THECASEOFCHINA
ZHUOCHEN,WALLACEE.HUFFMANandSCOTTROZELLE∗
Indevelopingagricultures,pastresearchhassuggestedaninverserelationshipbetweenfarmproductivityandsize.TherawdatafromChinashowsuchaninverserelationship.However,theinverserelationshipdisappearsafterweinstrumentforlandareausingthefactthatoneoftheobjectivesofthelandallocationprocessinruralChinaistoensurelocalhouseholdstomeettheirnutritionalneeds.Theempiricalinverserelationshipislikelyduetothefailuretoaccountfortheunobservedlandqualitythatisunevenlydistributedacrossthefarmsizecontinuum,ratherthaninherenttoChina’sagriculture.(JELO13,Q12,Q15)
I.
INTRODUCTION
Inthedevelopmentliterature,itisallegedthatfarmsize(e.g.,landarea)andproductivity(e.g.,totalfarmoutputdividedbylandarea)areinverselyrelatedindevelopingcountries(Sen1962).1AlthoughChayanov(1926)iscreditedwithbeingthefirsttosuggestthisrelationshipforprewarRussianagriculture,Sen(1962)is
authorsthankparticipantsattheAmericanAgri-culturalEconomicsAssociation2005AnnualConferenceinProvidence,RIforhelpfulcomments.TheauthorsalsowishtothankDrs.JinyongHahnandJerryHausmanforshar-ingtheircodesoftheHahn-Hausman(2002)testandDr.RobinBurgessforsharinghisworkingpapers.CommentsfromDrs.Xiao-YuanDongandDennisTaoYangaregreatlyappreciated.Theviewsexpressedinthispaperrepresentthoseoftheauthors.Noofficialsupportorendorsementbytheiremployersorfundingagenciesisintended,norshouldbeinferred.
Chen:VisitingScholar,ChicagoCenterofExcellenceinHealthPromotionEconomics,TheUniversityofChicago,1600CliftonRdNE,MS-E33,Atlanta,GA30333.Phone404-498-6317,Fax404-498-1111,E-mailzchen1@cdc.gov
Huffman:CharlesF.CurtissDistinguishedProfessorofAgricultureandProfessorofEconomics,DepartmentofEconomics,IowaStateUniversity,HeadyHall260,Ames,IA50011.Phone515-294-6359,Fax515-294-0221,E-mailwhuffman@iastate.edu
Rozelle:HelenF.FarnsworthSeniorFellow,TheFree-manSpogliInstituteforInternationalStudies,Stan-fordUniversity,EncinaHallEast,E407,Stanford,CA94305.Phone650-724-6402,Fax650-723-6530,E-mailrozelle@stanford.edu1.Notethat,Benjamin(1995)usesanalternativespeci-ficationbyregressingoutputonfarmsizeandsuggeststhataninverserelationshipexistsifthecoefficientestimateoffarmsizeislessthan1.Thetwospecificationsgenerallyproduceconsistentinference,buttheBenjaminspecificationcanbetteraddresstheissueofmeasurementerrorinfarmsizethantheotherbecausefarmsizedoesnotappearintheleft-handside.
580
ContemporaryEconomicPolicy(ISSN1465-7287)Vol.29,No.4,October2011,580–592OnlineEarlypublicationOctober8,2010
∗The
consideredtheearliestmodernreferenceonthesubject.Iftheinverserelationshipdoesexist,thenpoliciessuchas“smallbutefficient”comeintoplay,andlandredistributiontobreakuplargefarmsmaybeproposed.TheKeralaLandReformsActof1963imposesaceilingof15acresforlandownedbyindividualsandjointfarmingfamiliesintheIndianstateofKerala.InChina’sagriculture,resultsfrominvesti-gatingtherelationshipbetweenfarmproductiv-ityandfarmsizeareofinteresttopolicymakers.ManyChinesefarmsaretinyandtheirsmallsizelimitsmechanization.ArapidlygrowingChinesenonfarmeconomyhasattractedlargenumbersofrurallaborintothemajorcities,whichhasprovidedopportunitiesforlandcon-solidationinruralareastoreducefragmentationandtoincreasefarmsizethroughlandrentalmarkets(DeiningerandJin2005).If,indeed,smallfarmsaremoreefficientthantheirlargercounterparts,effortstopromoteconsolidationshouldnotbepursued.2ArecentresolutionoftheCentralCommitteeoftheCommunistParty
2.Notethattherelationbetweenfarmsizeandproduc-tivityshouldbedistinguishedfromthestudyoftechnicalefficiency,whichistheengineeringefficiencyinthepro-ductionprocess(BerryandCline1979).Theefficiencyinthe“smallbutefficient”referstotheoveralllandutilizationoftheavailablelandresource.
ABBREVIATIONS
HRS:HouseholdResponsibilitySystemIV:InstrumentalVariable
LRLC:LawonRuralLandContractOLS:OrdinaryLeastSquare
RCRE:ResearchCenterforRuralEconomy
doi:10.1111/j.1465-7287.2010.00236.x
©2010WesternEconomicAssociationInternational
CHEN,HUFFMAN&ROZELLE:INVERSEPRODUCTIVITYRELATIONSHIPINCHINA’SAGRICULTURE581
ofChinaquelledspeculationsthatlandpriva-tization,orotherradicalformsoflandreform,wasontheimmediateagenda.3Meanwhile,theresolutionassertedboththeimportancetoex-ploitthescaleeconomiesofmodernagricul-turaltechnologiesandanincreasingroleoflandrentalmarkets,whichareinlinewithrecom-mendationsmadebyDeiningerandBinswanger(1999).
WiththenotableexceptionsofBenjaminandBrandt(1997,2002),fewstudieshaveexaminedtherelationshipbetweenproductivityandfarmsizeinChina’sagriculture.4SincetheimplementationoftheHouseholdResponsibilitySystem(HRS)during1978–1984,mostarablelandinruralChinaisownedcollectivelybyruralcommunitiesanddistributedbyvillagecouncilstolocalfamiliesforfarming(Dong1996).Followingtheinitiativeinthe1982ChineseConstitution,theAgricultureLaw,approvedin1993andamendedin2002,stressescollectiveownershipbyruralcommunitiesoverrurallandandtheallocationofrightsofusetolandareatoruralhouseholds.
Theobjectiveofthisarticleistoexaminetheempiricalrelationshipbetweenfarmpro-ductivityandfarmsizeinChina’sagricultureover1995–1999whenfarmersweregainingbetteraccesstomodernfarmingtechnologiesandequipment.Weconcludethattheinverserelationshipbetweenfarmproductivityandfarmsizecanbeattributedtomeasurementerroraris-ingfromimpropermeasurementoffarmland,forexample,failuretoaccountforlandquality.Thisarticlecontributestotheliteratureinthefollowingareas.First,althoughChina’slandtenuresystemisunique,itisnotuncommontoobservesimilarcollectivelyheldlandown-ershipindevelopingortransitioncountries,forexample,inVietnam(DeiningerandJin2008),CentralEurope(RozelleandSwinnen2004),andincertaincountriesinAfrica(DeiningerandFeder2001).Hence,conclusionsfromthisstudymayprovideusefulinsightstolandpolicyand
3.The17thSessionoftheCentralCommittee,theCommunistPartyofChina.Resolutionsonadvancingruralreformanddevelopmentandothermajorissues(Zhong-gongzhongyangguanyutuijinnongcungaigefazhanruo-ganzhongdawentidejueding).ApprovedonOctober12,2008,atthethirdplenarymeeting.http://www.most.gov.cn/yw/200810/P020081021325175156415.doc.AccessedonFebruary4.Burgess27,2009.
(1997)appearedtohaveaddressedtheinverseproductivityissueusingdatafromtwoprovincesofChinaascitedinotherstudies.We,however,werenotabletoacquirethemanuscriptandthelaterversions(Burgess2001,2007)seemedtohavedroppedtheanalysis.
landreformsinothereconomies.Second,weexploitthefactthatanobjectiveofruralvillagecouncilsistoensurelocalfamiliestomeetbasicnutritionalneeds(Burgess2007;DeiningerandBinswanger1999)andthevariationinthenum-berofdependentsandhukouregistrationinruralhouseholds(notethatitispossibleforaruralhouseholdtohavememberswithurbanhukouregistration)tomotivateaninstrumentalvariableestimation.AlthoughouruseofinstrumentalvariableestimationissimilartothatinBenjamin(1995),themechanismwehypothesizedisnewtotheliterature.Third,usingneweconomet-rictechniques(HahnandHausman2002),weexplicitlyaddresstheweakinstrumentproblemthathasbeenraisedintheliterature(Heltberg1998).Fourth,weuseahousehold-leveldatasetthatallowsustoavoidthe“ecologicalfal-lacy”thatariseswhenaggregatedataareused(Heltberg1998).
II.
THERELATIONSHIPBETWEENFARM
PRODUCTIVITYANDSIZE
A.ExistingTheoriesandEmpiricalEvidenceAnumberoftheoriesandexplanationsfortheinverserelationshipbetweenfarmsizeandpro-ductivityexist.First,labormarketimperfectionsareapotentialsource.Forexample,Sen(1962)attributedtheinverserelationshiptolabordual-ism,inwhichlargeandsmallfarmsareassumedtohavethesametechnology,butsmall-scalefarmershaveloweropportunitycostsoftheirlaborthanoperatorsoflargefarms.DeiningerandFeder(2001)suggestthatafarmusingonlyfamilylaborismoreefficientbecauseitisfreeofprincipal-agentproblems,thatis,familymem-bershavealong-runinterestinthesuccessofthefarm.Whenafarmissmallandlabormarketsarenotfunctioning,small-scalefarmsuseonlyfamilylaboranddonothirelabororselllaborinthenonfarmlabormarket(TaylorandAdel-man2003).Inaneconomywithprivatelandownership,familymembershaveastrongincen-tivetowork,becausetheysharethefarmoutputdirectly,andinthelongrun,theymightexpecttoinheritthefarm.Second,Barrett(1996)arguesthatthecombinedeffectsofnondegener-ativelanddistributionandpriceriskcouldpro-duceaninverserelationship.Third,AssuncaoandGhatak(2003)presentatheoreticalmodelshowingthatendogenousoccupationalchoiceandheterogeneityinfarmingskills,coupledwithcreditmarketimperfections,canexplain
582CONTEMPORARYECONOMICPOLICY
theinverserelationshipwhenthereisaconstantreturntoscaleandnolabormarketimperfection.Fourth,BenjaminandBrandt(2002)attributetheinverserelationshipbetweenfarmpro-ductivityandsizeinChina’sagriculturetolocaladministrativelanddistributionpoliciesandunevenoff-farmworkopportunities.5Fifth,measurementerrorinlandinputduetohet-erogeneouslandcouldalsoexplainaninverserelationship.Forexample,BenjaminandBrandt(2002)andLamb(2003)investigatetheroleofmeasurementerrorinlaborandland,respec-tively,andshowhowitcreatestheobservedrelationship.Deolalikar(1981),BhallaandRoy(1988),andBenjamin(1995)suggestthatunob-servedlandqualityispositivelyrelatedtofarmproductivitybutinverselyrelatedtofarmsize.Lamb(2003)confirmsthatinclusionoflandqualityadjustmentslargelyexplainstheinverserelationshipbetweenfarmsizeandprofit.How-ever,Heltberg(1998)claimsthatBhallaandRoy’sconclusionsareunderminedbytheiruseofdistrictaggregatedataratherthanhousehold-leveldata,butCarter(1984)findsasignificantwithin-villageinverserelationshipbetweenfarmsizeandproductivityusingfarmleveldataforHaryana,India.Finally,Heltberg(1998)cau-tionsresearchersagainsttheweakinstrumentproblemthatmayhaveunderminedsomepre-viousinstrumentalvariableestimation.B.LandAllocationinChinaandInstrumentsforLand
Weproposethatheterogeneityinlandqual-ityintroducedbythelocalvillagecouncil’sallocationprocesscontributestotheobservedinverserelationshipbetweenfarmsizeandpro-ductivity.TheargumentissimilartothatbyBenjamin(1995),albeitviaadifferentmech-anism.Benjamin(1995)suggeststhatinruralJava,“[I]ffarmsweresubdividedthroughinher-itanceovertime,egalitarianmotivesonthepartofthebenefactorwouldresultinhigherqual-ityparcelsbeingdividedmoreoftenthanlowqualityparcels,”whichwouldimpartanegativecorrelationbetweenfarmsizeandlandquality,particularlyatthelocalorvillagelevel.Incon-trast,thecorrelationbetweenfarmsizeandlandqualityinChinaisaresult(atleastinthepast)ofaproactivelandallocationprocessinsteadofanaturalprocessofinheritance.
5.NotethatthevillagesinnortheastChinaappeartobeinthesamecountiesasvillagesstudiedbyBenjaminandBrandt(1997).UndertheHRS,themajorityofarablelandinruralChinaisownedbyruralcommunitiesandmanagedbyalocalvillagecouncil(Burgess2007).ThepoliticsinruralChinaarecharac-terizedbyamixtureofanauthoritariancom-mandsystemandgrass-rootsdemocracy,whichhasestablishedthelegalityofegalitarianprin-ciplesduringlanddistributionandensuresitsimplementation.TheLawonRuralLandCon-tract(LRLC),whichwasapprovedinAugust2002,andwentintoeffectinMarch2003,stip-ulatesthatequityisaprimeobjectiveduringlandallocations(the9thStandingCommittee2002).However,atthesametime,thelegis-lationalsostatesthatnolandreallocationisallowedbeforethecontractexpires,anditisille-galforlocalvillagecouncilstovoidHRScon-tracts.Therefore,ifduringthecontractperioddemographicoremploymentfactorschange,itispossiblethattheegalitarianallocationsoflandcouldbecompromised.Anylatechangesneedtoberatifiedbyatleasttwo-thirdsofavil-lagecouncil(Article48oftheLRLC).AlthoughLRLCbecameeffectiveafterthestudyperiod,itreflectsprevailingpractices(andchangesinprevailingpractices)oflandallocationinruralChinaduringthestudyperiod.
LandallocationinruralChinahadbeenatwo-stepprocess(ChenandBrown2001;Dong1996)duringthestudyperiod.First,eachruralhouseholdreceives“subsistence”land—enoughlandtoguaranteethebasicfoodandnutritionalneedsofitsmembers,andsecond,anyremain-inglandistoberentedouttowillingruralhouseholds.Amemberofahousehold,usuallythehouseholdhead,presentstothecouncilhisorherhousehold’sneeds,andtheremaybenego-tiations.Majorqualitydifferencesinthelocallandareduetotheamountofwaterandpotentialforirrigationaswellasthegeneralsoilquality.Thosedimensionsoflandplotsareconsideredduringthelandallocationprocess.
Burgess(2007),usinghousehold-leveldataforthetwodistinctprovincesofJiangsuandSichuan,couldnotrejectthehypothesisofuniversalandegalitarianaccesstolandandallocationssatisfyingnutritionalneedsofruralhouseholds.Withequitybeingoneofthepri-maryobjectivesduringthelandallocation,itislikelythataveragelandqualityandfarmsizeare,indeed,negativelycorrelated.Althoughlandallocationarrangementsacrossgeographicregionshavebeensomewhatheterogeneousinrecentyears,theegalitarianmotiveshavegen-erallypersistedtoserveequitypurposesand
CHEN,HUFFMAN&ROZELLE:INVERSEPRODUCTIVITYRELATIONSHIPINCHINA’SAGRICULTURE583
asapracticalsocialsafetynet.Hence,inruralChina,anegativerelationshipseemslikelytoexistbetweenlandareaandlandquality.
III.
DATAANDTHEECONOMETRICMODEL
Inthissection,wedescribethedatasets,thevariablesofinterest,andtheeconometricmodel.TheResearchCenterforRuralEconomy(RCRE),MinistryofAgriculture,China,con-ductedthesurveyusedinthisstudy.A.DescriptionoftheDataSetandMainVariables
Thedataaretakenfromalargecompre-hensivesurveyofChineseruralhouseholds,whichstartedin1986,covered29provinces,andincludedabout20,000households.Sam-pleattritionhasbeenlow,whichisconsideredasamajoradvantageofthedataset(Chen2003).However,thesurveywastemporarilydis-continuedin1992and1994forfinancialrea-sons.Thedatasetforthisstudyconsistsof591randomlyselected(andthenmadeavailabletotheauthors)farmhouseholdsfrom29vil-lagesand9provinces.ThenineprovincesareHebei,Shanxi,Heilongjiang,Liaoning,Anhui,Jiangsu,Shandong,Sichuan,andYunnan.Dataarepooledfrom1995to1999.
Samplingfortheoriginaldatasetwascon-ductedbyprovincialofficesundertheMinistryofAgriculture(seedetailsinBenjamin,Brandt,andGiles2005).Eachprovincialofficefirstselectedequalnumbersofupper,middle,andlowerincomecounties,thenchosearepresen-tativevillageineachcounty.Intotal,40–120householdswererandomlysurveyedwithineachvillage.Villageofficersandaccountantscom-pletedasurveyformongeneralvillagecharac-teristicseveryyear.RCREclaimedthat80%ofthehouseholdsthatenrolledin1986remainedin1999.AlthoughtheoriginalRCREsurveydoesnothaveanexplicitpanelstructure,anear-lierstudybyChen(2003)identifiedthepanelstructurebymatchinghouseholdcharacteristics.MoreinformationisavailablefromtheRCREWebsite.6
Thepaneloffarmsandfarmhouseholdsisunbalancedbecausenotallhouseholdsrespondedtothesurveyeveryyear.Wedeletedasmallnumberofobservations(anegligible
6.http://www.rcre.org.cn/RCRE/GDGC/gdgcposition.htm.
fractionofthesample)becauseofobviousdatarecordingerrorsandmissinginformation.Aftersuchadjustments,westillhaveatotalof2,693validobservations(yearsandfarmhouseholds).Totalfarmoutputisthetotalquantityofallgrainoutputproducedinareportingyear(e.g.,LinandWen1995;Kimhi2006).Specif-ically,totalfarmcropoutputisthesummationbythequantity(weight)ofproductionofrice,sorghum,wheat,andcorn.Becausewearecon-cernedwithproductivityandnotjustaverageyieldofaspecificcrop,wedonotconsidereachcropoutputandareasownseparately.Wealsorefrainfromusingtheaggregatedsalesofgraintoavoidintroducingnoiseduetoinventorychanges.
Farmlandismeasuredastotalcultivatedareaforgraincropsmeasuredatyearend.Similarmeasureshavebeenusedextensivelyintheliterature,see,forexample,FanandZhang(2002)andFan(1997),amongothers.
Table1providessummarystatisticsforvari-ablesusedintheeconometricanalysis.Averageannualgrainoutputperhouseholdis3,176kg.Averagefarmsizeisabout9.9Mu(1Mu=1/15hectare),soaveragefarmproductivityis320kg/Mu.Theaveragenumberofplotsculti-vatedissix.Atypicalfamilyhasoneortwodependents,onemalerurallaborer,andonefemalerurallaborer.Mosthouseholdheadsarebetweenages30and60.Themostcommon-leveleducationcompletedislessthan12years.About4%ofthehouseholdshaveadependentwhohasanurbanhukouregistration.B.ControlVariablesintheProductivityEquation
Besideslandareacultivated,otherexplana-toryvariablesincludedintheproductivityequationarethenumbersoffemaleandmalelaborersinthehousehold,householdhead’sage(proxyforfarmingexperience),andedu-cationoftheindividualinthehouseholdwhohascompletedthemostschooling(proxyformanagerialability,e.g.,Huffman1974;Yang1997a,1997b).Totalnumberofplotscultivatedisalsoincludedasanexplanatoryvariabletocontrolforlandfragmentation(FleisherandLiu1992;WanandCheng2001).Thisvariableisalsosquaredandincludedtopermitnonlinearproductivityeffectsoflandfragmentation.Vil-lagefixedandrandomeffectsareusedtocap-turelocation,climate,regionalirrigationsystem,croppingmix,outputandinputpricedifferences,
584CONTEMPORARYECONOMICPOLICY
TABLE1
ChineseFarmHouseholdData(RCRE):
SummaryStatistics
Variable
MeanSDContinuousvariablesOutput(1000kg)3.1763.628Land(Mu)
9.9148.785Numberofplots
5.8474.603Numberofdependents1.6851.129Malerurallabor1.3510.681Femalerurallabor
1.2120.734
Binaryvariables(1iftrue,0otherwise)Head’sage:<31(reference)0.078Head’sage:31–400.323Head’sage:41–500.383Head’sage:51–600.170Head’sage:>60
0.047Highesteducation:<9years0.320(reference)
Highesteducation:9–11years0.549Highesteducation:12+years0.131Head’seducation:<5years0.082(reference)
Head’seducation:5–8years0.398Head’seducation:9–11years0.407Head’seducation:12+years
0.112Presenceofdependentswithurban0.036(hukou)registrationNumberofobservations
2,693
Note:Authors’tabulationfromtheResearchCenterforRuralEconomySurvey1995–1999.
andmultiplecroppingfactors.Yearfixedeffectsareincludedtocaptureyear-specificeffects,suchasunusualweatherconditions.C.FarmProductivityandFarmSize
Theeconometricmodeloffarmproductiv-ityis(1)
lnyi=α+γlnli+Xiβ+ηi
whereyiistotalannualfarmgrainproductionbytheithfarminghousehold,lisameasureoffarmlandαcultivated,η,γ,andβareparametersXisasetoftobecontrolestimated,variables,andIdeally,isazerowewouldmeanlikerandomtohavedisturbanceameasurementterm.offarmlandthatishomogenous,butworkingwiththeexistingdataprovidesopportunitiesandchallenges.Ifγisequalto1inEquation(1),thena1%increaseinlandarearesultsina1%increaseintotalgrainoutput.Ifγisless(greater)than1,thena1%increaseinlandareawillresultinaless(more)than1%increase
infarmoutputoradecline(increase)infarmproductivity.
Toaccommodatethepanelnatureofthedataset,Equation(1)ismodifiedtobecome(2)
lnyit=α+γlnlit+Xitβ+ξi+ηit
whereηtistheindexoftime,y,l,X,β,γ,household-specificaredefinedasbefore,effectthatandcapturesξisafarm-andpossibleorheterogeneityatthefarm/householdlevel.Itcanbemodeledaseitherrandomorfixedeffects,dependingupontheassumptionofthecorre-lationbetweenξandcovariates(Wooldridge2002).
Inthefixed-effectsspecificationofEquation(2),theindividualfixedeffects(ξi)mightbecorrelatedwithfarmsize(i.e.,thebetweeneffects),andthenγreflectsonlythepartialeffect(withineffects)offarmarea.Therefore,thetotalchangeingrainoutputcanbedecom-posedasthesumofchangesduetoγandtochangesinfixedeffects.Notethatinthefixed-effectsspecification,γnolongerrepre-sentsthefulleffectofloglandareaonlogoutputbutisthewithineffectsidentifiedthroughchangesinlandareaamonghouseholds.Theissuewithusingthefixedeffectsinourcon-textisthattheycouldleadtooverparameteriza-tionandpoorout-of-sampleprediction.Alterna-tively,underrandomeffects,ξi∼N(0,σ2
ξ),weneedξitobeuncorrelatedwiththeregressors,includingthefarmsize.Lamb(2003)suggestedthattherandom-effects7modelisabetterrepre-sentation.Nonetheless,weprovideestimatesofbothrandom-andfixed-effectsmodelstoexam-inetherobustnessofourresults.With5yearsofdataandyearfixedeffects,autocorrelationinthedatacanbeignored,whichwasconfirmedbyhypothesistestingincludingtwoversionsof
7.Hausmantestinthiscaseprefersthefixed-effectsmodel,whichwasnotsurprisinggiventhatwehavealargenumberofhouseholdeffects.WehavenotreliedonaHausmantestinthiscaseforthreereasons.First,asignificantHausmanstatisticusuallysuggestsatensionbetweenthe“within”and“between”estimationapproaches.Fixed-effectsmodelestimatesonlythewithinvariation,whilelumpingthebetweenvariationintothefixedeffects,whichiscriticalinourcase.Thesecondreason,whichrelatestothefirstpoint,isthatthe“exchangeability”testwouldsuggestarandom-effectsmodel(Nerlove2007).Lastly,itisalsowellknownthatHausmanteststendtofavorfixed-effectsmodelduetotheirlargenumberofestimationparameters.Thelargenumberofparametersintroduceoverfittingandposedifficultyforpredictionorextrapolationofthesample.Moreontheprosandconsoffixed-andrandom-effectsmodelscanbefoundinthestudybyHsiao(2007).Weestimatebothmodelstoexaminetherobustnessoftheresults.
CHEN,HUFFMAN&ROZELLE:INVERSEPRODUCTIVITYRELATIONSHIPINCHINA’SAGRICULTURE585
Durbin-Watsontest,thatis,theBhargavaetal.(1982)testandBaltagi-Wu(1999)LBItest.D.IdentificationStrategy
Equation(2)maybeviewedasthe“truemodel”oftherelationshipbetweentotalfarmoutputandfarmsize.However,whatweobserveisameasureofcultivatedareathatisofheterogeneousquality,whichcreatespotentialmeasurementerrorproblemswhenthemeasure-menterroriscorrelatedwithoneormoreregres-sors.Onesolutionistoinstrumentfarmsize(Benjamin1995).Validinstrumentsshouldbecorrelatedwithfarmsize(cultivatedarea)butnotwiththedisturbancetermintheproductivityequation(Greene2005).
Theinstitutionalstructureofthefarmlandallocationprocessbyruralvillagecouncilssug-geststhatacorrelationexistsbetweenaruralhousehold’slandareaandnutritionalneeds(Burgess2007).Furthermore,thenumberofhouseholddependentsisstronglycorrelatedwithtotalnutritionalneedsofthehousehold(Burgess2007),andthereisnoreasontobelievethatthenumberofdependents8inahouseholdisrelatedtofarmproductivity.Inaddition,theresidencehukouregistrationsysteminChinaclassifiesres-identsintotwotypes:urbanandrural.Ruralresidentsareentitledtoreceivefarmlandsbuturbanresidentsarenot.Inthedataset,about3.6%oftheruralhouseholdshaveatleastonedependentwhohasurbanhukouregistration.However,weareunabletodeterminewhetherthesedependentsalsohaveanentitlementtofarmlandintheruralcommunity.Manycollegestudentsfromruralareasareregisteredasanurbanresidentbutcontinuetoreceivelandfromtheirruralhomecommunities.
Wearguethatthetotalnumberofdependentsinthehouseholdandthepresenceofadependentwithurbanhukouregistrationarevalidinstru-mentsforfarmlandarea.Thevariationinthenumberofdependentsandhukouregistrationservesasa“naturalexperiment”thatcanbeusedtoidentifythetruerelationshipbetweenfarmsizeandproductivity(AngristandKrueger2001,73).Thenumberofdependentsrelateslargelytothenumberofchildrenandofelderlydependents.Fertilitydecisionsweremadebasedonhowstrictlyfamilyplanningpolicieswereenforcedandthegenderofthefirstborn,among
8.Dependentsincludebothchildrenandelders,whodonotundertakefarmingactivitiesduetophysicallimitations.
otherfactors.9Thenumberofelderlydepen-dentsisdeterminedbymortalityshocks.Thereisnoreasontobelievethatmortalityshockandthefactorsrelatedtofertilitydecisionswouldberelatedtocurrentshocksinfarmproduc-tivity.Thenumberofdependentsrelatestotheeffectivelandthatisallocatedtoahouseholdtomeetthehousehold’snutritionalneeds(Burgess2007)butdoesnotdirectlyrelatetotheagricul-turalproductivityshocks.AsimilarstrategywasusedbyBenjamin(1995).
Thepresenceofahouseholdmemberwithurbanhukouregistrationderivesfromtwotypesofevents,beingadmittedtoacollegeandbeingrecruitedforapositioninstate-ownedenter-prisesorgovernmentagencies.10Thevariationassociatedwiththeseeventsdoesnotcorrelatetolandquality,northegrainproductionbutdoesreducethelikelihoodofreceivinglandfromthehomecommunity.We,hence,usethenumberofdependentsandadummyvariableindicat-ingpresenceofahouseholdmemberwithurbanhukouregistrationasthebasicsetofinstrumentsforland.Thequalityoftheinstrumentswillalsoundergoabatteryofstatisticaltests.
Notethattheinstrumentsarerelatedtolandareas,whichisintendedanddesired.Inexam-iningtherelationshipbetweenproductivityandfarmsize,wewouldbeinterestedtohaveameasurementoflandareathatishomogenousintermsoflandquality.Essentially,theinstru-mentsareusedtopredictahomogenousmeasureoflandareatobeusedinthemainestimationequation.Argumentsagainstusingtheinstru-mentsincludethatmoredependentsmightpushthehouseholdlaborstoworkharderandthatdependentsmayworkonthefarmtoincreaseproductivity.However,themainimpactofthedependentsongrainoutput(whichisthevari-ableofinterestinthisstudy)seemsthroughtheincreasedallocationof(qualityadjusted)land.Theincreasedlivingcostinducedbymoredependentsmightbecoveredthroughworksoncashcropsoroff-farmactivities.Nonethe-less,theoveridentificationtestinSectionIVaddressestheissueofthecorrelationbetweeninstrumentsanddisturbancetermsinthemainestimationequation.
9.InruralChina,thefamilyplanningpolicyallowsasecondchildifthefirstbornisfemale.Householdsofminorityethnicities(otherthanHan)maybeexemptedfromthe“one-child”policy.
10.Notethatthisreferstothecaseduringthestudyperiod.Theroleofhukouregistrationisdiminishingovertime.Severaladministrativeregionshaveattemptedtoabolishthehukouregistrationorlimititsrole.
586CONTEMPORARYECONOMICPOLICY
E.WhoRulestheRoost,andonWhat?RevisitingRolesofHouseholdHeadandVillageOfficers
Wealsoattemptedtoexamineanexpandedsetofinstruments.China’sagriculturehasundergonemanydramatictransitionsandhasitsownuniquefeaturesasidefromthelandallocationsystem.Theroleofeducationinagriculturalproductionhasbeenscrutinizedintensivelyintheliterature,see,forexample,Huffman(2001)forareviewofliterature.ForChina’sagriculture,inparticular,Yang(1997a,1997b)foundthatthehighestlevelofedu-cationamongallhouseholdmembershasastrongercorrelationwithagriculturaloutputthanhouseholdhead’seducationoraverageyearsofschoolingacrosshouseholdmembers.Hehypothesizedthatcollectivizeddecisionmak-ingwasusedinagriculturalhouseholdinruralChina,andhence,thehighestlevelofeducationmattersmostforproductiondecisions.How-ever,duringmeetingswiththevillagecouncildealingwithlandrequests,thehouseholdheadnormallyrepresentshisorherhousehold.Theexactamountandtypeofthelandahouseholdreceiveswouldthusdependonhisorhernego-tiatingskillsaswellashisorheraccesstoandabilitytoprocessinformation.Thus,thehouse-holdhead’seducationseemslikelytobemoreimportantinthelandacquisitionprocessthanindeterminingfarmproductivity.Thedummyvariablesindicatinghouseholdhead’seducationareincludedintheexpandedsetofinstrumentsforfarmsize.
Theroleofahouseholdhavingamemberwhoisavillageofficercouldaffectaccesstofarminputs,forexample,industrialinputs,andcommunityproductivecapital(Brandtetal.2002;WalderandZhao2006).However,withthedramaticdevelopment,expansion,andlib-eralizationofindustrialinputmarkets,villagecouncilshavelostmostoftheircontroloverindustrialinputstothemarketforces.Xu(1999)hasattributedalargeportionoffarmproductiv-ityincreaseintheearly1990stoevolvingindus-trialinputmarkets.Rentseekingamongvillageofficersmightcontinuetoexistbutshiftedtooff-farmeconomies(ChenandRozelle1999;WalderandZhao2006).However,underHRS,ahouseholdwithamemberholdingavillageofficerpositionmaybebetterinformedaboutthequalityoflandplotsbeingallocatedandmayalsohavebetterbargainingpowerornego-tiatingskills(RozelleandLi1998).Hence,
membershiponthevillagecouncilisanotherpotentialinstrumentforeffectivefarmsize.
IV.
RESULTS
A.BaselineResults
ThebaselineestimateoftheproductivityequationisreportedinTable2.Bothrandom-andfixed-effectsmodelsoftheoutputequationshowthattheestimateofγissignificantlylessthan1.Hence,ourresultssuggestaninverserelationshipbetweenfarmsizeandgrainoutputamongChinesefarmhouseholds.
TABLE2
RelationshipbetweenOutputandFarmSizefromtheRCREDataSet(DependentVariable:LogarithmofOutputQuantity;N=2,693)
RandomEffects
FixedEffectsLand(logarithm)0.841∗∗∗(0.014)
0.742∗∗∗(0.022)Malerurallabor0.016∗(0.009)0.026∗(0.015)
Femalerurallabor0.015∗(0.008)0.031∗∗
(0.013)Head’sage:<31(reference)Head’sage:31–40−0.003(0.023)0.008(0.041)Head’sage:0.001(0.022)
0.010
(0.041)41–50Head’sage:51–60
−0.025(0.024)−0.039(0.044)Head’sage:>600.011(0.033)−0.049
(0.060)
Highest
education:<9years
(reference)Highest0.007(0.013)−0.004(0.023)education:9–11years
Highest
0.012(0.020)
0.039
(0.048)
education:12+years
Villageofficer0.050∗∗(0.024)−0.010(0.044)
Numberofplots0.018∗∗∗(0.004)0.032∗∗∗(0.008)Numberofplots−0.001∗∗∗(0.000)−0.001∗∗∗(0.000)
squaredVillagefixedYesNo
effects
YearfixedeffectsYesYes
Constant5.760∗∗∗
(0.044)6.036∗∗∗(0.063)
σu
0.0750.379σe
0.2290.229ρ(u,v)0.098
0.733
Note:Thestatisticaltestandinferenceofthecoefficientoflandisbasedonthetestofthenullhypothesisthatthecoefficient∗∗∗
isequalto1(not0asin∗∗thestandardtvalue).∗Significantatthe1%level;significantatthe5%level;significantatthe10%level.
CHEN,HUFFMAN&ROZELLE:INVERSEPRODUCTIVITYRELATIONSHIPINCHINA’SAGRICULTURE587
Thecoefficientestimatesofcontrolvariablesagreewithourexpectationingeneral.Thereisanonlinearcorrelationbetweenthenumberofplotsandoutput.Itsinterpretation,however,iscomplicatedbythefactthatthenumberofplotsmayrelatetobothfarmsizeandlandfragmentation.Asarobustnesscheck,weexam-inedthespecificationwithoutthenumberofplotvariablesandobtainedqualitativelyconsistentresults.
B.InstrumentalVariableEstimation:Necessity,Appropriateness,andResults
Althoughthetheoryoutlinedinthepre-vioussectionssuggestspotentialgainsfrominstrumentalvariableestimation,weneedtoexaminethevalidityandqualityoftheinstru-ments.First,thespecificationtestofHausman(1978)isusedtohelpdeterminewhethertheordinaryleastsquare(OLS)estimatesarecon-sistentoraninstrumentalvariable(IV)estimatorisbetter.WefollowLamb’s(2003)versionofthistest,whichreducestoazstatistic.TheteststatisticfortheHausmantestimpliesthatOLSestimationisinconsistentandthatIVisbetter.Second,dotheorthogonalityconditionsbetweentheinstrumentsandtheerrortermhold,whichisanoveridentificationtest(Ruud2000,573).Thesamplechi-squarestatisticsanditsstatis-ticalsignificanceforthistestarereportedinTable3;andwecannotrejectthenullhypothe-sisthattheorthogonalityconditionholds.Third,
TABLE3
IVsEstimationoftheRelationshipbetweenOutputandFarmSize(N=2,693)
RandomEffects
Regressors
Land(logarithm)MalerurallaborFemalerurallabor
Head’sage:<31(reference)Head’sage:31–40Head’sage:41–50Head’sage:51–60Head’sage:>60
Highesteducation:<9years(reference)
Highesteducation:9–11yearsHighesteducation:12+yearsVillageofficerNumberofplots
NumberofplotssquaredNumberofdependents
Presenceofdependentswithurbanregistration
VillagefixedeffectsYearfixedeffectsConstantσuσe
ρ(u,v)
Bhargavaetal.(1982)Durbin-WatsontestBaltagi-Wu(1999)LBI
Hausmanspecificationtest(zvalue)Overidentificationtest:χ2(2)First-stagepartialR2
Hahn-Hausmanteststatistics
ln(Land)0.176∗∗∗0.056∗∗0.071∗∗∗0.053∗−0.0230.012−0.0090.064∗∗0.081∗∗∗−0.002∗∗∗0.136∗∗∗−0.101∗∗∗
(0.011)
0.162∗∗∗(0.010)
(0.028)(0.028)(0.030)(0.041)(0.017)(0.025)(0.031)(0.004)(0.000)(0.007)(0.037)
ln(Output)0.977−0.003−0.002−0.020−0.013−0.0280.0210.0020.0070.0360.005−0.000∗
(0.040)(0.010)(0.010)(0.023)(0.023)(0.024)(0.033)(0.014)(0.020)(0.025)(0.005)(0.000)
FixedEffects
ln(Land)0.157∗∗∗
(0.015)
0.119∗∗∗(0.012)−0.048−0.008−0.019−0.0850.0230.0560.0460.088∗∗∗−0.002∗∗∗0.094∗∗∗0.012
(0.040)(0.040)(0.042)(0.058)(0.023)(0.046)(0.042)(0.007)(0.000)(0.009)(0.046)
ln(Output)0.992−0.0070.0070.0130.006−0.029−0.011−0.0130.029−0.0210.008−0.001
(0.104)(0.021)(0.016)(0.042)(0.043)(0.045)(0.064)(0.024)(0.049)(0.045)(0.013)(0.000)
YesYes
0.172∗∗∗(0.056)YesYes
5.708∗∗∗(0.047)0.0750.2370.0911.7252.2883.653∗∗∗0.2690.1501.394
NoYes
1.062∗∗∗(0.057)0.7320.222∗∗∗0.916
NoYes
5.728∗∗∗(0.140)0.3380.236∗∗∗0.673
2.468∗∗∗
Note:Thestatisticalinferenceofthecoefficientoflandisbasedonthenullhypothesisthatthecoefficientisequalto1.∗∗∗Significantatthe1%level;∗∗significantatthe5%level;∗significantatthe10%level.
588CONTEMPORARYECONOMICPOLICY
aretheinstrumentssufficientlycorrelatedwiththeendogenousregressor?Thisisrelatedtothediscussionintheliteratureof“weakinstru-ments,”forexample,StaigerandStock(1997)andNelsonandStartz(1990).Weakinstrumentsmaymakethesecond-stageinferenceinvalid.Bound,Jaeger,2andBaker(1995)suggestedthatthepartialRandtheFstatisticoftheidentify-inginstrumentsinthefirst-stageestimationareusefulindicatorsofthequalityofinstruments.Asthefirst-stageR2(thepartialR2ifthereareincludedexogenousvariables)increases,theinstrumentalvariableestimationbiasbecomessmaller.Inthisstudy,thepartialR2(addedexplanatorypowerusingtheexcludedinstru-ments)inthefirststageis0.15,whichisrea-sonableconsideringthelargesamplesize,themicronatureof11thedataset,andthatweusefewinstruments.Third,anewtestproposedbyHahnandHausman(2002)addressesexplicitlytheissuewhethertheconventionalinstrumentalvariableasymptoticsarereliable.Itcomparesthedifferenceintheconventional2SLSestimateofthecoefficientoftheright-handsideendogenousvariablewiththereverse2SLSestimateofthesameunknownparameter.TheHahn-Hausmanstatistic,whichhasatdistributionunderthenullhypothesis,is1.394—smallerthanthecrit-icalvaluesatthe5%significancelevel.Hence,wecannotrejectthenullhypothesisthattheinstrumentalvariableestimatorprovidesreliableinferences.C.MainResults
TheinstrumentalvariableestimatesarereportedinTable3.Thefirst-stagefixed-effectsresultsprovideinterestingimplications.First,havingahouseholdmemberwhoisavil-lageofficerincreasestheamountoffarmlandallocatedtothehousehold.Householdheadswhoareof41–50yearsofageandthosewithmoredependentsreceivemoreland.Householdswhohaveamemberholdinganurbanhukouregistrationreceivelessland(alsoseeBurgess2007).
Thesecond-stagefixed-effectsresultsshowthattheestimatedcoefficientofthelogarithmofcultivatedlandareahasincreasedfromthebase-linemodelsof0.74withvillagefixedeffects(Table2)to0.992inTable3.Similarlarge
11.PartialR2indicatestheaddedexplanatorypoweroftheidentifyinginstruments(Halletal.1996).Itisusedtodeterminewhetherthesetofinstrumentsis“weak,”alongwiththeHahn-Hausmantest.
increasesarereportedforthemodelfittedwithrandomeffects.Hence,theinverserelationshipbetweentotalfarmoutputandtotalcultivatedareadisappearsoncelandheterogeneityiscon-trolledforusinganinstrumentforlandarea.D.RobustnessCheck
Weinvestigatetherobustnessoftheresultswithbothfixed-andrandom-effectsmodelsaswellaswithanexpandedsetofinstru-ments.Thefixed-andrandom-effectsIVmod-elsprovidesimilarresults.TheIVestimationwiththeexpandedsetofinstrumentsalsopro-ducesconsistentresults,asshowninTable4.Aninterestingobservationisthatthehouseholdhead’seducationandthevillageofficerdummyvariabledoappeartobeanimportantfactorinthelandallocationprocessinruralChinabutnotintheoutputequation(thelatterwasshownintheresultforthebaselinesetofinstruments).ThissubstantiatestheconclusionofYang(1997a,1997b)onthecollectivedeci-sionmakingwithinChineseruralhouseholdsandthathouseholdheadsandvillageofficersplaycriticalrolesinthelandallocationprocess.Wealsoexaminedtherelationshipbetweentotalgrainoutputandcultivatedlandareabyreplacingthenumberofdependentswithitslog-arithmandthemainconclusiondidnotchange.ThesespecificationsconsistentlysuggestthataninverserelationshipinChina’sagriculturedoesnotexist—outputisproportionaltofarmlandareacultivated.
V.
DISCUSSION
WhenDeiningerandFeder(2001)summa-rizedseveralstudiesthatdocumentedaninverserelationshipbetweenfarmsizeandproductiv-ity,theyarguedthatsupervisioncostsforhiredlaborthatcomewithalargerfarmarepartic-ularlylargeduetospatialdispersion,andthus,contributetotheinverserelationship.ThiscouldbeinterpretedasonereasonwhyChina’sagri-culturewastransformedfromcollectivefarmingtoHRSinthe1980s.Microeconomictheorysuggeststhatanoptimumsizeexistsformostproductionprocessesandinstitutionalarrange-ments.Empiricalevidence,asbyDeiningerandFeder(2001),indicatesthattheoptimalfarmsizeinmostdevelopingcountries,giventheexistenceofimperfectinput/output/creditmar-kets,lowrealwage,staticagriculturaltechnolo-gies,andlandheterogeneities,issmallrelative
CHEN,HUFFMAN&ROZELLE:INVERSEPRODUCTIVITYRELATIONSHIPINCHINA’SAGRICULTURE589
TABLE4
InstrumentalVariablesEstimationoftheRelationshipbetweenOutputandFarmSize(Expanded
SetofInstruments;N=2,693)
RandomEffects
Regressors
Land(logarithm)MalerurallaborFemalerurallabor
Head’sage:<31(reference)Head’sage:31–40Head’sage:41–50Head’sage:51–60Head’sage:>60
Highesteducation:<9years(reference)
Highesteducation:9–11yearsHighesteducation:12+yearsNumberofplots
NumberofplotssquaredVillageofficer
Head’seducation:<5years(reference)
Head’seducation:5–8yearsHead’seducation:9–11yearsHead’seducation:12+yearsNumberofdependents
Presenceofdependentswithurbanregistration
VillagefixedeffectsYearfixedeffectsConstantσuσe
ρ(u,v)
Bhargavaetal.(1982)Durbin-WatsontestBaltagi-Wu(1999)LBI
Hausmanspecificationtest(zvalue)Overidentificationtest:χ2(2)First-stagepartialR2
Hahn-Hausmanteststatistics
ln(Land)0.173∗∗∗0.051∗0.067∗∗0.050−0.010
(0.011)
0.160∗∗∗(0.010)
(0.029)(0.028)(0.031)(0.043)
ln(Output)0.986−0.004−0.003−0.019−0.012−0.0280.0220.0020.0090.005−0.000∗
(0.039)(0.010)(0.010)(0.023)(0.023)(0.025)(0.034)(0.014)(0.020)(0.005)(0.000)
FixedEffects
ln(Land)0.157∗∗∗
(0.015)
0.119∗∗∗(0.012)−0.050−0.017−0.035−0.099∗0.0340.088∗0.087∗∗∗−0.002∗∗∗0.053(0.033)(0.037)(0.068)(0.009)(0.046)
NoYes
1.108∗∗∗(0.066)0.7330.2220.916
NoYes
5.756∗∗∗(0.136)0.3380.2340.675
(0.041)(0.041)(0.043)(0.059)(0.023)(0.048)(0.007)(0.000)(0.042)
ln(Output)0.968−0.0040.0100.0130.006−0.029−0.015−0.0110.0300.010−0.001∗
(0.099)(0.020)(0.016)(0.042)(0.042)(0.045)(0.063)(0.024)(0.049)(0.012)(0.000)
0.023(0.019)0.022(0.037)0.080∗∗∗(0.005)−0.002∗∗∗(0.000)0.064∗∗(0.031)0.088∗∗∗(0.027)0.049∗(0.029)0.032(0.044)0.136∗∗∗(0.007)−0.103∗∗∗(0.037)YesYes
0.118∗(0.061)
−0.009−0.059−0.158∗∗0.094∗∗∗0.019
YesYes
5.703∗∗∗(0.047)0.0780.2360.0981.7252.2883.924∗∗∗0.5390.1600.752
2.339∗∗∗
Note:Thestatisticalinferenceofthecoefficientoflandisbasedonthenullhypothesisthatthecoefficientisequalto1.∗∗∗
Significantatthe1%level;∗∗significantatthe5%level;∗significantatthe10%level.
totheoptimalsizeoffarminhighwage,techni-callyadvanced,developedcountriessuchastheUnitedStates.
InChina,weseeacomplicatedpicture.First,Chinahasaverylargeruralpopulationrelativetotheamountofarableland.Thearablelandperruralpersoninthelater1990swasabout0.144hectare,comparedto0.221inIndia,2.729intheUnitedStates,andaworldaverageof0.426hectare.Second,arablelandinChinaiscollectivelyownedbyruralcommunitiesinsteadofindividualhouseholds.LargeprivatelyownedfarmsrarelyexistinChina,althoughasmallnumberofprivatelyoperatedfarmshaveemerged.Onewouldexpectlargefarmstobe“specialized”andtohavesubleasedlandfromthecommunityorotherhouseholds.Communi-tiesaremorelikelytoputupforleaseexcesslandwhichisofpoorerthanaveragequality,andtotheextentthathouseholdsleaseoutland,itismostlikelylessproductiveorremotelylocated.Thismaycontributetothespuriousinverse
590CONTEMPORARYECONOMICPOLICY
relationshipbetweenproductivityandfarmsizeinsomestudies.Third,EastandSouthChinahaveseenaneconomicboostinthelastfewdecades,andmanypreviouslyrurallaborersarenowemployedinindustrialorservicesectors.Alandrentalmarketmightsuccessfullytrans-ferlandsfromhouseholdsthatarelessinvolvedinfarmingtothosethataremore“specialized”infarming.TherapideconomicdevelopmentinChinainthe1990smayhaveimprovedthefunc-tioningoffarminputandoutputmarketsandmostlikelycontributestotheweakeningofanyinverserelationshipbetweenfarmproductivityandfarmsizethatmayhaveexistedearlier.AsChina’sagriculturebecomesmoremecha-nizedandtheinputsectorstartstoproduceasteadystreamofnewtechnologies,largerfarmsmayhaveacomparativeadvantageoversmallerfarms.
Ourresultsshowthatlandheterogeneitycontributestotheobservedinverserelation-ship.This,aswellasotherstudies(Benjamin1995;Carter1984;DeiningerandFeder2001),pointstoanimportantconclusion:theinverserelationshipbetweenfarmsizeandoutputperunitoflandisnotinherenttoChina’sagricul-turebutratheraconsequenceofheterogeneouslandunevenlydistributedacrosshouseholdsandintroducedduringthelandallocationprocess,aswellas(labor)marketimperfections,andunobservedfactors.Therefore,apublicpolicyofbreakinguplargefarmsisnotcurrentlyjus-tified(DeiningerandFeder2001).Thehiddenunemploymentproblemcanbeaddressedbygeneraleconomicdevelopmentandinvestmentsinruraleducation(Huffman2001;HuffmanandOrazem2007).Amechanismthatconsolidateslandthroughmoreactivelandrentalmarkettoexploitthebenefitsofmoreadvancedtechnol-ogyandtosharesuchbenefitsbetweenlandown-ersandfarmersisneeded.
Whennewagriculturaltechnologiesarebeingdevelopedanddisseminatedtofarmers,theadoptiondecisionincludesafixedcostoflearningaboutandexperimentingwiththenewtechnology.Thereturnstoadoptionarepositivelyrelatedtothesizeofthefarmingoperationandthelengthofthefarmer’splan-ninghorizon.Hence,otherfactorsbeingequal,largefarmshaveacomparativeadvantageoversmallerfarmswhenagriculturaltechnologiesaredynamic(HuffmanandEvenson2001).Tradi-tionally,Chinesefarmshavebeensmall.How-ever,villagecouncils’auctionsofexcesslandsandtheemerginglandrentalmarkethavemadeitpossibleforsomefarmerstotakeinmoreland,whichpermitsusingnewandlarge-scalemachines.Thisstudypresentssomeevidencethat,atpresent,thereisnoneedtodiscouragelargefarms,becausetheinverserelationshipscouldbeduetounobservedlandquality.AboldconjectureisthatChina’spotentialtoimproveagriculturalproductivitymayberealizedifthefragmentedlandparcelscanbeconsolidatedintolargerfarms,wheremoderntechnologiescanbeeasilyapplied.Deininger,Jin,andYu(2007)foundthataccesstolandoffersruralChi-nesehouseholdsprotectionagainstidiosyncraticrisks.DeiningerandJin(2005)showthatdecen-tralizedlandrentmarketsmightcontributetobothequityandefficiencyandhaveadvantagesoveradministrativereallocation.
Theimportanceofthisstudyforpolicymak-ers,weargue,liesintherejectionofanobservedinverseproductivityrelationship.Notonlyislandconsolidationthroughinstitutionssuchaslandrentalmarketsapressingpolicyissueincoastalareas,buttherearealsoreportssuggest-ingthatout-migrationhasleftonlywomenandtheelderlytoworkonthelandininlandvillages.Thishasledinsomecasestolandconsolidationandhigheragriculturalproductivity.
AlthoughweprovideeconometricevidencethatunobservedlandqualityexplainstheinverseempiricalrelationshipbetweenfarmsizeandoutputinChina,otherpossibleexplanationsincludemarketimperfectionsandinstitutionalfactors.BenjaminandBrandt(2002)suggestthatadministrativeallocationoflandmaycon-tributetotheinverserelationship.Thisarti-clediffersfromthatofBenjaminandBrandt(2002)inthatwehypothesizedacorrelationbetweenlandqualityandfarmsizethatwasintroducedduetoequityconcerns.Interestingly,theresultfromourfixed-effectsmodel,whichestimatestheimpactofchangesinlandareaonchangesintotalfarmoutput,seemstobeinlinewithBenjaminandBrandt’sconclusionthatlandreallocationalleviatestheinefficiencyoflaboruse.OurpolicyrecommendationtoimprovelandrentalmarketissimilartothatofBenjaminandBrandt(2002),whoarguedfortheimportanceofwell-functioningfactormar-kets.ExaminingtheeffectoffarmsizeontotalfactorproductivityasAhearn,Yee,andHuffman(2002)mightbeausefulfuturelineofresearch.
CHEN,HUFFMAN&ROZELLE:INVERSEPRODUCTIVITYRELATIONSHIPINCHINA’SAGRICULTURE
591
VI.
CONCLUSIONS
Thisarticlehasexaminedtheempiricalrela-tionshipbetweenfarmproductivityandfarmsizeinChina’sagriculture.GiventhatthelocalruralvillagecouncilistheinstitutionthatholdsthemajorityofChina’sfarmlandandmakeslandallocationdecisions,wechoosetouseaninstru-mentalvariableestimationproceduretoexamineindetailtherelationshipbetweenfarmsizeandproductivityinsmall-scaleagriculture.ThemaindatasetthatweusedarefromtheRCREsurveyinthelate1990s.Intherawdata,weobservedaninverserelationshipbetweentotalgrainout-putandcultivatedarea.However,afterweusedeconometricmethodstocontrolforunobservedlandquality,theinverseempiricalrelationshipbetweengrainoutputandcultivatedareadisap-pears.Hence,farmoutputisproportionaltofarmsizecurrentlyinChina,andthereisnourgencytodeveloppoliciesthatwouldreducefarmsize.
REFERENCES
Ahearn,M.C.,J.Yee,andW.E.Huffman.“R&D,Pro-ductivity,andStructuralChangeinU.S.Agriculture,1960–1996.”TheU.S.DepartmentofAgriculture,EconomicResearchServiceReport,July,2002.
Angrist,J.D.,andA.B.Krueger.“InstrumentalVariables
andtheSearchforIdentification:FromSupplyandDemandtoNaturalExperiments.”JournalofEconomicPerspectives,15(4),2001,69–85.
Assuncao,J.J.,andM.Ghatak.“CanUnobservedHetero-geneityinFarmerAbilityExplaintheInverseRelation-shipbetweenFarmSizeandProductivity?”EconomicsLetters,80(2),2003,189–94.
Baltagi,B.H.,andP.X.Wu.“UnequallySpacedPanel
DataRegressionswithAR(1)Disturbances.”Econo-metricTheory,15(6),1999,814–23.
Barrett,C.B.“OnPriceRiskandtheInverseFarmSize-ProductivityRelationship.”JournalofDevelopmentEconomics,51(2),1996,193–215.
Benjamin,D.“CanUnobservedLandQualityExplainthe
InverseProductivityRelationship?”JournalofDevel-opmentEconomics,46(1),1995,51–84.
Benjamin,D.,andL.Brandt.“Land,FactorMarkets,and
InequalityinRuralChina:HistoricalEvidence.”Explo-rationsinEconomicHistory,34(4),1997,460–94..“PropertyRights,LaborMarkets,andEfficiencyinaTransitionEconomy:TheCaseofRuralChina.”CanadianJournalofEconomics—RevueCanadienneDEconomique,35(4),2002,689–716.
Benjamin,D.,L.Brandt,andJ.Giles.“TheEvolutionof
IncomeInequalityinRuralChina.”EconomicDevel-opmentandCulturalChange,53(4),2005,769–824.Berry,R.A.,andW.R.Cline.AgrarianStructureandPro-ductivityandDevelopingCountries.Baltimore,MD:JohnsHopkinsUniversityPress,1979.
Bhalla,S.S.,andP.L.Roy.“Mis-SpecificationinFarm
ProductivityAnalysis—TheRoleofLandQuality.”OxfordEconomicPapers,40(1),1988,55–73.
Bhargava,A.,L.Franzini,andW.Narendranathan.“Serial
CorrelationandtheFixedEffectsModel.”ReviewofEconomicStudies,49(4),1982,533–49.
Bound,J.,D.A.Jaeger,andR.M.Baker.“Problemswith
InstrumentalVariablesEstimationWhentheCorre-lationbetweentheInstrumentsandtheEndogenousExplanatoryVariableIsWeak.”JournaloftheAmeri-canStatisticalAssociation,90,1995,443–50.
Brandt,L.,J.Huang,G.Li,andS.D.Rozelle.“LandRights
inRuralChina:Facts,FictionsandIssues.”TheChinaJournal,47,2002,67–97.
Burgess,R.“Land,WelfareandEfficiencyinRuralChina.”
LondonSchoolofEconomics.Mimeo,1997.
Burgess,R.“LandandWelfare:TheoryandEvidencefrom
China.”LondonSchoolofEconomics.Mimeo,2001..“Mao’sLegacy:AccesstoLandandHungerinModernChina.”PresentedattheJointDevelopmentEconomicsSeminarorganisedbyIDPMandEco-nomics,UniversityofManchester,March20,2007.Carter,M.R.“IdentificationoftheInverseRelationship
betweenFarmSizeandProductivity:AnEmpiricalAnalysisofPeasantAgriculturalProduction.”OxfordEconomicPapers,36(1),1984,131–45.
Chayanov,A.V.inTheTheoryofPeasantEconomy,edited
byD.Thorner,B.Kerblay,andR.E.F.Smith.Home-wood,IL:IrwinPublishers,1926.
Chen,H.,andS.D.Rozelle.“Leaders,Managers,andthe
OrganizationofTownshipandVillageEnterprisesinChina.”JournalofDevelopmentEconomics,60(2),1999,529–57.
Chen,J.“EconomicDevelopmentandtheEvolutionof
BackyardLivestockProduction:ACaseStudyofHogProductioninChina.”Ph.D.dissertation,UniversityofCalifornia,Davis,2003.
Chen,K.,andC.Brown.“AddressingShortcomingsin
theHouseholdResponsibilitySystem—EmpiricalAnalysisoftheTwo-FarmlandSysteminShandongProvince.”ChinaEconomicReview,12(13),2001,280–92.
Deininger,K.,andH.Binswanger.“TheEvolutionofthe
WorldBank’sLandPolicy:Principles,Experience,andFutureChallenges.”TheWorldBankResearchObserver,14(2),1999,247–76.
Deininger,K.,andG.Feder.“LandInstitutionsandLand
Markets,”inHandbookofAgriculturalEconomics,Vol.1,editedbyBruceGardner,andG.ordenRausser.NewYork:ElsevierScience,2001,287–331.
Deininger,K.,andS.Jin.“ThePotentialofLandRental
MarketsintheProcessofEconomicDevelopment:EvidencefromChina.”JournalofDevelopmentEco-nomics,78(1),2005,241–70.
.“LandSalesandRentalMarketsinTransition:EvidencefromRuralVietnam.”OxfordBulletinofEconomicsandStatistics,70(1),2008,67–101.
Deininger,K.,S.Jin,andX.Yu.“RiskCopingandStar-vationinRuralChina.”AppliedEconomics,39(11),2007,1341–52.
Deolalikar,A.B.“TheInverseRelationshipbetweenPro-ductivityandFarmSize:ATestUsingRegionalDatafromIndia.”AmericanJournalofAgriculturalEco-nomics,63(2),1981,275–79.
Dong,X.-Y.“Two-TierLandTenureSystemandSustained
EconomicGrowthinPost-1978RuralChina.”WorldDevelopment,24(5),1996,915–28.
Fan,S.“ProductionandProductivityGrowthinChinese
Agriculture:NewMeasurementandEvidence.”FoodPolicy,22(3),1997,213–28.
Fan,S.,andX.Zhang.“ProductionandProductivityGrowth
inChineseAgriculture:NewNationalandRegionalMeasures.”EconomicDevelopmentandCulturalChange,50(4),2002,819–38.
592CONTEMPORARYECONOMICPOLICY
Fleisher,B.M.,andY.Liu.“EconomiesofScale,PlotSize,
Human-Capital,andProductivityinChineseAgricul-ture.”QuarterlyReviewofEconomicsandFinance,32(3),1992,112–23.
Greene,W.H.EconometricAnalysis.5thed.UpperSaddler
River,NY:PrenticeHall,2005.
Hahn,J.,andJ.Hausman.“ANewSpecificationTestfor
theValidityofInstrumentalVariables.”Econometrica,70(1),2002,163–89.
Hall,A.R.,G.D.Rudebusch,andD.W.Wilcox.“Judging
InstrumentRelevanceinInstrumentalVariablesEsti-mation.”InternationalEconomicReview,37(2),1996,283–98.
Hausman,J.“SpecificationTestsinEconometrics.”Econo-metrica,46(6),1978,1251–71.
Heltberg,R.“RuralMarketImperfectionsandtheFarm
Size–ProductivityRelationship:EvidencefromPak-istan.”WorldDevelopment,26(10),1998,1807–26.Hsiao,C.“PanelDataAnalysis—AdvantagesandChal-lenges.”Test,16(1),2007,1–22.
Huffman,W.E.“DecisionMaking:TheRoleofEduca-tion.”AmericanJournalofAgriculturalEconomics,56,1974,85–97.
.“HumanCapital:EducationandAgriculture,”inHandbookofAgriculturalEconomics,Vol.IA,editedbyBruceL.Gardner,andGordonC.Rausser.NorthHolland,Amsterdam:ElsevierScience,2001,333–81.Huffman,W.E.,andR.E.Evenson.“StructuralandPro-ductivityChangeinU.S.Agriculture.”AgriculturalEconomics,24(2),2001,1950–82.
Huffman,W.E.,andP.F.Orazem.“TheRoleofAgricul-tureandHumanCapitalinEconomicGrowth:Farmers,Schooling,andHealth,”inHandbookofAgriculturalEconomics,Vol.3,editedbyRobertEvenson,andPrabhuPingali.NorthHolland,TheNetherlands:Else-vierScience,2007,2281–341.
Kimhi,A.“PlotSizeandMaizeProductivityinZambia:
IsThereanInverseRelationship?”AgriculturalEco-nomics,35(1),2006,1–9.
Lamb,R.L.“InverseProductivity:LandQuality,Labor
Markets,andMeasurementError.”JournalofDevel-opmentEconomics,71(1),2003,71–95.
Lin,J.Y.,andJ.G.Wen.“China’sRegionalGrainSelf-SufficiencyPolicyandItsEffectonLandProductiv-ity.”JournalofComparativeEconomics,21(2),1995,187–206.
Nelson,C.R.,andR.Startz.“TheDistributionoftheInstru-mentalVariablesEstimatorandItsT-RatioWhentheInstrumentIsaPoorOne.”JournalofBusiness,63(1),1990,S125–40.
Nerlove,M.“CommentsOn:PanelDataAnalysis—
AdvantagesandChallenges.”Test,16(1),2007,42–46.
Rozelle,S.D.,andG.Li.“VillageLeadersandLand-Rights
FormationinChina.”AmericanEconomicReview,88(2),1998,433–8.
Rozelle,S.D.,andJ.F.M.Swinnen.“SuccessandFailure
ofReform:InsightsfromtheTransitionofAgricul-ture.”JournalofEconomicLiterature,42(2),2004,404–56.
Ruud,P.AnIntroductiontoClassicalEconometricTheory.
Oxford:OxfordUniversityPress,2000.
Sen,A.K.“AnAspectofIndianAgriculture.”TheEco-nomicWeekly,14,1962,243–46.
Staiger,D.,andJ.H.Stock.“InstrumentalVariablesRegres-sionwithWeakInstruments.”Econometrica,65(3),1997,557–86.
Taylor,J.D.,andI.Adelman.“AgriculturalHousehold
Models:Genesis,Evolution,andExtensions.”ReviewofEconomicsoftheHousehold,1(1),2003,33–59.
The9thStandingCommitteeoftheNationalPeople’s
CongressofthePeople’sRepublicofChina.“LawonRuralLandContract.”2002.AccessedDecember30,2002.http://www.agri.gov.cn/zcfg/t20020916_3250.htm.
Walder,A.G.,andL.Zhao.“PoliticalOfficeandHousehold
Wealth:RuralChinaintheDengEra.”TheChinaQuarterly,186,2006,357–76.
Wan,G.,andE.Cheng.“EffectsofLandFragmentation
andReturnstoScaleintheChineseFarmingSector.”AppliedEconomics,33(2),2001,183–94.
Wooldridge,J.M.EconometricAnalysisofCrossSection
andPanelData.Cambridge,MA:MITPress,2002.Xu,Y.“AgriculturalProductivityinChina.”ChinaEco-nomicReview,10(2),1999,108–121.
Yang,D.T.“EducationandOff-farmWork.”Economic
DevelopmentandCulturalChange,45(3),1997a,613–32.
.“EducationinProduction:MeasuringLaborQualityandManagement.”AmericanJournalofAgriculturalEconomics,79(3),1997b,764–72.
因篇幅问题不能全部显示，请点此查看更多更全内容