The impact of growth and inequality on rural poverty in China

www.elsevier.com/locate/jce

Theimpactofgrowthandinequality

onruralpovertyinChina

YinZhanga,∗,GuanghuaWanb

aDepartmentofEconomicStudies,UniversityofDundee,3PerthRoad,Dundee,DD14HN,UK

bUNU-WIDER,Helsinki,Fin-00160,Finland

Received8August2006Availableonline26September2006

Zhang,Yin,andWan,Guanghua—TheimpactofgrowthandinequalityonruralpovertyinChinaThispaperanalyzestheevolutionofpovertyinChinafromthelate1980stothelate1990s,employingaversionofShapleydecompositiontailoredtounit-recordhouseholdsurveydata.Thechangesinpovertytrendsareattributedtotwoproximatecauses—incomegrowthandshiftsinincomedistribution.Differentdatasets,povertylines,povertymeasures,andequivalencescalesareusedtoexaminetherobustnessoftheresults.Potentialbiasesarisingfromignoringdifferentialregionalpricesandinflationarealsoinvestigated.Notwithstandingsomeambiguitiesintheresults,itisconsistentlyfoundthatruralpovertyincreasedinthesecondhalfofthe1990sandadversedistributionalchangesarethemaincause.JournalofComparativeEconomics34(4)(2006)694–712.DepartmentofEconomicStudies,UniversityofDundee,3PerthRoad,Dundee,DD14HN,UK;UNU-WIDER,Helsinki,Fin-00160,Finland.

©2006AssociationforComparativeEconomicStudies.PublishedbyElsevierInc.Allrightsreserved.

JELclassification:O15;O53

Keywords:Ruralpoverty;Shapleydecomposition;China

1.Introduction

EastAsiaisthefirst,andremainstheonly,regionwherethefirsttargetoftheMillenniumDevelopmentGoals(MDGs)—halvingextremepovertybetween1990and2015—hasbeenmet(UNMillenniumProject,2005).WhilethedramaticpovertyreductionsinChinaareessential

*Correspondingauthor.Fax:+441382384691.

E-mailaddress:y.x.zhang@dundee.ac.uk(Y.Zhang).

0147-5967/$–seefrontmatter©2006AssociationforComparativeEconomicStudies.PublishedbyElsevierInc.Allrightsreserved.

doi:10.1016/j.jce.2006.08.008

Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712695

tothatattainment,asof2001aroundonefifthoftheworld’spoorstillresidethere(ChenandRavallion,2004).WithinChina,ruralpovertydominatesthescene.Theoverallreductionofruralpovertysincetheinitiationofmarket-orientedreformhasbeennothingbutimpressive.Notwithstanding,somerecentresearchsuggeststhattheprocesssloweddownsignificantlyinthe1990sandhasevenshownsignsofreversal(ChenandRavallion,2004).SuccessinbattlingruralpovertyclearlyholdsthekeytofurtherprogresstowardpovertyalleviationinChinaandtheachievementoftheMDGpovertytargetonaglobalscale.ItisthusimportanttounderstandwhatdrivesthelatestpovertytrendinruralChina.

TheagriculturalsectoriswhereChina’sremarkablegrowthstorystarted.Bysimplyde-collectivizingproductionandallowingfarmerstoselltheirsurplusproduceonthemarket,Chinapropelledtheannualgrowthrateofagriculturefromanaverageof2.5percentin1952–1977to7.4percentin1978–1984.Asaresult,ruralpercapitaincomerosebyastunning270per-cent.However,theagriculturalsectormoveddownthegovernment’sprioritylistaroundthemid-1980swhenthefocusofreformwasshiftedtotheurbanareaandindustrialsector.Notonlyweregovernmentprocurementpricesoffarmproducesetbelowmarketprices,buttheyalsofailedtokeepupwithpriceincreasesinothersectorsduring1989–1993(Fig.1).Agricul-turalgrowthsloweddown,andthegrowthofgrainoutputsufferedparticularlyseverely.In1993,acombinationoffloodsanddroughtledtofoodshortage.Attemptstoraisegrainprocurementpricestomarketlevelsin1994triggeredasharpincreaseingrainprices.Concernedaboutin-flationpressuresandgrainself-sufficiency,thegovernmentrespondedbyincreasinginvestmentinruralinfrastructureandreassertingcontrolovertheproductionandmarketingofseveralba-siccommodities.Inaddition,agovernor’sgrainbagresponsibilitysystemwasimplementedin1995,whichmadeprovincialgovernorspersonallyresponsibleforensuringadequatesupplyof

Fig.1.ChangesinruralCPIandprocurementpriceindexoffarmproduce.

696Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712

domesticgrainwithintheirjurisdictions.Thesepoliciessucceededinstabilizingfoodpricesandincreasinggrainreserves.Inthelate1990s,governmentcontrolsovertheimportsofagriculturalproductsweregraduallyeasedasChinapreparedforentryintotheWTO.Increasedimportsalongwithconsecutiveyearsofgoodharvestboostedfoodsupply.Thedemandforfood,how-ever,didnotriseasfast,partlyduetoadecliningshareoffoodinhouseholdexpenditureinducedbyhigherincome.Foodpricesslumpedandruralincomestagnated.

Theaboveagriculturalpolicieshaveimplicationsforruralincomegrowthaswellasforitsdistribution.ComparingthetrajectoriesofpovertyreductionandGDPgrowthafterthelate1980swiththoseintheearlieryears,anumberofstudieshaveemphasizedtheroleofrisingin-comeinequalityinslowingdownpovertyreduction.Theseinclude,amongothers,Khan(1999),GustafssonandWei(2000),Yao(2000),ChenandWang(2001),andRavallionandChen(2004).Givenapovertyline,anypovertytrendcanalwaysbeattributedtoincomegrowthandshiftsinthedistributionofincome.Thedecisionfacingpolicymakersisoftenoneofallocatinglimitedresourcesbetweengrowthpromotionandredistributiontowardthepoor.ThispaperseekstoanalyzethecontributionstoChina’sruralpovertytrendattributabletoincomegrowthanddistri-bution.ThreefeaturessetthispaperapartfrompreviousstudiesonChina’spoverty.Firstly,moststudiesrelyonhouseholdsurveydatafromtheNationalBureauofStatistics(NBS).TheNBSdatafrompublishedsourcesareingroupedformat.Estimatingpovertymeasuresfromgroupeddatanecessarilyinvolvesinterpolationinordertogeneratedatabetweengroupboundaries.1Thisentailserrorsunlesstheinterpolationmethodhappenstoagreewiththeunderlyingincomedistri-bution.2Besides,theNBSdatahavebeencriticizedfortheirexclusionofsuchimportantincomeitemsasimputedrentsofowner-occupiedhousing,subsidies,incomeinkind,andsoon(Khan,1999).ExploringalternativedatasourcescanthereforeservetheusefulpurposeofcheckingtherobustnessofresultsbasedontheNBSdata.Thetwodatasetsemployedinthispaperaresuchexamples,bothofwhichprovidehousehold-levelincomedataandarefromsourcesdistinctfromtheNBShouseholdsurvey.3

Theseconddepartureofthispaperliesinourmethodofquantifyingtherelativeimpor-tanceofgrowthanddistributionalchangesinformingpovertytrend.Becauseincomegrowthanddistributionalchangesareinterrelated,theirunconditionalcorrelationwithpovertychangesisuninformativeoftheirmarginalimpactonpoverty.Somestudiesregressthelogarithmsofapovertyindexonthoseoftheaverageincomeandanaggregateinequalitymeasure,typicallytheGiniindex,toobtainthemarginalimpactsofgrowthandredistribution(e.g.,Khan,1999;Yao,2000;RavallionandChen,2004).Implicitly,thisassumesthattherelationshipamongthethreevariablesisapproximatelylog-linear.Theaccuracyofthisapproximationaside,4amajorproblemwiththisapproachisthattheGiniindexuniquelydeterminestheLorenzcurveonly

1Intherarecaseofthepovertylinecoincidingwithoneoftheincomegroupboundaries,groupeddatalendthemselves

readilytothecalculationofpovertymeasuressuchasthefirstthreeindicesoftheFoster–Greer–Thorbecke(FGT)family.2Someresearchersgainedaccesstohousehold-levelNBSdata,butthesearetypicallyforisolatedprovincesandoflimitedyears(e.g.,Yao,2000).TheexceptionisarecentWorldBankstudy(RavallionandChen,2004),whichhasmeticulouslyassembledfromhousehold-levelNBSdatathe1981–2001seriesofthreeFGTindicesbothatthenationallevelandseparatelyforurbanandruralareas.

3Anotherdatasetfeaturingunit-recordincomedataistheChinaHouseholdIncomeProject(CHIP)dataset.ItcontainsinformationfromtwohouseholdsurveysconductedbytheChineseAcademyofSocialSciences(CASS)inselectedprovincesin1988,1995,and2002.

4Iftheregressionhasapoorfit,muchofthechangesinpovertywillbeassignedtotheresidualtermandthusgounaccountedfor.

Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712697

underrestrictedconditions.5AgivenchangeintheGiniindexmaybecausedbyredistributionamongthenonpoor,amongthepoor,orbetweenthepoorandthenonpoor.Inthefirstcase,povertywillnotbeaffectedatall.Povertychangesinthelattertwocaseswillalsodiffer.Hence,theregressioncoefficientontheGiniindexdoesnotidentifytheeffectsofdistributionalchangesonpovertyasitispurportedto.Another,andalsomoreappropriateinourreview,approachtopinningdowntherelativecontributionofgrowthanddistributionalchangesisthatproposedbyDattandRavallion(1992)todecomposeachangeinapovertymeasureintogrowthandredis-tributioncomponents.6ThispaperadoptstheDatt–Ravalliondecompositionmethodologywithtwoextensions.First,wedrawontheShapleyvaluedecompositionframeworkpropoundedbyShorrocks(1999)tomakethedecompositionsymmetricandexact.Second,intheDatt–RavalliondecompositiondistributionalchangesareidentifiedwithchangesintheestimatesofLorenzcurveparameters.Weshallshowthat,withmicroleveldata,decompositioncanbeconductedwithoutresortingtoparametricLorenzfunctions.

Finally,measurementissuesloomlargeinpovertyresearch,especiallyforstudiesabouttran-sitionanddevelopingeconomies.Apartfromtheproblemswiththeavailabilityandqualityofhouseholdsurveydata,thereareoftenambiguitiesabouttheappropriatepovertyline,equiv-alencescale,povertymeasure,andpriceindextouse.Assomearbitrarinessisinevitableinmakingthechoice,weexaminethesensitivityofpovertytrendanddecompositiontoalternativemeasurementassumptionsratherthanfocusingexclusivelyonaparticularoneofthem.

Theremainderofthepaperisorganizedasfollows.Thenextsectionexplainsthedecom-positionmethodologyanddiscussesvariousuncertaintiesinvolvedinassessingpovertytrendanddecomposition.Section3introducesthedataandpresentsthetimeprofilesofthreepovertymeasuresoftheFoster–Greer–Thorbecke(FGT)family.InSection4,wediscussthedecompo-sitionresults.Particularattentionisgiventoresultsthatareconsistentacrossdifferentdatasetsandpovertymeasuresandunderalternativeassumptionsaboutthepovertylineandequivalencescale.SomeconcludingcommentsareofferedinSection5.

2.Growth–redistributiondecompositionandtherobustnessofpovertymeasurementApovertymeasurePisafunctionoftheincomedistributionYandpovertylinez,i.e.,Pt=P(Yt,zt).Ifthepovertylineisheldconstantovertime,achangeinpovertybetweenperiod0andperiodTcanbewrittenas

󰀃P=P(Yt)−P(Y0),

(1)

wheretheletterzrepresentingthepovertylineisdroppedforsimplicity.7Thebasicideabehindthegrowth–redistributiondecompositionisthat,atanypointoftimet,theincomedistributionYtcanalwaysbefullydescribedbyitsmeanincomeμtandLorenzcurveLt,thelatterofwhichisuniquelydeterminedbytheprobabilitydensityfunction(PDF)ofrelativeincome.8Thus,

5Ifthesizedistributionofincomeislog-normal,forexample,thenthereisaone-to-onerelationshipbetweentheGini

indexandtheLorenzcurve.However,empiricalevidenceshowsthatthelog-normaldistributiondoesnotdescriberealincomedatawell(McDonald,1984).

6ForapplicationsoftheDatt–RavalliondecompositiontotheChinesecontext,seeFangetal.(2002)andChenandWang(2001).

7Changesinthepovertylinecanbeeasilyaccommodatedintheframeworkdescribedbelow.8Weusetheterm‘relativeincome’torefertoincomevaluesnormalizedbythemeanincome.

698Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712

󰀃PstemsfromchangesineitherofthetwodeterminantsofY.Ifincomegrowthisdistribu-tionneutral,ortheincomeofeveryindividualgrowsbythesameproportion,thentheLorenzcurve(or,equivalently,thePDFofrelativeincome)willstayunchangedand󰀃Pisdueen-tirelytochangesinthemeanincome.Conversely,whenthemeanincomeneithergrowsnorcontracts,achangeinpovertywilloccurifandonlyiftheLorenzcurveshifts,i.e.,thereisincomeredistributionamongsomeindividuals.Appliedtothegeneralcasewithbothincomegrowthanddistributionalchanges,theprecedingreasoningimpliesthat󰀃Pcanbeseparatedintotwocomponentsmeasuringrespectivelythegrowthanddistributionaleffects.Toexpressthetwocomponentsmathematically,letY(μi,Lj)beahypotheticalincomedistributionwithmeanincomeμiandLorenzcurveLjtakenfromdifferentdistributions,i.e.,i=0orT,j=0orT,andi=j.AndletP(μi,Lj)representthepovertylevelcorrespondingtoY(μi,Lj).Thegrowthcomponentof󰀃Pcanbedefinedas

growthcomponent≡P(μT,L0)−P(Y0),or,alternatively,as

growthcomponent≡P(YT)−P(μ0,LT).

Similarly,theredistributioncomponentcaneitherbedefinedas

redistributioncomponent≡P(μ0,LT)−P(Y0),oras

redistributioncomponent≡P(YT)−P(μT,L0).

(3A)(3)(2A)(2)

Fourdecompositionsof󰀃Pcanbeformedbycombiningthealternativedefinitionsofthegrowthandredistributioncomponentsdifferently.Ifdefinitions(2)and(3)areused,period0istheref-erenceperiod;if,instead,definitions(2A)and(3A)arechosen,thereferenceperiodisperiodT.Theresultsfromthetwodecompositionsneednotagree,andbothareinexactinthatthegrowthandredistributioncomponentsdonotaddupto󰀃P.Ifthecombination(2)and(3A)or(2A)and(3)isused,thedecompositionwillbeexactsince

P(YT)−P(Y0)=growthcomponent+redistributioncomponent

󰀄󰀃󰀄󰀃

=P(μT,L0)−P(Y0)+P(YT)−P(μT,L0)

󰀄󰀃󰀄󰀃

=P(YT)−P(μ0,LT)+P(μ0,LT)−P(Y0).

(4)(5)

However,thegrowthandredistributioncomponentsinexpressions(4)and(5)aremeasuredagainstdifferentreferenceperiods.Again,thetwodecompositionsproducedifferentresultsingeneral,andthusareequallyarbitraryorequallyjustified.

Thedecompositionmethodsusedinpreviousstudies,suchasthoseofDattandRavallion(1992),KakwaniandSubbarao(1990),andJainandTendulkar(1990),essentiallycompriseoneortwooftheabovedecompositions.Hence,theyaresensitivetothechoiceofthereferenceperiod,andareinexactorhaveanonvanishingresidualtermwhenevergrowthanddistributionalchangesarebothpresent.Asolutiontotheseproblemscreatedbythereferencepointistotaketheaverageofexpressions(4)and(5)toarriveat

󰀅󰀃󰀄󰀃󰀄󰀆

󰀃P=0.5P(μT,L0)−P(Y0)+P(YT)−P(μ0,LT)

󰀄󰀃󰀄󰀆󰀅󰀃

+0.5P(YT)−P(μT,L0)+P(μ0,LT)−P(Y0).(6)

Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712699

AsarguedinShorrocks(1999)andKolenikovandShorrocks(2005),thedecompositioninex-pression(6)isnotanarithmeticgimmick,buthasitstheoreticalrootsinthecooperativegametheory.Apartfromnotationaldifference,expression(6)isidenticaltowhatShorrocks(1999)derivedusingtheShapleyvalue.ThegrowthcomponentGandtheredistributioncomponentRoftheShapleyvaluedecompositionof󰀃Parethus

󰀄󰀃󰀄󰀆󰀅󰀃

G≡0.5P(μT,L0)−P(Y0)+P(YT)−P(μ0,LT),(7)

󰀄󰀃󰀄󰀆󰀅󰀃

R≡0.5P(YT)−P(μT,L0)+P(μ0,LT)−P(Y0).(8)ItcanbeeasilyseenthattheShapleydecompositionissymmetricandexact.

ThereremainsthequestionofhowtoobtainthepovertyindicesP(μT,L0)andP(μ0,LT)ofthehypotheticaldistributions.ThemethodusedinpreviousstudiesistoassumeaparametricLorenzcurveorPDF.Then,theformulaforthepovertymeasureasafunctionofthemeanincomeandparametersoftheLorenzcurveorPDFisderived.Theparametersareestimatedeconomet-ricallyforbothperiods0andT.Pluggingintothederivedformulatheparameterestimatesforperiod0andmeanincomeofperiodTgiveP(μT,L0).P(μ0,LT)isobtainedsimilarly.TheweaknessofthisparametricprocedureisthatthespecificationandestimationoftheLorenzcurveorPDFcangiverisetoerrorsthatbiassubsequentestimatesofthepovertymeasure.Whengroupeddataareallthatisavailable,theLorenzcurve(andthePDF)willhavetobeestimatedandtheparametricprocedureisatleastagoodplacetostartwith.However,ifunit-recorddataareavailable,whichisthecaseofthisstudy,asimplersolutionexists.TokeeptheLorenzcurveofanincomedistributionintactbutgiveitanewmean,onecansimplyscaleeveryobservationbythenewmeandividedbytheoldmean.Inotherwords,thetwohypotheticaldistributionscanbeconstructedasY(μ0,LT)=YT×(μ0/μT)andY(μT,L0)=Y0×(μT/μ0).ThepovertyindicesP(μ0,LT)andP(μT,L0)canthenbecalculateddirectlyfromtheconstructeddistributions.Evenwithunit-recorddata,assessingpovertytrendisstillsubjecttoahostofuncertainties,whichinturnaffectpovertydecomposition.Weconsiderthreesuchuncertaintieshere:povertymeasures,povertylines,andequivalencescales.ThemostwidelyusedpovertymeasuresarethefirstthreeFGT(Fosteretal.,1984)povertyindicesthatcanbegenericallyexpressedas

󰀈󰀇

1󰀁z−Yiα

,α󰀁0.Pα=(9)

Nz

Yi󰀁z

P0,thehead-countratio,givestheproportionofthepopulationwhoseincomesfallbelowthe

povertylinez.ThepovertygapindexP1measurestheaverageincomeshortfallinmeetingthepovertyline,wheretheshortfallisexpressedasaproportionofthepovertylineandthepovertygapofthenonpoorisassignedzero.ThesquaredpovertygapindexP2isthesumoftheproportionatepovertygapsweightedbythemselves,andisthusmoresensitivetotheincomechangesofpoorerindividuals.Thethreeindicesreflectdifferentaspectsofthesamepovertyexperience,measuringrespectivelytheincidence,depth,andseverityofpoverty.Therefore,themagnitudeanddirectionoftheirchangesneednotalwaysconcur.Thiswillthenleadtodifferentassessmentsoftherelativeroleplayedbyincomegrowthandredistributioninaffectingpoverty.9

9Totakeasimpleexample,supposethatanincomedistributionhaschangedfrom(1,2,3,4)to(2,2,2,4)andthe

povertylineissetat2.5.Thehead-countratiowouldindicateanincreaseinpoverty(from0.5to0.75)whereasthepovertygapindexwouldshowadecrease(from0.2to0.15).Decomposingthechangeinthehead-countratioaccordingtodefinitions(7)and(8)wouldputthecontributionofgrowthatzeroandthecontributionofredistributionaspovertyworsening(R>0).Thesamedecompositionappliedtothechangeinthepovertygapindexwouldgiveanegativeredistributioncomponent.

700Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712

Theevaluationofpovertytrendmayalsobesensitivetowherethepovertylineisdrawn.Forexample,ifthepovertylinehappenstobenearalocalmodeoftheincomedistribution,anim-materialshiftofthepovertylinemightcausealargeswingofmeasuredpoverty,especiallyforpovertymeasuressuchasP0andP1,whicharenotcontinuousatthepovertyline.Giventheinevitablearbitrarinessindefiningthepovertyline,anassessmentofthepovertytrendthatcanbeeasilyreversedbyaslightchangetothepovertylinewillhardlyinspiremuchconfidence.Inthispaper,weconsidersixnationalandinternationalpovertylines.10TheseincludetheUS$1.08andUS$2.15percapitaperdaypovertylinesin1993PPP,theUS$1andUS$2percapitaperdaypovertylinesin1985PPP,theruralpovertylineproposedinRavallionandChen(2004)(850yuanin2002prices),andtheofficialruralpovertylineof530yuanin1995prices.An-otherconcernaboutthepovertylineiswhetherapovertylineshouldbeapplieduniformlytoallregionsunderexamination.ThecostsoflivingvaryacrossChineseprovincessometimesbywidemargins.OfficialCPIspublishedbytheNBS,availableattheprovinciallevel,allowonetotracechangesinthecostsoflivingwithinaprovince,butnotthedifferencesacrossprovinces.UsingofficialCPIsandpricedatafor1990,BrandtandHolz(2004)constructedseveralpanelsofprovincialpricelevels.Oneoftheseisadoptedinthispapertoconvertnationalpovertylinestotheirprovincialcounterpartsor,equivalently,toconvertnominalincomefigurestorealincomesmeasuredinnationalpricesofthebaseyear.11

InmostpovertystudiesonChina,theindicatorofindividualwelfareisonapercapitabasis—totalhouseholdincome(orconsumption)dividedbythenumberofpeopleinthehousehold.Thispracticeassumesawaythepossibilitythatthepercapitacostofsemipublicgoodssuchashousing,utilities,transportation,andsoonisnegativelyrelatedtohouseholdsize.Evenforri-valgoodssuchasfood,theunitpricepaidbylargehouseholdsmaybelowerthanthatpaidbysmallhouseholdsbecausetheformeraremorelikelytomakebulkpurchases.Itisalsopossi-blethatthecostsforreachingagivenwelfarelevelaredifferentfordemographicallydifferentbutotherwiseidenticalhouseholds.Studiesonotherdevelopingcountriesshowthatthescopeforeconomiesofscaleinhouseholdconsumptioncanbeconsiderablylargebysomemeasuresbutnegligiblebyothers,andthattheeffectsofdemographiccompositionsonhouseholdcon-sumptionareinsignificant(LanjouwandRavallion,1995).Inastudyofurbanresidentsin12Chinesecities,Gustafssonetal.(2004)findthatthesizeandagecompositionofhouseholdshaveamodestimpactonhouseholds’perceptionofminimumlivingexpenditure.Whethertheirfindingalsoappliesinruralareasisanopenquestion,sinceruralhouseholdstendtobelargerandhavedifferentconsumptionpatternthanurbanhouseholds.

Theaveragehouseholdsizesofbothourdatasetsexhibitadeclinetrendovertheperiodof1988to1999,onefrom4.71to4.13persons,andtheotherfrom4.33to3.74.Ifthereex-istsignificanteconomiesofscaleandpoorhouseholdsaregettingsmalleronaverage,thenapovertytrendbasedonpercapitaincomewillunderstatetheincrease(oroverstatethereduction)inpoverty.12Toexaminehowallowingforeconomiesofscalewouldaffecttheassessmentof

10Povertylinesaresometimesdefinedrelativetothemeanincome.Wewillconfineourselvestoabsolutepovertylines.11Theruralpricelevelsbyprovinceswereobtainedbyapplyingtoa1990ruralconsumptionbaskettheofficialrural

CPIsadjustedforconsumptionofself-producedproducts.NotethatBrandtandHolz(2004)usedthesamecompositionoftheconsumptionbasketforallprovincesthroughout1984–2000.Asaresult,regionaldifferencesinandchangesovertimeofconsumptionpatternsareignored.Inaddition,theconsumptionbasketusedforderivingCPIsismeanttoberepresentativeoftheconsumptionpatternofallruralresidents,andhencemaywelldifferfromtheconsumptionpatternoftheruralpoor.

12Wedonothavesufficientinformationtoinvestigatetheexistenceofeconomiesofscaleinthedatasets.

Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712701

povertytrendanditsdecomposition,weemploythreeconstant-elasticityequivalencescalestonormalizehouseholdsizes.Morespecifically,ifnirepresentsthenumberofpeopleinhouse-holdi,thenormalizedhouseholdsizeisgivenbyki=nθi,whereθisalternativelysetto1,0.8,and0.5.

3.Dataandpovertytrend

Inthissection,wewillexaminepovertytrendinruralChinafromthelate1980stotheendof1990s,usinghousehold-levelincomedatafromtwodifferentsurveys.13ThefirstsurveywasadministeredannuallybytheResearchCentreforRuralEconomy(RCRE)oftheMinistryofAgricultureofChinaineightprovincesbetween1987and1999except1992and1994.Thenumberofhouseholdscoveredbythesurveywasexceptionallysmallin1990and1991.Fortheotheryears,itvariedbetween6200and6900households.14OurseconddatasourceistheChinaHealthandNutritionSurvey(CHNS).15FiveroundsofCHNSwereconductedin1989,1991,1993,1997,and2000.Eachroundcoveredaround15,000individualsfromabout4000householdsspreadovernineprovinces.Ofthese,abouttwothirdsofthehouseholdsand70percentoftheindividualsareclassifiedasrural.Theincomefiguresreportedineachroundappertaintotheyearimmediatelypriortothesurveyyear.

Observationsinbothdatasetsarecheckedforcompletenessandinternalconsistency.Toretainasmanydatapointsaspossible,onlythosewithmissingordubiousentriesforhouseholdsizeortotalincomearedropped.Weselectfromeachsurvey4yearsofdatatomakethesampleperiodsofthetwodatasetscomparable.ForCHNS,thefourroundschosencontaindatafor1988,1992,1996,and1999;forRCREsurvey,thesampleyearsare1988,1993,1996,and1999.

ThattheRCREsurveyandtheCHNSeachcoveronlyeightornineprovincesmightraisecon-cernsabouthowrepresentativetheyareofthe30oddChineseprovincesintotal.TheprovincesincludedinthetwodatasetsaregiveninTable1.Theseprovincesarediverseineconomicstruc-ture,levelofdevelopment,andgrowthperformanceduringthesampleperiod.Agoodfewofthemareamongthemostpopulousprovinces.Accordingtotheend-of-yearpopulationstatisticsfor2003(NationalBureauofStatistics,2004),thecombinedpopulationoftheeightprovincesintheRCREsurveyisabout40percentofthenationaltotal,whilethepopulationshareofthenineprovincesintheCHNSexceeds43percent.Theshareoftotalruralpopulationofthesampledprovincesmaybeevenhigher.Inourview,thesurveyedprovincesineachdatasetconstituteafairlybalancedrepresentationofChina’seconomicgeography.Thepovertytrendin

Table1

ProvincesincludedinthedatasetsRCRECHNS

Anhui,Gansu,Guangdong,Henan,Jiangsu,Jilin,Shanxi,Sichuan

1988,1992:Guangxi,Guizhou,Henan,Hubei,Hunan,Jiangsu,Liaoning,Shandong1996:Guangxi,Guizhou,Heilongjiang,Henan,Hubei,Hunan,Jiangsu,Shandong

1999:Guangxi,Guizhou,Heilongjiang,Henan,Hubei,Hunan,Jiangsu,Liaoning,Shandong

13Sinceweareinterestedinpovertyamongindividuals,individual-levelincomedatawouldbeideal.Thedatasetsdo

containsomeinformationonindividualearnings.However,householdmembersusuallypooltheirearnings,andlittleisknownaboutintrahouseholdredistribution.Followingstandardpractice,weassumeperfectincomeequalityamongindividualsofthesamehousehold.

14AbriefdescriptionofthehistoryoftheRCREsurveycanbefoundinWanandZhou(2005).15Thedatacanbedownloadedfromhttp://www.cpc.unc.edu/projects/china.

702Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712

theseprovincesshouldalsobeabletoconveythevariedpovertyreductionexperienceacrossthecountry.Thatsaid,itisnecessarytostressthatthissectionisnotaboutestablishingtheexactval-uesofcertainpovertymeasuresintheruralareasofChina,thoughwedopresentestimatesofthethreeFGTindicesbasedonthetwodatasets.Toundertakethattaskrequiressurveyswithamuchmoreexpansivesamplingframebothingeographicalcoverageandpreferablyinthenumberofparticipatinghouseholds.Atpresent,theNBSurbanandruralhouseholdsurveysareprobablytheonlyonesthataresufficientlycomprehensiveforsuchapurpose.UsingtheNBSsurveydata,RavallionandChen(2004)havedoneameticulousjobassemblinglongseriesoftheFGTindicesbothatthenationallevelandseparatelyfortheruralandurbanareas.Ratherthantryingtobettertheirestimates,theobjectivehereistoassessthetrendofincomepovertyinabroadspectrumofprovinces.TheresultsderivedhereinareindicativeofwhetherthepovertytrendobservedintheNBSdataissensitivetogeographicalcoverageand,sinceourdataarenotculledfromtheNBSsurveys,whetherthetrendissurveyspecific.Thatthetwodatasetsthemselvescomefromdifferentsourcesservesananalogouspurpose—asensitivitycheckonthepovertytrendpresentineachdataset.

Table2providestheestimatesofthehead-countratio,povertygap,andsquaredpovertygapindicesforthetwodatasetsunderdifferentcombinationsofalternativeassumptionsaboutpovertymeasurement.ThelefthalfofthetableshowstheresultsfortheRCREdata,andtherighthalffortheCHNSdata.Horizontally,thetableisdividedintotwopanels,eachcontainingresultsbasedonadifferentequivalencescale.16Hence,intheupper-leftblockwefindthevaluesofP0,P1,andP2fortheRCREdatawhenpercapitaincome(θ=1)isthewelfareindicator.Everypovertyindexismeasuredagainstthesixpovertylinesstatedearlier.Toexplorethesen-sitivityoftheresultstointerprovincialdifferencesinthecostsofliving,nominalincomevaluesarealternatelyadjustedforprovincialdeflatorsconstructedbyBrandtandHolz(2004)andtheruralCPI.

Itcanbeseenfromtheestimatedhead-countratios(P0)thatthesixpovertylinesrepresentroughlythreedifferentclassesofpoverty.Thelowestpovertylinez1(530yuanin1995prices)cutstheincomedistributionatthebottom5to15percentofthepopulation,thuscanbeviewedasthethresholdforextremepovertyinthiscontext.Usingthesecond,third,orfourthpovertyline,around20to30percentofthesampledruralresidentswouldbeclassifiedaspoor,anassessmentinagreementwiththegeneralperceptionofruralpovertyinChina.Theothertwolines,US$2.15in1993PPPandUS$2in1985PPPperpersonperday,designate60to70percentofthepopulationaspoor.Theyseemrathertoohighatthepresentstagetoserveaspovertythresholdsforpolicypurpose.17

Povertytrendsappeartodifferwhenmeasuredagainstthethreeclassesofthresholds.Forexample,inthefirstfourcolumnsoftheupper-leftblockwhereinterprovincialpricedifferencesarefactoredin,theprevalence(P0)ofextremepoverty(measuredagainstz1)startedrisingasearlyas1993.Thedownwardtrendinthesecondpovertyclass(measuredagainstz2–z4)didnotreverseuntil1996,whereasthedeclineinthethirdpovertyclass(measuredagainstz5–z6)wassustainedthroughoutthisperiod.Thechoiceofapovertyindexcanalsoaffecttheassessmentofpovertytrend.Judgingbythehead-countratio,forinstance,asizablereductioninpovertycan

16Settingθto0.5doesnotchangeourmajorconclusionabouttheevolutionofpovertytrendinthisperiod.However,

giventhecurrentincomelevelandexpenditurepatterninruralChina,thescopeforscaleeconomiesisprobablynotlargeenoughtojustifyθ=0.5.Avalueof0.8isarguablymoreappropriate.Theresultswithθ=0.5areavailableuponrequest.

17Wethankananonymousrefereeforhighlightingthispoint.

Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712703

Table2

Alternativeestimatesofpovertyindices

RCREdataProvincialprices

P0z1

z2z3z4z5z6

5.4919.8323.2327.1766.6171.20

5.2017.9421.2525.3363.7768.56

5.6115.2817.8621.2858.6163.22

7.9217.9520.1423.0155.8560.65

Nationalprices5.9720.6724.1227.9964.0568.15

5.0217.8220.8024.6160.6264.95

4.8813.9616.4119.4655.5759.74

θ=17.0716.9318.9921.6552.8357.51

CHNSdataProvincialprices14.8127.7130.2132.6159.3963.17

7.2617.8219.8021.9149.3352.98

7.4415.1116.9719.4243.9447.49

10.8620.3522.2224.0845.7449.38

Nationalprices14.4727.2229.8332.3358.4562.45

6.6516.1818.5720.3146.5450.64

5.9912.4714.0815.9939.4743.08

9.7317.9820.0721.5343.2846.31

1988199319961999198819931996199919881992199619991988199219961999

P1z11.401.482.013.291.491.361.452.59

z25.274.974.926.815.554.804.176.04z36.355.935.697.636.675.764.896.83z47.717.176.698.648.076.995.847.81z526.4425.3622.3723.1526.2724.2820.8521.75z629.7028.5025.3225.8529.3327.2323.6824.32P2z10.610.711.172.150.640.590.731.52

z22.172.152.523.892.302.021.943.20z32.632.572.884.322.782.432.273.61z43.253.123.344.853.422.982.704.13z513.6212.9411.5513.0113.8012.4810.5212.02z615.7615.0013.3714.7115.8714.4412.2613.64

θ=0.8

P0z12.142.272.784.652.362.152.213.93

z28.027.877.6410.768.657.576.629.85z39.889.568.7612.2210.519.437.6811.30z412.4311.8310.4513.8213.0011.719.4212.96z545.3244.9138.3037.8745.0543.0736.2036.24z650.4750.4243.8642.6149.7647.9141.4640.77

5.682.772.424.405.452.681.993.9711.366.225.568.3911.105.834.647.5312.527.046.259.2512.256.615.218.3013.888.037.1010.2613.617.545.929.2028.8020.8218.5121.8628.4619.6816.0220.1131.2723.1320.6023.8130.9021.9117.9622.003.021.641.182.462.881.620.952.196.403.342.894.856.193.192.404.357.103.763.285.366.893.582.724.817.964.293.755.977.734.073.125.3618.0611.8810.4913.5717.7911.198.9012.3219.8213.3711.8214.9319.5412.6110.1013.619.5318.2120.2722.1744.9148.33

4.289.6411.2712.5234.5737.11

4.059.4910.8611.6530.5633.96

7.4013.4414.9916.8234.7237.57

8.9017.5820.0921.9144.2747.44

4.023.646.699.208.0612.0510.429.0613.4912.0710.0215.0133.3826.3632.1435.6229.9935.16

P1z10.610.731.202.290.640.600.791.663.361.801.322.893.181.791.052.56

z22.082.152.614.162.232.012.023.457.053.623.265.596.813.462.695.04z32.542.582.984.642.712.442.363.927.854.073.726.167.613.883.075.55z43.153.153.465.243.373.032.804.518.814.654.266.868.564.403.546.17z514.9214.4012.6914.4615.3014.0311.6313.4820.2913.4211.9415.6720.0312.5510.0714.16z617.4516.9414.8616.4317.7516.4513.7215.4022.3015.1413.5017.2322.0014.2311.4615.65P2z1

z2z3z4z5z6

0.300.901.081.326.848.22

0.400.991.171.406.607.94

0.801.451.621.846.227.33

1.682.592.823.127.999.09

0.300.941.141.407.128.51

0.280.871.041.286.437.73

0.430.991.141.345.466.52

1.071.691.953.842.174.302.464.877.2112.138.2713.48

1.132.072.302.597.278.31

0.591.531.601.603.193.681.843.534.132.143.954.686.489.4111.897.4010.4413.24

1.132.022.232.506.837.80

0.461.311.511.765.446.22

1.352.863.173.548.489.42

Notes.(1)z1=theofficialruralpovertylineof530yuanin1995prices,z2=850yuanin2002prices,equivalentto833.85yuanin1995ruralprices,z3=US$1.08perdayin1993PPP,equivalentto892.85yuanin1995ruralprices,z4=US$1perdayin1985PPP,equivalentto962.39yuanin1995ruralprices,z5=US$2.15perdayin1993PPP,equivalentto1777.40yuanin1995ruralprices,z6=US$2perdayin1985PPP,equivalentto1924.80yuanin1995ruralprices.(2)SeeSection2fordefinitionsofP0,P1,andP2.

beconcludedfortheentireperiod(1988–1999)againstafairlywiderangeofpovertylines.18If,instead,P1andP2arechosentomeasurepovertytrend,onewouldconcludethattherewasno

18Nevertheless,thehead-countratioforzincreasedoverthisperiod.Theincreaseisstatisticallysignificantwiththe

1

t-statisticat11.73.ThestandarderrorsarecalculatedfollowingKakwani(1993).

704Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712

(orareversalin)povertyreductionunlessthepovertylineissetatarelativelyhighincomelevel.Regardlessofwhichpovertylineorpovertyindexisused,however,allestimatessuggestthattheearlytomid-1990ssawdivergentfortunesamongthepoor:whilesomeofthepoorforgedahead,thecircumstancesofthepoorestgroup(thosewithincomesfallingshortofthelowestpovertyline)worsened.Inaddition,povertywentuponnearlyallaccountsbetween1996and1999.Thenextfourcolumnsoftheupper-leftblockshowthevaluesofthepovertyindiceswhenthenominalincomesofallhouseholdsaredeflatedbytheruralCPIregardlessoftheirplacesofresidence.Interestingly,whilstignoringprovincialpricedifferencestendstoraisethevaluesofthepovertyindicesfor1988,itreducesthevaluesoftheindicesforalltheotheryears.Thecauseforthesecontrastingeffectsturnsouttobethatin1988thelocalpricelevelsinthemajorityoftheeightprovinces(excepttherelativelywell-offGuangdongandJiangsuprovinces)werelowerthanorclosetothenationalaverage.SuchwasnotthecaseinthelateryearswhenthepricelevelsinAnhuiandGansu,twoofthelow-incomeprovinces,stoodabovethenationalaverage.Ascanbeseen,thischangetoauniformnominalincomedeflatoraffectstheassessmentofpovertytrendbothquantitativelyand,insomeoccasions,qualitatively.Forinstance,withprovince-specificdeflators,thechangesinP1andP2between1988and1993forthelowestpovertylinearepositive,indicatinganincreaseinpoverty;usingtheruralCPIasthedeflator,thecorrespondingvaluesarenegative,suggestingadecreaseinpoverty.Apparently,muchofthiswouldturnonthedifferentialsbetweentheprovincialandnationalinflationrates.Italsoseemsthatwhethersuchinflationdifferentialsareacknowledgedbearsmoreonthedistribution-sensitivemeasureP2.MovingdowntotheotherblockontheleftsideofTable2,thevaluesofthepovertymeasuresbecomesmallerwiththedecreaseofthesizeelasticityoftheequivalencescale,θ.Thisisonlytobeexpected.Asmallerθallowsforgreatereconomiesofscale,whichresultsinhigherincomeperequivalentadultforagivenamountofhouseholdincome.Withasmallerθ,thesignsofpovertychanges,whetherinthethreeshortertimeintervalsorovertheentireperiod,arealsomorelikelytosuggestincreasedpovertyforlowerpovertylines.Underlyingthistendencyistheaforementioneddiminishingofhouseholdsizesoverthesampleperiod.Asahouseholdbecomessmaller,itlosessomeofthebenefitsfromeconomiesofscale.Thelossgrowswiththedecreaseinθ.Itisthusoflittlesurprisethattheindicationforariseinpovertyin1996–1999andin1988–1999isstrongerfromtheresultswithθat0.8.

PovertyestimatesfortheCHNSdataarereportedintherighthalfofTable2.IncomparisonwiththeresultsfortheRCREdata,thefollowingpointsarenoteworthy.First,thetwodatasetsaresourcedfromindependentsurveys.Eventhegeographiccoverageofthesurveysoverlapsonlyslightly(seeTable1).Inviewofthedistinctnessofdatasources,thetwosetsofestimates,especiallythosefor1996and1999,aresurprisinglyclose.Second,applyingprovince-specificdeflatorsincreasesthevaluesofthepovertyindices,butonlymarginally.Infact,severalestimatesarestatisticallyindistinguishablefromtheircounterpartsbasedonnominalincomesdeflatedbytheruralCPI.19Thisisbecause,asintheRCREsurvey,provincialpricelevelsintheCHNStypicallyhoverabovethenationalaverage.However,thediscrepanciesbetweenthelocalandnationalpricesarenotlargeenoughtosignificantlyaffectthevaluesofpovertymeasures.Third,consistentwiththeresultsfortheRCREdata,theCHNSdatamanifestedadeclineinallthreepovertyindicesfromthelate1980stotheearly1990sandareversalbetween1996and1999.Thedeclineinthehead-countratioandthepovertygapindexcontinuedupto1996.Finally,thetwosetsofresultssharethecharacteristicthatthequalitativeassessmentofpovertytrendis

19Thet-testresultsareavailableuponrequest.

Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712705

byandlargeindifferenttothechoiceofpovertyline,povertymeasure,andequivalencescale.Nonetheless,itisadvisabletobecautiouswhendrawingconclusionsatlowpovertylines.

Howdotheresultsherecomparewiththefindingsinotherstudies?RavallionandChen(2004)provideestimatesofthethreeFGTindicesintheruralareasfor1980–2001.ThetwopovertylinesusedintheirstudyarethesameasthefirsttwopovertylinesinTable2,andtheirwelfareindicatorisincomepercapita.Moreimportantly,theirestimatesarebasedontheunitrecorddata(atthehouseholdlevel)fromtheNBSruralhouseholdsurvey.AcomparisonoftheestimatesinTable2withtheirsis,therefore,ofinterest.Beforeproceedingfurther,however,itisnecessarytobearinmindthattheNBSsurveyhasnationwidecoverage,whichprecludesdirectcomparabilityofthevaluesofthepovertyindices.Also,RavallionandChenusetheruralCPIasthedeflator.Interprovincepricedifferencesarethusnottakenintoaccountintheirstudy.Withregardtochangesinpoverty,theirestimatesindicatelittlereductionorevenincreasesinpovertyinthelate1980sandearly1990s,rapidreductionintheearlytomid-1990s,andaslowdownofthereductioninthelate1990swithrisingpovertyin1999–2001.SeeTable2inRavallionandChen(2004).Hence,despitesomediscrepanciesinthetimingandmagnitudeofpovertychanges,theestimatesinTable2andthoseinRavallionandChen(2004)agreethatpovertydecreasedinthemid-1990sandincreasedinthelateryearsofthedecade.Regardingthepovertytrendfromthelate1980stotheearly1990s,however,theassessmentsofferedinthetwostudiesdonotsquarewellwitheachother.

TwootherstudiesofrelevanceareKhan(1999,2004),whichexaminepovertychangesinruralandurbanareasbetween1988and1995andbetween1995and2002,respectively.Bothutilizeunit-recordhouseholddatafromCASSsurveys(seefootnote3).RuralpovertyismeasuredbytheFGTindicesagainsttwopovertythresholds:theloweroneisabout5percentbelowthesecondpovertylineinTable2,andtheupperoneisabout20percentabovethefourthpovertyline.TheestimatedFGTindicesfor1988and1995inKhan(1999)arelargelyinlinewiththeestimatesfor1988and1995/1996inTable2.However,theestimatesfor2002inKhan(2004)implyrapidpovertyreductionbetween1995and2002.20Technicallyspeaking,thisinitselfdoesnotconstitutecontradictiontothefindinginTable2thatruralpovertyincreasedinthelate1990s,aspovertyreductioncouldhaveoccurredafter1999.Nonetheless,themagnitudeofthereductioninKhan(2004)pointstomajordiscrepanciesbetweenthepost-1996incomedatacollectedinCHNSandRCREsurveysandthosecollectedinCASSsurveys.ItisworthmentioningthatKhan(2004)considersmorevigorousincomeredistributionbythegovernmentafterthe1997Asianfinancialcrisisasthemajorcontributortothepovertytrendbetween1995and2002.ThisconjectureeffectivelyassignsthebulkofpovertyreductionasexhibitedbytheCASSdatatotheyearsafter1997,andhenceisatoddswiththepovertyestimatesinRavallionandChen(2004),whichshowrisingpovertyaroundtheturnofthecentury.4.Impactsofgrowthandredistribution

AsstatedinSection2,temporalchangesinpovertyindicescanbeapportionedbetweendistribution-neutralincomegrowthandshiftsinthedistributionofrelativeincome.ApplyingtotheRCREandCHNSdatatheShapleyvaluedecompositionprocedureproducestheresultsinTable3,wherethepovertytrendsinthethreetimeintervals(1988–1992/93,1992/93–1996,

20ThelatestCHNSincomedataarefromthe2000round,andhenceprovidenoinformationonhowpovertyhas

evolvedsince1999.However,datafor2000–2002areavailableintheRCREdataset.TheFGTindicesbasedonthesedataindicatethatpovertyindeeddeclinedbetween1999and2002,albeitatananemicrate.

Table3

Shapleyvaluegrowth–redistributiondecomposition

RCREdataCHNSdataProvincialpricesNationalpricesProvincialpricesNationalprices

z1

z2z3z4z1z2z3z4z1z2

z3

z4z1z2z3z4θ=1P0

1988–1992/3G−1.46−4.05−4.62−4.75−2.26−5.85−6.32−6.99−5.21−9.24−10.05−10.41−5.59−10.06−10.85−11.52R1.162.162.632.911.303.013.003.61−2.34−0.65−0.36−0.30−2.23−0.97−0.42−0.511992/3–1996G−0.22−0.58−0.72−0.960.260.660.810.57−0.86−2.22−2.14−2.21−1.16−2.67−3.11−3.10R0.62−2.07−2.67−3.09−0.40−4.52−5.20−5.711.04−0.49−0.69−0.280.50−1.04−1.39−1.221996–1999G−0.62−1.43−1.71−1.79−0.61−1.23−1.46−1.570.581.091.021.200.701.151.631.42R2.944.093.993.522.804.204.043.762.854.154.233.473.034.364.374.121988–1999G−2.45−5.68−6.25−7.10−2.50−5.40−6.40−6.63−5.77−9.30−9.88

−10.28−6.27−10.20−10.70−11.77R4.883.803.162.944.152.151.750.761.821.931.881.751.520.960.940.96P1

1988–1992/3G−0.34−1.22−1.43−1.66−0.52−1.79−2.08−2.41−1.96−3.96

−4.34−4.77−2.06−4.19−4.59−5.06R0.420.921.011.120.391.041.171.33−0.96−1.18−1.14−1.08−0.71−1.08−1.04−1.011992/3–1996G−0.06−0.18−0.21−0.250.060.200.230.27−0.34−0.75−0.84−0.94−0.43−0.96−1.10−1.24R0.580.13−0.03−0.240.03−0.83−1.10−1.42−0.010.090.040.01−0.26−0.23−0.31−0.381996–1999G−0.18−0.46−0.52−0.60−0.15−0.38−0.44−0.520.230.430.480.530.290.540.600.66R1.462.342.462.561.292.262.382.481.742.402.522.631.692.352.492.621988–1999G−0.64−1.83−2.10−2.43−0.63−1.78−2.05−2.37−2.41−4.38−4.72−5.11−2.57−4.75−5.13−5.58R2.533.373.383.362.312.822.762.651.121.421.451.481.091.191.181.17P2

1988–1992/3G−0.14−0.52−0.62−0.75−0.21−0.77−0.92−1.11−1.05−2.24−2.48−2.77−1.10−2.36−2.62−2.93R0.230.500.560.630.160.500.570.67−0.33−0.82−0.86−0.90−0.16−0.64−0.69−0.741992/3–1996G−0.02−0.08−0.09−0.110.030.090.100.12−0.17−0.39−0.45−0.51−0.22−0.51−0.58−0.66R0.480.450.400.320.11−0.16−0.26−0.40−0.29−0.05−0.04−0.03−0.45−0.29−0.28−0.291996–1999G−0.08−0.22−0.26−0.30−0.07−0.19−0.22−0.250.130.250.270.310.160.310.350.39R1.071.601.701.810.861.441.551.681.161.711.811.921.081.641.741.861988–1999

G−0.29−0.86−1.00−1.18−0.28−0.84−0.98−1.15−1.42−2.65−2.90−3.18−1.51−2.85−3.12−3.43R

1.83

2.58

2.68

2.78

1.74

2.31

2.37

2.42

0.86

1.11

1.15

1.20

0.83

1.02

1.04

1.06

(continuedonnextpage)

706Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712Table3(continued)

RCREdataCHNSdataProvincialpricesNationalpricesProvincialpricesNationalpricesz1

z2z3z4z1z2z3

z4

z1z2z3z4z1z2z3z4θ=0.8

P0

1988–1992/3G−0.50−1.72−2.24−2.41−0.77−2.80−3.22−3.78−3.20−6.37−7.23−7.97−3.15−6.54−7.50−8.14R0.631.571.921.810.561.722.132.49−2.05−2.20−1.78−1.67−1.74−1.84−2.17−1.701992/3–1996G−0.01−0.10−0.13−0.170.120.560.670.69−0.39−0.92−0.99−0.94−0.38−1.61−1.48−1.76R0.52−0.13−0.66−1.20−0.06−1.51−2.42−2.990.160.760.580.070.000.470.12−0.291996–1999G−0.22−0.71−0.61−0.99−0.22−0.60−0.67−0.800.280.820.761.070.361.031.131.22R2.093.834.074.361.933.834.284.353.073.133.374.102.692.973.293.771988–1999G−0.87−2.41−2.89−3.55−0.84−2.43−2.82−3.38−3.66−6.62−7.06−7.52−3.64−7.06−7.60−8.40R3.385.155.224.942.984.164.133.851.531.861.772.171.431.541.001.50P1

1988–1992/3G−0.11−0.44−0.54−0.67−0.19−0.71−0.86−1.05−1.15−2.37−2.65−3.00−1.18−2.53−2.83−3.19R0.230.500.580.670.150.490.590.72−0.42−1.06−1.12−1.16−0.22−0.82−0.90−0.971992/3–1996G−0.01−0.03−0.04−0.040.040.140.170.21−0.15−0.36−0.40−0.45−0.22−0.51−0.58−0.65R0.480.490.440.350.15−0.14−0.26−0.44−0.33−0.010.040.06−0.51−0.26−0.23−0.211996–1999G−0.08−0.23−0.26−0.31−0.06−0.19−0.22−0.260.170.340.370.410.220.410.460.50R1.171.781.922.090.941.621.781.971.401.992.072.191.291.942.022.121988–1999G−0.24−0.75−0.87−1.04−0.24−0.74−0.86−1.02−1.47−2.69−2.97−3.28−1.59−2.97−3.26−3.60R1.922.822.983.131.832.532.642.730.991.241.281.330.971.201.201.20P2

1988–1992/3G−0.05−0.18−0.22−0.27−0.08−0.28−0.35−0.43−0.61−1.31−1.46−1.65−0.64−1.38−1.55−1.75R0.150.270.310.350.050.210.250.310.06−0.46−0.54−0.620.18−0.28−0.35−0.431992/3–1996G0.00−0.01−0.02−0.020.020.060.070.09−0.08−0.19−0.21−0.24−0.12−0.26−0.30−0.34R0.410.470.470.460.130.060.03−0.02−0.46−0.29−0.25−0.21−0.56−0.45−0.42−0.401996–1999G−0.04−0.11−0.13−0.15−0.03−0.09−0.10−0.120.100.200.220.240.120.240.260.29R0.921.251.331.420.671.051.131.230.841.391.471.560.771.311.401.491988–1999G−0.11−0.34−0.40−0.47−0.11−0.33−0.39−0.46−0.85−1.60−1.76−1.95−0.92−1.74−1.92−2.13R

1.49

2.03

2.14

2.27

1.46

1.91

1.99

2.09

0.69

0.94

0.99

1.03

0.66

0.92

0.96

0.99

SeenotestoTable2.

Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712707708Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712

1996–1999)andduring1988–1999aredecomposedintocontributionsbyincomegrowthandthosebyredistributionaccordingtoexpressions(7)and(8),respectively.Table3isstructuredinthesamewayasTable2exceptthatthecolumncategoryineachblockisnowthepovertylinesinsteadofthesampleyears.21Foragiventimeinterval,thegrowthcomponentsunderalternativepovertylinesarearrayedintherowlabeledGandtheredistributioncomponentsintherowla-beledR.Notethatnegativevaluesinthetabledenotereductionsinpoverty,andpositivevaluesdenoteincreasesinpoverty.

Theresultsintheupper-leftblockshowhowgrowthandredistributionwouldhaveaffectedthepovertytrendsintheRCREdataiftherewerenoeconomiesofscalewithinhouseholds.Ascanbeseen,thegrowthcomponentsinthefirstfourcolumnsarenegativeacrossallpovertylinesandpovertymeasures.Thisimpliesthatifnominalincomesaremeasuredagainstlocalprices,anaveragehouseholdintheRCREsurveywouldhaveexperiencedgrowthinrealpercapitaincomethroughoutthesampleperiod.Bycontrast,theredistributioncomponentsaremostlypositive.Onlyinthedecompositionsfor1993–1996arenegativeredistributioncomponentspresent,andthenonlyforhigherpovertylines.22Hence,whilethechangesinthedistributionofrelativeincomeduringthesampleperiodof1988–1999werehardlyfavorabletothepoor,itwasthepoorestgroupwhosawsustainedworseningoftheirrelativepositionsintheincomedistribution.Therelativemagnitudesofthetwocomponentsindicatethatthereductionofpovertyin1988–1993wasdrivenbyincomegrowth.Sincethen,however,distributionalchangeshadbecomethedominantfactor.Theincreaseinpovertyatlowerpovertylinesin1993–1996,aswellastheincreasein1996–1999,wasfullyattributabletoadversedistributionalchanges.

ItisseeninTable2thatreplacingprovincialpriceswithnationalaveragepricescanaffectnotonlytheestimatedvaluesofthepovertyindicesbutalsotheassessmentofpovertytrend.Whenitcomestopovertydecomposition,ignoringregionalpricedifferentialsisnotinconsequential,either,astheresultsintheotherfourcolumnsintheupper-leftblockshow.Between1988and1993,theinflationratesinmostoftheprovincesintheRCREsurveywerehigherthanthena-tionalaverage.UsingtheruralCPIasthedeflatorthusresultsinanexaggerationofthegrowtheffectsrelativetotheredistributioneffects.Theoppositeinflationscenariooccurredinthetwointervalsafter1993,whentheinflationratesinmostofthesampledprovinceswerebelowthenationalaverage.UsingtheruralCPIinthesetwocasesleadstoanunderstatementofthegrowthcomponents.Forthe1993–1996interval,inparticular,becausetheactualgrowthofrealincomewasratherweak,theaverageofthe1996incomefiguresaresmallerthanthatofthe1993figuresafterbeingdeflatedbytheruralCPI.Consequently,thegrowthcomponentsbasedontheoverde-flatedincomevaluesarepositive,suggestingacontractioninrealincomewhenthereisinfactrealgrowth.

Comparingthedecompositionresultsunderθ=0.8withthoseunderθ=1,itisclearthatallowingforeconomiesofscalewithinhouseholdstendstodampenthepoverty-reducingeffectsofthegrowthcomponent.Asdiscussedearlier,thisisbecausehouseholdsizeshadbeenshrinkingduringthesampleperiod.Withthepresenceofeconomiesofscale,thediminutionofhouseholdsizesoffsetssomeofthegrowthinhouseholdincomesothatthegrowthofincomeperequivalent

21AstheestimatesinTable2indicatethatthetwo$2perdaypovertylinesaretoohighformeasuringChina’srural

poverty,theirdecompositionresultsareomittedheretoavoidclutteringupthetable.

22Apossibleexplanationfortherespitefromworseningdistributionis,asmentionedearlier,governmentinterventionintensifiedinthegrainmarketafter1993.Aslow-incomeruralhouseholdsareheavilyreliantonfarmingincomes,measuressuchasraisingprocurementprices,ensuringthesupplyofagriculturalinputsandguaranteeingthepurchaseofcontractedquotastendtobenefitthemespecially.

Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712709

adultislowerthanthegrowthofincomepercapita.Howgreatereconomiesofscale,workingincombinationwithdiminishinghouseholdsizes,affecttheredistributioncomponentanditsrelativeimportancevis-à-visthegrowthcomponentisnotreadilyclearfromTable3.Partoftheanswerwilldependonhowhouseholdsizesarecorrelatedwiththegrowthofrealincomepercapita.

ThedecompositionresultsofusingCHNSdata,presentedintherighthalfofTable3,havemanysimilaritiestotheresultsusingtheRCREdata.Bothshowthatthereductioninpovertyin1988–1992/93wasprimarilydrivenbyincomegrowthwhilsttheincreaseinpovertyin1996–1999wasmainlyattributabletoadversedistributionalshifts.AsintheRCREresults,theuseofsmallersizeelasticitiesofequivalencescaledetractsfromthepoverty-reducingeffectsofincomegrowth.However,therealsoexistsomemajordifferencesbetweentheCHNSandRCREresults.FortheCHNSdata,theredistributioncomponentsofthe1988–1992intervalaremostlynegativeexceptforthetwohighestpovertylines.Thisindicatesthatdistributionalchangesbe-tweentheseyearswerefavorabletothepoor(atleasttothoseattheverybottomoftheincomedistribution,ifnotallthepoor),thoughtheameliorativeeffectsofthesechangesseemrathersmall.FortheRCREdata,theredistributioncomponentsofthecomparableinterval1988–1993arepositiveacrosspovertylinesandpovertyindices,exhibitingnosignofpro-poordistributionalchanges.Forthe1992/93–1996interval,theRCREandCHNSresultsdifferintherelativeim-portanceofthegrowthandredistributioncomponents.TheRCREresultspointtothedominanceoftheredistributioncomponentwhereastheoppositeistruefortheCHNSresults.Yetanotherdifferenceconcernsthecausesfortheincreaseinpovertyinthe1996–1999interval.Bothsetsofresultsindicatethatdistributionalshiftsaremainlytoblame.However,theCHNSresultsalsosuggestthatstagnationornegativegrowthofrealincomecompoundedtheproblem.TheRCREresults,incontrast,suggestthatincomegrowthwasstillpositive,butthepoverty-reducingeffectsofgrowthwereoutweighedbythepoverty-increasingeffectsofdeterioratingincomedistribu-tion.

Tosumup,thedecompositionresultsinTable3indicatethattheyearsbetweenthelate1980sandthelate1990ssawchangingbalanceofimportancebetweenincomegrowthanddistributionalshiftsinaffectingpovertytrend.Thepoverty-reducingeffectsofincomegrowthdiminished,givingprominencetotheimpactsofdistributionalchanges.Theevolutionofruralincomegrowthpartlyreflectsthecyclicalfluctuationsofaggregategrowthinthisperiod,asthegrowthofrealGDPpercapitadeclinedfromjustunder13percentin1992toalittleover6percentin1999.ItishoweveralsoindicativeoftheshrinkingGDPshareofagriculture,from25.7percentin1988to17.6percentin1999.AscanbediscernedfromFig.1,therelativepricesofagriculturalproductshavebeenonthedeclineexceptfortheperiodof1993–1996whenthegovernmenttemporarilyshelvedtheliberalizationofmarketsofsomestrategicallyimportantcrops.

Theperennialrelativedeclineofagricultureentailsthatruralincomegrowthhastocomein-creasinglyfromnonfarmactivities.However,notallruralhouseholdsinallregionsareequallyequippedtomakethattransition.Table4showsthechangesofpovertyincidencesinthreeprovincesintheRCREdata.ThecoastalJiangsuprovince,wheretownshipandvillageenter-prises(TVEs)flourishedduringthisperiod,sawthepercentagehalvedofthosewithincomebelowUS$1perday.InJilin,partofChina’sheavy-industryheartland,thesameratiomorethandoubledprobablybecausethelargestateindustrialsectordidnotleavemuchroomforthedevelopmentofruralindustry.ThesouthwesternSichuanprovincehadoneofthenation’shighestpovertyincidenceshistorically.Itspovertyratein1999remainedhighbynationalstandard,buthadcomedownsubstantiallyfromthe1988level.Sichuanwasnotparticularly

710Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712

Table4

Changesinpovertyincidencesinthreeprovinces

z11988

JiangsuJilinSichuan

0.875.007.69

19991.3921.202.34

z3198813.5014.5340.05

19996.3038.8317.00

Note:fordefinitionsofz1andz3,seenotestoTable2.

successfulinpromotingTVEsduringtheseyears,butithadthelargestnumberofmigrantwork-erswhoseremittancesin1995alreadyamountedto7percentoftheprovincialGDP(Gilley,1996).

Hence,duringtheperiodof1988–1999realincomesfromagriculturalproduction,tradition-allythemainstayofruralincome,stagnatedandsometimesevendecreased.Inthemeantime,regionsthatwereabletodiversifyintononfarmactivitiesstillachievedlargepovertyreduc-tion.ThissuggeststhatslowincomegrowthandunfavorabledistributionalchangesidentifiedinTable3arebothrelatedtochangesintheruraleconomyatlarge.Povertyalleviationpoliciesshouldthereforebeformulatedwithinthesamecontext.5.Concludingremarks

ThispaperhasexaminedthechangesinChina’sruralpovertyfromthelate1980stothelate1990s,employingtwodatasetsthatprovidehousehold-levelincomedata.Toquantifytherelativecontributionsofincomegrowthandredistribution,weadoptedtheShapleyvalueversionofthegrowth–redistributionpovertydecompositionadvancedbyDattandRavallion(1992),andextendeditfortheusewithunit-recorddata.Wealsoexploredthesensitivityofresultstothechoiceofpovertymeasures,povertylines,andequivalencesscales,andtodifferentialprovincialprices.

Ourresultsrevealthatboththeassessmentofpovertytrendanditsdecompositioncanbequalitativelyaffectedbydifferentmeasurementassumptions.Nonetheless,someresultsappeartoberobust.Theperiodof1988–1999isfoundtobeoneofprogressandreversalinruralpovertyreduction.Incomegrowthensuredsubstantialpovertyreductionintheearlyyearsoftheperiod.Asincomegrowthtaperedoffinthelate1990s,theimpactofadversedistributionalchangeseitheroutweighedtheeffectsofweakincomegrowthoraggravatedthoseoffallingincomes,leadingtoariseinpoverty.AllthesecompareratherunfavorablywithChina’sexperienceinthelate1980swhenagriculture-ledgrowthraisedruralincome,improvedincomedistribution,and,asaresult,achievedunprecedentedreductioninruralpoverty.23Couldthatexperiencebethemodelfordesigningpovertyalleviationpolicytodaytoresurrectruralincomegrowthandarrestworseningdistribution?

Notlikely.China’sruraleconomyismuchdifferentnowthanitwasintheearly1980s.TheGDPshareofagriculturedroppedto13percentin2004,yettheagriculturalsectorstillemploys50percentofthelaborforce.Ifeconomichistoryisanyguide,theagriculturalsectorwillcon-tinueitsrelativedecline.Relyingonagriculturalgrowthtoraiseruralincomeisnotsustainable.

23RavallionandChen(2004)estimatethatthehead-countratioofruralpovertydeclinedfrom41percentin1980to9

percentin1985bytheofficialpovertyline,andfrom76percentto23percentbytheirruralpovertyline.

Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712711

Giventheconstraintsofland,water,andothernaturalresources,theultimatewayofraisingagri-culturallaborproductivity(andhencepercapitaagriculturalincome)istotransferlaborfromagriculturetoindustryandservices.

Unliketheagriculture-basedgrowthintheearly1980swhereequalizedallocationoflandkeptincomegapsatbay,theprocessofdiversificationintononagriculturalactivitiestendstoincreasedisparitybothbetweenregionsandbetweenindividuals.Thisisbecausenotallruralhouseholdsareendowedwiththefinancialandhumancapitalrequiredformakingthattransition,andnotallregionsarewellpositionedtohaveeasyaccesstointernationalandurbanmarkets.Itappears,therefore,thatamongtheforcesresponsibleforflagginggrowthandincreasingdisparityinruralChinainthelate1990s,atleastsomesharethesameorigin.Thatoriginistheage-olddevelopmentproblemofsurplusrurallabor.Thus,policiesthatenable,encourage,andfacilitatethetransferofrurallaborcanhaveapositiveimpactonbothgrowthanddistribution,therebyhelpingrestoremomentumtoruralpovertyreduction.References

Brandt,Loren,Holz,CarstenA.,2004.SpatialpricedifferencesinChina:Estimatesandimplications.Manuscript.SocialScienceDivision,HongKongUniversityofScienceandTechnology,HongKong.

Chen,Shaohua,Ravallion,Martin,2004.Howhavetheworld’spoorestfaredsincetheearly1980s?WorldBankEco-nomicObserver19(2),141–169.

Chen,Shaohua,Wang,Yan,2001.China’sgrowthandpovertyreduction:Trendsbetween1990and1999.WorkingpaperNo.2651.PolicyResearch,WorldBank,Washington,DC.

Datt,Gustav,Ravallion,Martin,1992.Growthandredistributioncomponentsofchangesinpovertymeasures:Adecom-positionwithapplicationstoBrazilandIndiainthe1980.JournalofDevelopmentEconomics38,275–295.

Fang,Cheng,Zhang,Xiaobo,Fan,Shenggen,2002.EmergenceofurbanpovertyandinequalityinChina:Evidencefromhouseholdsurvey.ChinaEconomicReview13,430–443.

Foster,James,Greer,Joel,Thorbecke,Erik,1984.Aclassofdecomposablepovertymeasures.Econometrica52,761–766.

Gilley,Bruce,1996.Irresistibleforce:Migrantworkersarepartofasolution,notaproblem.FarEasterEconomicReview159(14),19.

Gustafsson,Björn,Wei,Zhong,2000.HowandwhyhaspovertyinChinachanged?Astudybasedonmicrodatafor1988and1995.ChinaQuarterly164,983–1006.

Gustafsson,Björn,Li,Shi,Sato,Hiroshi,2004.CanasubjectivepovertylinebeappliedtoChina?Assessingpovertyamongurbanresidentsin1999.JournalofInternationalDevelopment16,1089–1107.

Jain,L.R.,Tendulkar,SureshD.,1990.Roleofgrowthanddistributionintheobservedchangeinheadcountratiomeasureofpoverty:AdecompositionexerciseforIndia.IndianEconomicReview25,165–205.

Kakwani,Nanak,1993.Statisticalinferenceinthemeasurementofpoverty.ReviewofEconomicsandStatistics75,632–639.

Kakwani,Nanak,Subbarao,Kalanidhi,1990.RuralpovertyanditsalleviationinIndia.EconomicandPoliticalWeekly25,A2–A16.

Khan,AzizurRahman,1999.PovertyinChinaintheperiodofglobalization:Newevidenceontrendandpattern.Dis-cussionpaperNo.22.IssuesinDevelopment,InternationalLabourOffice,Geneva.

Khan,AzizurRahman,2004.GrowthinequalityandpovertyinChina.DiscussionpaperNo.15.IssuesinEmploymentandPoverty,InternationalLabourOffice,Geneva.

Kolenikov,Stanislav,Shorrocks,Anthony,2005.AdecompositionanalysisofregionalpovertyinRussia.ReviewofDevelopmentEconomics9,25–46.

Lanjouw,Peter,Ravallion,Martin,1995.Povertyandhouseholdsize.EconomicJournal105,1415–1434.

McDonald,JamesB.,1984.Somegeneralizedfunctionsofthesizedistributionofincome.Econometrica52,647–663.NationalBureauofStatistics,2004.StatisticalYearbookofChina2004.NationalBureauofStatistics,Beijing,China.Ravallion,Martin,Chen,Shaohua,2004.China’s(uneven)progressagainstpoverty.WorkingpaperNo.3408.PolicyResearch,WorldBank,Washington,DC.

Shorrocks,Anthony,1999,Decompositionproceduresfordistributionalanalysis:AunifiedframeworkbasedontheShapleyvalue.Mimeo.UniversityofEssex.

712Y.Zhang,G.Wan/JournalofComparativeEconomics34(2006)694–712

UNMillenniumProject,2005.Investingindevelopment:Apracticalplantoachievethemillenniumdevelopmentgoals.Overview.

Wan,Guanghua,Zhou,Zhangyue,2005.IncomeinequalityinruralChina:Regression-baseddecompositionusinghouse-holddata.ReviewofDevelopmentEconomics9,107–120.

Yao,Shujie,2000.EconomicdevelopmentandpovertyreductioninChinaover20yearsofreforms.EconomicDevelop-mentandCulturalChange48,447–474.