zanzibar.Thisstudyisvitalandtimely,especiallyasregionalandglobalbodiespushfortheharmonizationofpovertystatisticstoensurecomparability.WhileBeegleetal.(2012)providegeneralguidanceonhowthediaryandrecallmethodswork,thedatausedisnotnationallyrepresentativeinnature.Thisstudybuildsontheirfindingsbutusesexistingmodulesandnewmethodsofdatacollection,followingtheevolutionofmethodsinthisarea.

WeusenationallyrepresentativehouseholdsurveydatafromarandomizedexperimentinTanzaniatoexploredifferencesinhouseholdfoodconsumptionexpenditurebetween14-daydiaryand7-dayrecalldesigns.Weemploybothdescriptiveandregressionanalyticalapproaches.Theresultsprovideevidenceofdifferencesinfoodandtotalconsumptionexpenditureacrossthediaryandrecallarms.Weobserveverylittlesignificantdifferencesinnon-foodconsumptionexpenditure,allowingforattributingvariationsintotalconsumptionexpendituretodifferencesinfoodexpenditureemanatingfromthedisparitiesinmethodologicaldesigns.Theresultsshowthatwhilethediarymethodcapturesmorediversefoodconsumptionitems,theoverallconsumptionexpenditureappearssignificantlylowerthantherecallarms,evenatdifferentpercentiles.Povertyratesvariedsignificantlyacrossthedesignsandareverymuchconditionalonthepovertylinechosen.

Therestofthepaperisorganizedasfollows.InSection2wepresentcountrycontextforthestudy.Section3providesadescriptionofthedataandstudydesign.Section4presentstheanalyticalframeworkemployedtogeneratetheresultsofthestudy,whileSection5containsthediscussionofresults.TheconclusionandimplicationsofthestudyarepresentedinSection6.

5

2.Countrycontext–macroeconomicindicatorsandrecentpovertysurveysinTanzania

Tanzaniahasexperiencedsubstantialimprovementsinmacroeconomicindicatorsinrecentyears.Atthenationalthreshold,povertyheadcounthasdecreasedfrom34.4%in2007,toabout26.4%

in2018,anditstoodat44.9%in2018usingthe2017internationalpovertylineofUS$2.15.

PovertyinTanzania,likeothercountriesintheregion,ispredominantlyrural.SimilartootherSSAcountries,foodconstitutedabout59.9%oftheconsumptionbasketin2017inTanzania,about4percentagepointsincreasefrom2011,withruralandurbansharesbeing63.2%and52.6%,respectively(MinistryofFinance,NBS,andWorldBank2020).FoodsecurityisamajordevelopmentalchallengeinTanzania.TheFAO(2021)estimatesthatabout34.4millionTanzaniansweremoderatelyorseverelyfoodinsecurein2021,anincreasefrom25.1millionin2014.Asanagrarianeconomy,theagriculturesectorcontinuestobecriticaltoTanzania,especiallytheruralpoor.Thesectorcontributedabout30%toGDPin2021(NBS2022a),employednearly62%ofthepopulationin2021(NBS2022b),andgeneratedabout30%offoreignexchangeearnings(GovernmentofTanzania2017).

Theapexgovernmentagencies,NBSinMainlandandOCGSinZanzibar,aremandatedwiththeproductionofofficialstatistics,includingpovertyandinequalityinthecountry.TheNBSandOCGShave,inthepast,independentlypublishedofficialpovertystatisticsfortheirjurisdictionsusingdatafromtheHBSwherefoodconsumptioniscapturedusingdiaries.BesidestheHBS,theNBSandOCGSalsoimplementtheNPSaspartoftheLivingStandardsMeasurementStudyIntegratedSurveyonAgriculture(LSMS-ISA)project,whichuses7-dayrecallforfoodconsumptiondatacollection.

BoththeNPSandHBSaremulti-topichouseholdsurveys.ThoughtheHBSandNPSareimplementedbytheNBSandOCGS,thefocusofeachsurveyisdifferent–HBSprovidesdata

6

forofficialpovertystatisticswhiletheNPShelpsmonitorwelfaredynamicsinthecountry–andtheyarealwaysimplementedseparatelyandindifferentyears.TheHBSisimplementedevery5years,whiletheNPSisimplementedmorefrequently.Historically,theHBSuseda30-daydiaryuntil2017/18whentheinstitutionsswitchedtoa14-daydiary.ThenatureofadiaryimplementationmeansthattherespondentisnotrestrictedbyanitemlistintheHBS,whileintheNPStherespondentislimitedtoalistofbothfoodandnon-fooditems(thoughvariationsofsmallerandlargerlistshavebeenutilizedintheNPS).Bothsurveysarecollectedovera12-monthperiod,tocaptureseasonality.

3.Dataandstudydesign

3.1.Data

ThispaperusesdatafromtheTanzaniaHouseholdConsumptionMethodologicalstudyconductedinMay-June2022bytheNationalBureauofStatistics(NBS)andOfficeoftheChiefGovernmentStatistician(OCGS)Zanzibar,withfundingandtechnicalassistancefromtheWorldBank.Themethodologicalstudyadoptedamulti-stagesamplingapproachdesignedtoprovideestimatesatthenationallevel,aswellasruralandurbanstratification.Thefirststageinvolvedtheselectionof143enumerationareas(EAs)spreadacrossallregionsoftheUnitedRepublicofTanzania(bothMainlandandZanzibar)usingaprobabilityproportionaltosize(PPS)approach.TheselectionensuredtheinclusionofbothurbanandruralEAsineachregion.OncetheEAswereselected,afreshhouseholdlistingwasconductedtogetthetotalhouseholdcompositionineachoftheEAs.Followingthelistingexercise,15householdswereselectedusingasystematicrandomsamplingprocedureforinclusioninthestudy,with5householdsassignedtoeachofthethreearmsofthestudy.Intheend,atotalof2,119householdswereincludedintheanalysis.Thefinalsamplewas

7

weightedtoadjustfornon-responseandothersamplingerrors.SeeTable0fortheresponseratesbytreatment.

ThesurveywasimplementedusingtheWorldBank’sSurveySolutionssoftwareforComputerAssistedPersonalInterviews(CAPI).Thisallowedfordailyuploadofcompletedinterviewsbythefieldteams,aswellasfrequentdatadownloadandreviewbytheNBSandOCGSheadquartersteamstoensurethequalityofthedatabeingcollected,whileprovidingfeedbacktotheenumeratorsforfurtherqualitycontrol.

3.2.Studydesign

Threeapproachestocollectinghouseholdfoodconsumptiondata–14-daydiary,7-dayrecallwithshortitemslist,and7-dayrecallwithlongitemslist–arecomparedinthisstudy.Tobetterunderstandthedifferencesinhouseholdfoodconsumptiondatacollection,thismethodologicalstudyreplicatedtheHBS2017/18(14-daydiary)andtheNPS(7-dayrecallwithshortlist)methodologies,aswellasavariant/combinationoftheNPS(7-dayrecallwithlonglist).GiventhattheHBSisthesurveythatNBSandOCGSusetoproduceofficialpovertystatistics,thediarymethodologyservedasthebenchmarkandhouseholdsassignedtothisgroupservedasthecontrolgroup,andthetwo7-dayrecallarmsarecomparedtothediary.Inaddition,allhouseholdsinthestudyreceivedaFoodAwayfromHome(FAFH)moduleimplementedattheindividualleveltocaptureexpenditureonfoodconsumedoutsidethehome,suchasfromrestaurants,joints,etc.Householdswererandomlyassignedtoeachstudyarminallclusterstoensurenon-contamination(Beegleetal.2012;Brzozowskietal2017).Thedesigndetailsofeachmethodaredescribedbelow.

8

3.2.1.Treatment0:14-daydiary

Thedesignandimplementationofthe14-daydiary(henceforth,diaryorT0)inthisstudyfollowedexactlythemethodologyinthe2017/18HBS(seeMinistryofFinance,NBS,andWorldBank2020).Themethodologyrequiredadministeringthe14-daydiaryaspartofamulti-topicquestionnairewithmultiplevisitstothehousehold.Duringthefirstvisit,theenumeratorintroducedthefoodconsumptiondiarytothehousehold,gavethemtrainingonhowtocompleteit,andleftitwiththehousehold/individualstocompleteonadailybasis.Dailyexpenditureinformationonfooditemspurchased,thequantitypurchasedandcorrespondingunit(includingnon-standardunits),valueofthequantitypurchased,sourceanddestinationofthequantitypurchasedwererecorded.Inaddition,householdswereaskedtorecordseparatelytheactualdailyfoodconsumption,aswellasFAFH.Theenumeratorrevisitedthehouseholdatdifferenttimesduringthe14daystosuperviseandtransfertheinformationfromthedailyrecordbookintothemainCAPIquestionnaire.Attheendofthe14days,theenumeratorrevisitedthehousehold(onthe15thday)toclosethediarywiththem.Oncetheinformationwastransferredintothemainquestionnaire,COICOPcodeswereassignedtoeachofthefooditemsthatthehouseholdindicatedtohaveconsumed/purchased.Theothermodulesinthemainquestionnairewereadministeredanytimebetweenthefirstandthelastvisittothehousehold.Thus,thenumberoffooditems,thenumberandphrasingofquestions,andmethodologyofdiaryimplementationinthisstudy(includingfrequencyofenumeratorsvisitstothehousehold)wereexactlythesameasthoseusedinthe2017/18HBS.AseparateFAFHmodulewasalsoimplementedattheindividuallevelforthisgroup.Foodconsumptionaggregateswerethengeneratedusingthequantityconsumedandpriceinformationcollectedinthismodule,whileadjustingfortemporal(withinsurveymonths)

9

andspatialpricevariabilities.Residentenumerators(stationedintheEAs)wereusedtoimplementthisarm,thusmakingdiaryimplementationandsupervisionmorefeasible.

3.2.2.Treatment1:7-dayrecallwithshort-listoffooditems

Thefoodconsumptionmoduleinthisarm(henceforth,recallshortlistorT1)wasimplementedaspartofamulti-topicquestionnaireina“singlevisit”tothehousehold.Duringtheinterview,householdswereaskedtorecalltheirfoodconsumptionoverthe7daysprecedingthedayoftheinterview(excludingthedayofinterview).Atotalof75fooditemswereincludedinthismodule.Therespondentofthefoodconsumptionmoduleisgenerallyanadultfemalehouseholdmemberresponsibleforfoodpurchasesandpreparation.TherespondentisfirstpresentedwithaYES/NOfilterquestionforall75fooditemsincludedinthemodule.Forthoseitemsthattherespondentindicatedtheirhouseholdconsumedwithinthelast7days(YESresponseinthefirstfilterquestion),additionalquestionsonthequantityconsumed,sourcesofconsumption(purchased,ownproduction,andgifts),andtotalvalueofpurchasesifanyofthehousehold’sfoodconsumptioncamefrompurchases.

Informationonquantityconsumedwascollectedonlyinstandardunits(exceptforeggswhere“piece”wasallowed).Inadditiontofoodconsumedwithinthehousehold,informationonFAFHattheindividuallevelwasalsocollected.Foodconsumptionaggregateswerethengeneratedusingthequantityconsumedandpriceinformationcollectedinthismodule,whileadjustingfortemporal(withinsurveymonths)andspatialpricevariabilities.Mobileteamsthatoperateinarovingmannerwereusedfordatacollectioninthisstudyarm.

10

3.2.3.Treatment2:Modified7-dayrecallwithlonglistoffooditems

Thismodule(henceforth,recalllonglistorT2)wasdesignedtofollowthemostwidelyused7-dayfoodconsumptionrecallincountriesthatarepartoftheLSMS-ISAproject.Householdsinthisgroupwereadministeredamodifiedversionofthe7-dayfoodconsumptionrecallmoduleinT1.Themodulecaptures7-dayfoodconsumptionrecallofquantityconsumed,andthesourcesofthefoodconsumed(purchases,own-productionandgifts).Householdswerealsoaskediftheypurchasedagivenfooditeminthelast30days,andwhereapplicable,thequantityandvalueoftheirmostrecentpurchasewithinthelast30dayswerecollected.Respondentswereallowedtoprovidequantityconsumedandpurchasedinnon-standardunits(NSU),andenumeratorsusedNSUphotoalbumswhereapplicable.

GiventheunavailabilityofconversionfactorsforsomeoftheNSUs,effortsweremadetoonlyallowrespondentstoprovideresponsesinunitsforwhichconversionfactorswereavailable.Inaddition,forthoseitemswhererespondentsreportedinNSUs,theywereaskedtoprovideanestimateofthatinstandardunits(kilograms,grams,liters,orcentilitersasapplicable)aswell.Alonglistoffooditems(167)thatconstitutedthelargestshareofconsumptioninHBS2017/18wereincludedinthisgroup.FAFHattheindividuallevelwasalsoimplementedforthisgroup.Foodconsumptionaggregateswerethengeneratedusingthequantityconsumedandpriceinformationcollectedinthismodule,whileadjustingfortemporal(withinsurveymonths)andspatialpricevariabilities.SimilartoT1,mobileteamsthatoperateinarovingmannerwereusedfordatacollectioninthisstudyarm.

11

3.3.Additionalimplementationconsiderations

3.3.1.Questionnairedesign

Inadditiontothetreatment-specificfoodconsumptionmodulesdescribedearlier,allhouseholdsincludedinthethreetreatmentgroupsalsoansweredquestionsonhouseholdroster,education,employment,non-foodconsumption(7-day,30-day,and12-monthreferenceperiods),householdparticipationinagriculture(cropsandlivestock),asset,labor(anabridgedversion),credit,groupformation,housing,cognitiveskills,andotherrelevantbalancingvariables.Foreachofthemodulesincludedinthequestionnaire,informationwascollectedonthemainrespondent,aswellasthestartandendtimes.

3.3.2.Ensuringcomparability

Giventhatdifferentsetsofteamarrangementswereusedforthediaryandrecallcomponentsofthestudy,andthedifferentreferenceperiods,itwasvitaltoensurethattherewasoverlapinthereferenceperiodbetweenthediaryandtherecallarms.IneachEA,themobileteamresponsibleforthatEAliaisedwiththeirresidentenumeratorcounterparttoensuretheyvisittheEAandstartadministeringthefoodconsumptionmoduleatleast7daysafterthediaryhasbeenopenforthediaryhouseholdsinthatEA.Thisensuredanoverlapinthedaysforthefoodconsumptiondiaryandtherecallmodulesimplementation.ThiswasmadepossiblesincetheEA-specificresidentenumeratorandmobileteamhadthesamesupervisor.

3.4.Constructionofconsumptionaggregatesandtreatmentofoutliers

Theconsumptionaggregateswereconstructedinanidenticalmannerandguidedbythesametheoreticalandpracticalconsiderationsforallthreedesigns.Thisincludesthecomponents,the

12

treatmentofoutliers,theinclusionofbenefitsbeyondfoodandnon-food(i.e.,rent,housing,durables),andadjustmentsforbothadultequivalent,andspatialandtemporalpricedifferences.Forallthreedesigns,thefinalconsumptionaggregateisrepresentedbythesumofallitscomponents:i)foodconsumedinsidethehome–whetherpurchased,self-produced,orreceivedin-kindasgift;ii)foodconsumedoutsidethehome;iii)non-foodexpendituresofvariousreferenceperiods–oneweek,onemonths,12monthsinourcase;iv)educationexpenditure;v)healthexpenditure–excludinghospitalizationcosts;vi)theimputedvalueofconsumerdurables;andvii)theimputedvalueofdwellingbenefits.

Constructionofthefoodcomponentwasguidedbythreeprinciples.First,thatallpossiblesourcesoffoodconsumptionwereincluded(i.e.,purchasesinmarket,producedbyhousehold,receivedasgiftorpayment,etc.).Second,non-purchasedfoodthatwasconsumedmustbeassignedavalue.Finally,onlyfoodthatwasconsumed–notnecessarilythetotalamountoffoodpurchased–isincluded.Non-foodconsumptionisslightlymoredifficulttomeasure,andwereliedonafewwidelyacceptedassumptionstodoso.First,thatthecostsofacquisitionofanon-fooditemisequaltoitsconsumption,andsecond,thatthevalueofconsumptionofnon-foodgoodsandservicesfromownproductionorgiftsisnegligible.Finally,unlikefood,non-foodexpenditureswerevaluedatthepurchaseorself-reportedacquisitioncost,asmostnon-fooditemsaretooheterogeneoustocollectquantitiesandcalculateprices.Expenditureonsomenon-foodgoodsandserviceswasexcludedfromthecalculationoftheconsumptionaggregate,includingpaymentsofmortgagesordebts(financialtransactions),lossestotheft(neitherconsumptionnorexpenditure),expenditureonmarriages,dowries,births,andfunerals(tooirregular).Giventhatconsumptiondatafromhouseholdsurveysisbothextensiveandcomplex,dataerrorsandoutliersareexpectedregardlessofdatacollectionmethod.Thefollowingruleswereappliedtothemanagementof

13

outliersforallthreedesigns:i)outliersweredealtwithsimilarlyacrosstreatmentarms;ii)outlierswereidentifiedattheitemandrural-urbanlevelforgreatestaccuracy;andiii)outliersweredefinedasfallingoutsideoftwopredeterminedtoleranceintervals(thevariable-specificinterquartilerange(IQR)andthreetimesthestandarddeviationfromthelogarithmicmean).Forbothcostandquantityvalues,moderateoutlierswerepreservedwhileextremeoutlierswerereplacedwiththestrata-levelmedianvalueorsomefactorthereof,i.e.,forlargeoutliersthevaluewasreplacedwithfourtimesthestrata-levelmedianvalue.Allconsumptionexpenditurecomponentsareconvertedtoonemonthusinganappropriateadjustmentfactordependingonthereferenceperiodandthenaggregatedtogeneratethefinalmonthlyconsumptionaggregate.

4.Analyticalapproach

Thestudyusesbothdescriptiveandregressionapproaches.Thedescriptiveapproachinvolvesbasicttestsandtheconstructionofdensityfunctionstounderstanddifferencesinconsumptionexpendituresandtreatments.Followingthedescriptiveanalysis,weestimatetwosetsofOrdinaryLeastSquares(OLS)regressions.Inthefirstspecification,werunthelogarithmoffoodconsumptionexpenditureandlogtotalconsumptionexpenditure(bothfoodandnon-food)ondummiesforthetworecallarms(longandshort),usingthediaryasthecomparisoncategory.Weestimatethefollowingmodel:

yij=βjDj+eij(1)

where,yijisthelogarithmoftotalpercapitafoodconsumptionexpenditureandlogtotalpercapitaconsumptionexpenditureofhouseholdifortreatmentj.Djisavectorofdummyvariablesfortreatmenttypesj=0,1,2(0forthediary,1fortherecallshortlist,and2fortherecalllonglist).

14

βjcapturesthedifferencesinexpenditureofthe