THREEESSAYSONFERTILITY,LABORMARKETPERFORMANCE,ANDPARENTAL MENTALHEALTH By HuiWang ADISSERTATION Submitted toMichiganStateUniversity inpartialoftherequirements forthedegreeof EconomicsŒDoctorofPhilosophy 2015 ABSTRACT THREEESSAYSONFERTILITY,LABORMARKETPERFORMANCE,ANDPARENTAL MENTALHEALTH By HuiWang Inthisdissertationresearch,theempiricalanalysesaredevelopedtoinvestigatethecausalre- lationshipsamongfertility,laborsupplyandparentalhealth.Asthesethreefactorsareclosely intertwined,ofcausalityareachievedthroughvariousmethods.Theandthird chaptersconstructexogenousvariationinfertilitythroughanaturalexperiment,theOne-Child PolicyinChina.Thentheconstructedexogenousvariationisusedasaninstrumentforfertility toidentifythecausalseffectsoffertilityonfemalelaborsupplyaswellasparentalmentalhealth. Thesecondpaperanalysestheimpactsofjobdisplacementsonfertility.Severaldifferentspeci- includingtimetrendmodel,edeffectpropensityscorematchingandregressionwith narrowerofjobdisplacementareusedtoverifytherobustnessofthecausaleffects. IChapterOne,Itrytoanswerthequestion,fiDoesfertilityplayadifferentroleinfemalelabor forceparticipationinChinathanintheU.S.?flThischapterexploitsplausiblyexogenousvari- ationsinfertilitycreatedbytheafveOne-ChildPolicyinChinatoestimatetheeffectof havingtwoormorechildrenonthemother'slaborforceparticipation.Usingalargedatasetfrom the1990PopulationCensus,IthatOLSunderestimatesthenegativeeffectsoffertility,and 2SLSestimatesimplythatconditionalonhavingonechild,additionalchildrendecreasesmother's femalelaborforceparticipationby8-15percentagepointsinruralChina.Recently,Chinarelaxed itsOne-ChildPolicytoTwo-ChildrenPolicy,ourhereprovidesaperspectiveforthepo- tentialeffectsofsuchpolicyrelaxationsonfemalelaborsupply. Historicalmacrodatashowanegativeassociationbetweenunemploymentandfertility.Individual levelpaneldataisneededtoexplainthecausalrelationshipbehindthenegativeassociation.Using microdatafromNationalLongitudinalSurveyofYouth1979,ChapterTwostudiestheeffectof jobdisplacementsonfertilityintheU.S.Aftercontrollingindividualtime-invariantheterogeneity, themainregressionresultsindicatethatdisplacementsofmenwillleadtoreducedfertilityinthe followingyears,whiletheeffectofdisplacementsforwomendependsonthewomen'seducation levels.Forwomenwithoutcollegeeducation,theirfertilitywillincreasefouryearsafterdisplace- ment.Forwomenwithcollegeeducation,however,noeffectonfertilityis Theempiricalarerobusttoseveraldifferentincludingtimetrendmodel, edeffectpropensityscorematchingandregressionwithnarrowerofjobdisplacement. Thereisanoldsaying,fiMorechildren,moreblessingsfl.Doeshavingmorechildrenreallypromote mentalhealthoftheparentsinChina?Toanswerthisquestion,ChapterThreeexploitsplausibly exogenousvariationsinfertilitycreatedbytheafveOne-ChildPolicyinChinatoestimate thelong-termeffectofhavingmorechildrenontheparent'smentalhealthforpeopleage45and aboveinruralareas.Usingdatafromthe2011ChinaHealthandRetirementLongitudinalStudy (CHARLS),resultsshowthat,aftercontrollingendogeneityinfertility,motherswithmorechil- drenarefoundtohaveahigherprobabilityofexperiencingdepressionsymptomsinruralChina, whiletheeffectsonfathersaregenerallynot ACKNOWLEDGEMENTS Thisdissertationhasgreatlyfromguidance,comments,andsupportfromToddElderand SongqingJin,myadvisors.Theirgenerosityoftimeandthoughtfulcommentswereprofoundly importantinthecompletionofthisdissertation.Theirkeenintuitionandunderstandingsofeco- nomicissueshavegreatlysharpenedmyfocus.Allofmyfuturedevelopmentwillbebuiltupon whattheyhavetaught,orattemptedtoteach,me. Iwouldalsoliketothanktheothermembersofmyguidancecommittee:LeahLakdawala,Thomas Reardon,JohnGiles,andJeffreyWooldridgefortheirinvaluableguidanceandinsightsatvarious stagesoftheprocess. ItisimperativethatIthankmyhusband,ShangSun,whoconsistentlystimulatesandaccompanies meontheroadofacademicadventure.Itwouldnotbeasmuchfunandenjoymentinmyve yearsofPhDlifewithouthim.Ialsomustthankmyparentsfortheirencouragementandsupport duringmywholelife. Inaddition,manypeopledeservemythanksforhavingofferedhelponvariouspartsofthisdisser- tation.Inparticular,Iamgratefulforextensivecommentsandsuggestionsfromfacultymembers andfellowgraduatestudentsatMichiganStateUniversity. Finally,IhavetothankforthesupportfromEconomicsdepartmentandAgricultural, FoodandResourceEconomicsdepartment.Withouttheassistance,thisresearchwould havebeenimpossible. iv TABLEOFCONTENTS LISTOFTABLES ....................................... vii LISTOFFIGURES ....................................... xi CHAPTER1FERTILITYANDFEMALELABORFORCEPARTICIPATION:EVI- DENCEFROMTHEONE-CHILDPOLICYINCHINA ........... 1 1.1Introduction......................................1 1.2TheOne-ChildPolicyinChina............................5 1.3EstimationStrategy..................................7 1.3.1DIDusingethnicity..............................8 1.3.2DIDusinggenderof........................10 1.3.3OtherresearchusingtheOCPtoconstructinstrumentsforfertility.....11 1.4DataandDescriptiveStatistics............................13 1.5RegressionResults..................................19 1.5.1OLSestimatesoftheeffectofadditionalchildrenonfemaleLFP......19 1.5.2FirstStage:theeffectsoftheOneChildPolicyonfertility.........20 1.5.3Reducedform:theeffectsoftheOne-ChildPolicyonfemaleFLP.....21 1.5.42SLSestimatesoftheeffectofadditionalchildrenonfemaleLFP.....22 1.5.5HeterogeneousEffects............................23 1.6ValidityofInstruments................................24 1.6.1DIDusingethnicity..............................24 1.6.2DIDusinggenderofthe......................26 1.6.3Placebotestson2-Childrenprovinces....................27 1.7RobustnessCheck...................................29 1.7.1Usingtripledifferencesasinstrument....................29 1.7.2Usetwinningasinstrument..........................30 1.8Conclusion......................................30 CHAPTER2HOWDOJOBDISPLACEMENTSAFFECTFERTILITYINTHEU.S. .. 32 2.1Introduction......................................32 2.2WhyWouldDisplacementAffectFertility......................36 2.2.1RelatedLiterature...............................36 2.2.2ConceptualModel..............................37 2.3DataandDescriptiveStatistics............................41 2.4RegressionResults..................................45 2.4.1FixedEffectsModel.............................45 2.4.2TimeTrendModel..............................48 2.4.3RobustnessCheck...............................49 2.5PlantClosure.....................................50 2.6PropensityScoreEstimates..............................51 v 2.7Conclusion......................................53 CHAPTER3THEMORETHEMERRIER?THEEFFECTOFFAMILYSIZEON PARENT'SMENTALHEALTHINRURALCHINA ............ 55 3.1Introduction......................................55 3.2TheOne-ChildPolicyinChinaandEstimationStrategy...............59 3.3DataandSummaryStatistics.............................63 3.4RegressionResults..................................65 3.5PhysicalHealthandLivingArrangements......................69 3.5.1Self-ReportedHealthandChronicDiseases.................69 3.5.2LivingArrangements.............................71 3.6RobustnessCheck...................................72 3.7Conclusion......................................73 APPENDICES ......................................... 74 AppendixATablesforChapter1 ............................ 75 AppendixBFiguresforChapter1 ............................ 93 AppendixCTablesforChapter2 ............................ 99 AppendixDFiguresforChapter2 ............................ 110 AppendixETablesforChapter3 ............................ 120 AppendixFFiguresforChapter3 ............................ 133 AppendixGAppendicesforChapter1 .......................... 136 AppendixHAppendicesforChapter2 .......................... 148 AppendixIAppendicesforChapter3 .......................... 153 BIBLIOGRAPHY ........................................ 165 vi LISTOFTABLES TableA.1DescriptiveStatistics,Womenaged16-45withatleastonechild.........75 TableA.2SummaryStatisticsforEachSample........................76 TableA.3HanVs.non-HanandFirst-BornGirlVs.First-BornBoy............77 TableA.4DIDEstimatesRegardingEthnicity........................78 TableA.5DIDEstimatesRegardingGenderofFirstBirth..................79 TableA.6OLSand2SLSEstimatesoftheEffectofAdditionalChildrenonFemaleLFP..80 TableA.7CoefofInteractionTermsforthe1stStageRegressionsRegardingEthnicity81 TableA.8CoefofInteractionTermsforthe1stStageRegressionsRegarding Genderof1stBirth.................................82 TableA.9CoefofInteractionTermsfortheReducedFormRegressionsRegard- ingEthnicity....................................84 TableA.10CoefofInteractionTermsfortheReducedFormRegressionsRegard- ingGenderof1stBirth...............................85 TableA.11HeterogeneousEffectofAdditionalChildrenonFemaleLFP...........87 TableA.12CoefofInteractionTermsfortheRegressionsofEducationandGender ofFirstBirth....................................88 TableA.13CoefofInteractionTermsforthe1stStageRegressionsRegardingEth- nicity(2-ChildrenProvinces)............................89 TableA.14CoefofInteractionTermsforthe1stStageRegressionsRegarding GenderoftheFirst-Birth(2-ChildrenProvinces).................90 TableA.15RobustnessCheckoftheEffectofAdditionalChildrenonFemaleLFP.....92 TableC.1People'sCharacteristicsbyTheirDisplacementStatus..............99 TableC.2ImpactofDisplacementontheProbabilityofHavinganAdditionalChild (FixedEffect)....................................100 vii TableC.3HeterogeneousImpactofDisplacementontheProbabilityofHavinganAd- ditionalChild(FixedEffect)............................101 TableC.4InteractionTermsofCollegeEducationandDisplacementStatus.........102 TableC.5ImpactofDisplacementontheProbabilityofHavinganAdditionalChild (TimeTrendModel)................................103 TableC.6ImpactofDisplacementontheProbabilityofHavinganAdditionalChild (CorrelatedRandomEffectProbitModel).....................104 TableC.7ImpactofDisplacement_ExcludingPeopleSufferingFirstDisplacementAf- ter1994.......................................105 TableC.8ImpactofDisplacementforObseravationsduring1984-1992...........106 TableC.9ComparisonsbetweenNon-displacedPeopleandPeopleLostaJobDueto FirmClosure....................................107 TableC.10ImpactofJobLossDuetoFirmClosureontheProbabilityofHavingan AdditionalChild(FixedEffect)..........................108 TableC.11HeterogeneousImpactofDisplacementontheProbabilityofHavinganAd- ditionalChild(FixedEffectPropensityScoreMatching).............109 TableE.1SummaryStatisticsforKeyVariables.......................120 TableE.2DIDEstimatesRegardingGenderofFirstBirth..................121 TableE.3FirstStageResultsforTwoorMoreChildren...................122 TableE.4FirstStageResultsforNumberofChildren....................123 TableE.5OLSResultsforEffectsofFertilityonParent'sMentalHealth..........124 TableE.62SLSResultsforEffectsofFertilityonParent'sMentalHealth..........125 TableE.72SLSEstimatesofEffectsofTwoorMoreChildrenonParent'sMentalHealth byParents'Education...............................126 TableE.82SLSEstimatesofEffectsofNumberofChildrenonParent'sMentalHealth byParents'Education...............................127 TableE.92SLSResultsforEffectsofFertilityonParent'sSelf-ReportedHealth......128 TableE.102SLSEstimatesofEffectsofFertilityonDiagnosisofChronicDiseases.....129 viii TableE.112SLSResultsforEffectsofFertilityonParent'sMentalHealth(Controlling Self-ReportedHealth)...............................130 TableE.122SLSResultsforEffectsofFertilityonCo-residencewithChildren.......131 TableE.132SLSResultsforEffectsofFertilityonParent'sMentalHealth(Controlling LivingArrangements)...............................132 TableG.1CoefofInteractionTermsforFirstStageandReducedFormwhenFo- cusingonMothers 30...............................136 TableG.2OLSand2SLSEstimatesoftheEffectofAdditionalChildrenonFemaleLFP whenFocusingonMothers 30..........................138 TableG.3FertilityonFemaleLFP: T i non-HanasIV....................139 TableG.4FertilityonFemaleLFP: T i First-BornGirlasIV.................140 TableG.5FertilityonFemaleLFP:DIDBasedonEthnicityasIV..............141 TableG.6CoefofTrippleInteractionTerms.....................142 TableH.1ImpactofDisplacementontheProbabilityofHavinganAdditionalChild (FirstDifference)..................................148 TableH.2ImpactofDisplacementontheProbabilityofHavinganAdditionalChild (Includingmarriagestatus).............................149 TableI.1ListofCES-D-10itemstomeasurementalhealth.................153 TableI.2CoefofInteractionTermsforthe1stStageRegressiononTwoorMore Children.......................................154 TableI.3CoefofInteractionTermsforthe1stStageRegressiononNumberof Children.......................................156 TableI.4CoefofInteractionTermsforthe1stStageRegressiononTwoorMore Children.......................................158 TableI.5CoefofInteractionTermsforthe1stStageRegressiononNumberof Children.......................................160 TableI.62SLSResultsforEffectsofHavingTwoChildrenonParent'sMentalHealth..162 TableI.7OLSEstimatesofEffectsofTwoorMoreChildrenonParent'sMentalHealth byParents'Education...............................163 ix TableI.8OLSEstimatesofEffectsofNumberofChildrenonParent'sMentalHealth byParents'Education...............................164 x LISTOFFIGURES FigureB.1FemaleLaborForceParticipationandTotalFertilityRateintheU.S.......93 FigureB.2FemaleLaborForceParticipationandTotalFertilityRateinChina.......94 FigureB.3TotalFertilityRateinChina............................95 FigureB.4CoefofInteractionsofAgeCohortsandEthnicity/First-BornGirl (HeterogenousEffectsforWomenwithDifferentEducationLevels).......96 FigureB.5GirlPercentageforFirst-BornsinDifferentYears(Ages).............97 FigureB.6CoefofInteractionsofAgeCohortsandEthnicity/First-BornGirl....98 FigureD.1TheU.S.FertilityRateHasFallenDuringRecessions..............110 FigureD.2UnemploymentandBirthRateintheU.S.....................111 FigureD.3EffectsofWageDecreaseforWomenwithVeryHighWage..........112 FigureD.4DisplacementRatefromNLSY79Vs.OfUnemploymentintheU.S. (1984-1994)....................................113 FigureD.5DisplacementRateforDifferentAgeCohorts.(1984-1994)...........114 FigureD.6DisplacementandBirthRateforMen.......................115 FigureD.7DisplacementandBirthRateforWomen.....................116 FigureD.8DisplacementandBirthRateforWomenwithCollegeEducation........117 FigureD.9DisplacementandBirthRateforWomenwithoutCollegeEducation......118 FigureD.10PropensityHistogrambyDisplacementStatusin1984_Men...........119 FigureF.1NumberofChildrenandParent'sCES-DScore.................133 FigureF.2NumberofChildrenandParent'sVignetteQuestionScore............134 FigureF.3CES-DScoreandVignetteQuestionScore....................135 FigureG.1Correlationbetweennumberofchildrenandlaborforceparticipationrate...144 FigureG.2Averagenumberofchildrenforwomenatdifferentage..............145 xi FigureG.3CumulativeDistributionofAgeWhenGivingSecondBirth...........146 FigureG.4SexRatioatBirthandAbortionRateinChina(1978-1991)...........147 FigureH.1DisplacementandBirthRateforMenwithCollegeEducation..........150 FigureH.2DisplacementandBirthRateforMenwithCollegeEducation..........151 FigureH.3DisplacementandBirthRateforMenwithoutCollegeEducation........152 xii CHAPTER1 FERTILITYANDFEMALELABORFORCEPARTICIPATION:EVIDENCEFROM THEONE-CHILDPOLICYINCHINA 1.1Introduction Inspiredinpartbyacontemporaneousincreaseinfemalelaborforceparticipationandadecrease infertilityintheU.S.(FigureB.1),manyeconomistshaveinvestigatedtheeffectsofnumber ofchildrenonmother'slaborsupply(Klerman,1999;AngristandEvans,1998;Bailey,2006). AngristandEvans(1998)andmanyothersnegativeeffectsoffertilityonfemalelaborforce participationintheU.S. 1 Ontheothersideoftheworld,ChinahasimplementedtheOne-Child Policysince1979aimingatcurbingitsfastpopulationgrowth;however,surprisingly,withfewer children,itappearsthatwomeninChinaparticipatelessinthelaborforce(FigureB.2).Inthis study,IexaminetheeffectsoffertilityonfemalelaborforceparticipationinruralChina. 2 Arethe effectsoffertilityinChinathesameasfoundintheU.S.,oraretheyopposite,assuggestedbythe informationfoundinFigureB.1andFigureB.2?Thispapercanhelpusanswerthisquestion. Empiricallyanalyzingtheeffectoffertilityonfemalelaborforceparticipationischallenging duetotheendogeneityoffertilitydecisionswithrespecttolaborsupply.Willis(1974)showsthat femalelaborforceparticipationandfertilityarealwaysjointlydetermined.Forexample,ifomitted variablessuchaswomen'spreferencesforworkarenegativelycorrelatedwiththeirpreferences forhavingmorechildren,orifthetimeandeffortspentonworkdiscouragesfertility(generating 1 Anegativecorrelationbetweenfertilityandfemalelaborforceparticipationhasbeenfound bothovertimeandcrosscountries(AgüeroandMarks,2011). 2 IexploittwosourcesofvariationinfertilitythatariseduetotheOne-ChildPolicy:gender atbirthandethnicity.Sincegenderatrstbirthisrelevantfortheimplementationofthe OCPonlyinruralareas(Qian,2009),IfocusonruralChinaforthisstudy.Moreoversincethe penaltiesimposedonthosewhoexceedtheirquotainurbanareasmayinvolvejobloss,theOCP maydirectlyaffectlaborforceparticipationintheseareasandthusIexcludeurbanareasfromthis analysis. 1 reversecausality),thentheestimatesoftheeffectofnumberofchildrenonfemalelaborsupply wouldbebiaseddownward.Thebiascouldalsogointheoppositedirection;forexample,some havefoundthatfertilityindevelopingcountriesisdeterminedthroughacollectivebargaining processatthehouseholdlevelandhouseholdmemberswithmorebargainingpowerhavemore onthetotalnumberofchildren(Rasul,2008).Inaddition,menaregenerallyfoundto prefermorechildrenthanwomenindevelopingcountries(MasonandTaj,1987).Ifwomenwith lessbargainingpowerareforcedtoworkaswellashavemorechildren,thennaiveOLSregressions willunderestimatethenegativeimpactofchildrenonmother'slaborforceparticipation. Toaddressthisendogeneityproblem,economistshaveexploitedvarioussourcesofexogenous variationinfamilysizesuchastwinning(RosenzweigandWolpin,1980;Jacobsenetal.,1999),sex compositionofthetwobirths(AngristandEvans,1998), 3 state-levelaccesstocontraception (Bailey,2006),andstate-levelabortionlaws(Klerman,1999;Levineetal.,1999;Angristand Evans,2000).AlloftheseU.S.-basedstudiesfoundsolidevidenceofanegativeeffectoffertility onfemalelaborsupply. Incontrast,studiesbasedondevelopingcountriesshowmixedresultsontheeffectoffertility onmaternallaborsupply.UsingdatafromarandomizedsocialexperimentinMatlab,Bangladesh, Schultz(2009)foundanegativeeffectforfamilyplanningprograms(whichleadtolowerfertility) onthelikelihoodoffiworkforpayflforwomen.Usingson-preference 4 asaninstrument,Eben- stein(2009)negativeeffectsoffertilityonmaternallaborforceparticipationinTaiwan. BasedondatafromDemographicandHealthSurveyscovering59developingcountries(excluding China),usingson-preference,thesexcompositionofandsecondbornchildren,andtwinning asinstruments,PorterandKing(2010)reportedthatwhileinmanydevelopingcountrieswomen arelesslikelytoworkwhentheyhavemorechildren,insomecountries,motherswithmorechil- drenaremorelikelytoworkduetoreasonssuchasthecostsofraisingmorechildren. 3 Theyfoundparentsoftwochildrenofthesamesexhavehigherprobabilityofhavingathird birththanparentsofoneboyandonegirl. 4 IneastAsiancountries,familiesarefoundtohavestrongpreferenceformalechildren.In particular,familieswithtwogirlshavehigherprobabilitytohavethirdchild. 2 CombiningdatafromDemographicandHealthSurveyswithabortionlegislationdocumentsin eachcountry,Bloometal.(2009)thesemixedresultsontheeffectsoffertility.Ina cross-countrystudy,AgüeroandMarks(2011)analyzed26low-andmiddle-incomecountries, andusinginfertilityasaninstrumentforfamilysize.Theyfoundthatnumberofchildrenhasno effectonlaborforceparticipationandworkintensity. Iamawareofonlyoneexistingstudyontheimpactoffertilityonfemalelaborsupplyinthe contextofChina.Basedoninformationfrom1,315womenfromtheChinaHealthandNutrition Surveyconductedin1993,Lee(2002)usedthesexofthechildasaninstrument 5 andfound noanteffectsoffertilityonruralfemalelaborsupplyinChina. 6 Themainconcernwhen usingson-preferenceasaninstrumentfortotalfertilityisthatthegenderofthechild maynotsatisfytheexclusionrestrictioncondition,sincepreviousworkhasshownthatchildren's genderaffectsmothers'laborsupplydirectly(CrucesandGaliani,2007). Inordertocurbrapidpopulationgrowth,theChinesegovernmenthasimplementedtheOne- ChildPolicy(OCP)since1979,possiblythelargestsocialexperimentinhumanhistory.Under thispolicy,amarriedcouplecanhaveonlyonechildinmostareasofthecountry.However, thereareseveralimportantexceptions.TheisthattheOCPwasnotappliedtonon-Han Chineseuntil1988.Thesecondexceptionarosewhen,insomeareas,itwasobservedthatthe OCPwasassociatedwithfemaleinfanticide,forcedabortionandforcedsterilization.Toprevent theseextremecases,19provincesadoptedthefi1-boy-2-girlflrulein1984,whichstatedthatrural couplesintheseprovinceswereallowedtohaveasecondchildifthechildwasagirl(Qian, 2009). Inthispaper,IexploittwosourcesofexogenousvariationinfertilitygeneratedbytheOne- ChildPolicyinChina.More,Iusethedifferencesinthepolicy'simplementation 5 Withstrongson-preference,theparentsinChinaarefoundtohavemorechildrenwhenthey havenosonsyet(Ebenstein,2009). 6 Lee(2002)alsoattemptedtousetheOne-ChildPolicy,inparticular,filocalfamilyplanning rulesflasinstruments,butheadmittedthisinstrumentwaspotentiallyendogenous.Idiscussthisin moredetailinSection3.3. 3 betweentheHanandnon-HanChinese,aswellasbetweenruralcoupleswithgirlsand thosewithboysinthe19provincesthatadoptedthefi1-boy-2-girlflruletoinstrument forthelikelihoodawomanhastwoormorechildren.Thusthecomesfromvariation infertilitygeneratedbytheOCPacrosscohortsofwomen(thoseaffectedandnotaffectedbythe policyduringtheirfertileyears),ethnicities(Hanandnon-Han),andfamiliesbasedonthegender oftheirchildren.AssumingthatwithoutthevariationintheOCPthedifferencesbetween womenintheunaffected(old)andaffected(young)cohortswouldbethesamefortheHanand non-Han,thedifferences-in-differencesestimatesbasedonethnicitywillcapturetheexogenous variationoffertilityduetothevariationintheOCP.Similarly,undertheassumptionthatthediffer- encesbetweenwomenindifferentagecohortswouldbethesameforwomenwitheither boysorgirlswithoutthevariationintheOCP,DIDestimatesbasedongenderof birthcancapturetheexogenousvariationsoffertilitygeneratedbytheOCP.Usingalargerandom samplefromthe1990ChinaPopulationCensus,theresultsshowthatOLSregressionsunderesti- matethediscouragingeffectsofchildrenonfemalelaborforceparticipation.Afteraccountingfor theendogeneityoffertility,thelikelihoodoflaborforceparticipationformotherswithtwoormore childrenis8to15percentagepointslowerthanformotherswithonlyonechildinruralChinain 1990. Thispapermakesseveralcontributions.First,itaimstotheknowledgegaponeffectsof fertilityonfemalelaborsupplyinChina,themostpopulouscountryintheworldwithacontro- versialpopulationcontrol.Alongwiththesubstantialdropinfertilityexperiencedoverthelast30 years,Chinawitnessedadeclineinfemalelaborforceparticipation,raisingdoubtsastowhether theobservednegativerelationshipbetweenfertilityandfemalelaborforceparticipationobserved indevelopedcountriesextendstothedevelopingcountrysetting.Byidentifyingthecausaleffects offertilityonfemalelaborforceparticipationinChina,thisresearchcanenrichourunderstanding oftheeffectsoffertilityonfemalelaborsupply. Second,themajorityofpastresearchfocusedontheeffectsofanadditionalchildfortheselect sampleofwomenwhohavehadtwochildren.Weknowlittleabouttheeffectofincreasingthe 4 numberofchildrenfromonetotwo,whichisimportantsincesomeresearchshowsthattheeffect ofincreasingthenumberofchildrenisnonlinear(Black,Devereux,andSalvanes,2004)(Figure G.1thatinourdataset,thecorrelationbetweennumberofchildrenandfemalelaborforce participationisnonlinear). Third,thereisanimportantlimitationofusingbothsamesexandtwiningasinstrumentsto identifytheeffectsoffertilityonfemalelaborforceparticipation.Bothstrategiesonlyidentifythe LocalAverageTreatmentEffect(LATE)forwomenwhoseunderlyingdesiretohavemorechildren islow.UsingsamesexastheinstrumenttofamilysizeestimatestheLATEforwomenwhose underlyingdesiretohavemorechildrenislowbutwhoareinducedtohaveanadditionalchildto balancethesexcompositionofthefamily.Similarly,usingtwinningastheinstrument,estimates theLATEforwomenwhoprefertohavesmallfamiliesbuthappentohaveanadditionalchilddue tofiluckfl(AgüeroandMarks,2011).Usinganexogenousvariationintheimplementationofthe One-ChildPolicyallowsustolookattheLATEforwomenwhohaveahighpersonaldemandfor childrenbutarerestrictedbythepolicy.Consideringtherelativelyhigherdemandforchildrenin developingcountries,thecompliersinourstrategymaybeamorerelevantpopulation. Theremainderofthepaperislaidoutasfollows.Section2providesbackgroundinformation onChina'sOne-ChildPolicy.Section3introducestheestimationstrategyandtheconstruction oftheinstrumentsforfamilysize.Section4describesthedataandtwosetsofbasicdifferences- in-differencesestimates.Section5discussesthemainregressionresultsandexploresthehetero- geneouseffectsoffertilityforwomenwithdifferentlevelsofeducation.Section6vthe validityoftheinstruments.Section7providesadditionalrobustnesschecksforthemainresults. Section8concludes. 1.2TheOne-ChildPolicyinChina Between1962-1970,thepopulationgrowthrateinChinareached27 : 5 h peryear,andthetotal populationreached816millionin1970(Yang,2004).Toalleviatethesocial,economic,and 5 environmentalproblemscausedbyincreasingpopulationpressure,theChinesegovernmentbegan tocurbpopulationgrowthasearlyas1972usingapolicyknownasfiLater,Longer,andFewerfl. Thispolicyencouragedpeopletogetmarriedandhavechildrenatalaterage(therecommended ageofmarriagewas25yearsoraboveformenand23yearsoraboveforwomen),suggested longerbirthspacing(atleastthreeorfouryears),andalsorecommendedthatcoupleshaveat most2children.Implementationofthispolicyreliedprimarilyonpropagandaandsocialpressure (McElroyandYang,2000).Asaresult,thetotalfertilityrateinChinastartedtodecreasein1972 (FigureB.3). In1979,Chinainitiatedthewell-knownfiOne-ChildPolicy"(OCP),theimplementationof whichwasmoreforceful.Underthispolicy,marriedcoupleswereallowedonlyonechildinmost areas,exceptforthoselivinginruralareasinveprovinces(Hainan,Yunan,Qinghai,Ningxia, Xinjiang),whowereallowedtohavetwochildren(Peng,1996). 7 Inpractice,implementationof thispolicyinsomeregionsbeganasearlyasin1978,andenforcementwastightenednationally in1980.InareassubjectedtotheOCP,asecondbirthwaspermittedonlyiftheonechildwould causethehouseholdfirealdifsuchaspoorhealthconditionsofthechild.Couples whohadanabove-quotabirthwithoutpermissionweresubjecttopenaltiessuchassevere jobloss,andlossofaccesstopublicgoods. 8 Localcadresweregiveneconomicandpromotion incentivestoimplementthepolicy. AsaresultoftheOCP,intheearly1980sfipartsofthecountryweresweptbycampaignsof forcedabortionandsterilizationandreportsoffemaleinfanticidebecamewidespreadfl(Green- halgh,1986).Topreventfemaleinfanticide,forcedabortionandforcedsterilization,andtobetter addressreconditions,theCentralPartyCommitteeissuedfiDocument7flinApril 1984,allowingregionalvariationinfamilyplanningpolices.Themainchangeinthepolicyfol- 7 TherearenorestrictionsonnumberofchildrenforruralcouplesinTibet. 8 InurbanChina,wheremostworkinstate-ownedenterprises,andmakeuseofpublicgoods, joblossandlostaccesstopublicgoodsarethemostcommonlyusedpunishments.Inruralareas, wheremostpeopleworkontheirownlandandprivateenterprises,theone-timeistheprimary penaltyusedbylocalgovernmentofThereislocalvariationin(WeiandZhang,2011). 6 lowingfiDocument7flwasthefi1-boy-2-girlflrulein19provinces,whichallowedruralcouples inthese19provincestohaveasecondchildifthewasagirl(Qian,2009).According toWhite(1991),thesekindsofpermissionsbegantobeissuedasearlyas1982,suggestingthat relaxationexistedevenbefore1984. 9 AnimportantfeatureoftheOCPwasthatitwasonlyappliedtoHanChinesebefore1988.This afverulehadbeenmadeattheconceptionofthebirth-controlpolicyinChinaandrecorded inalldocumentsrelatedtopopulationpolicy(Peng,1996).In1988,whenthepopulationofthe Zhuangethnicgroupreached10million,theybecamesubjecttotheOCPaswell.EthnicManchus weresimilarlyaddedin1990(Li,ZhangandZhu2005).Withimportantvariationacrossgroups, theOne-ChildPolicygivesusanopportunitytoinvestigatetheeffectsofexogenouschangein familysize. 1.3EstimationStrategy Followingtheliterature,themainregressionmodelweareinterestedincanbewrittenas: LFP ict = b kids 2 ict + X 0 ict d + a 1 + g t + y c + e ict (1.1) where LFP ict isthelaborforceparticipationindicatorforwoman i incounty c ,agecohort t .Follow Maurer-Fazio,HughesandZhang(2005), LFP ict = 1ifthewomanhadajobonthedayofthe censusorifshewasunemployedbutwascodedasfiwaitingforworkfl.Itwouldbeidealifwecan measurewomensupplyinbothextensivemarginandintensivemargin.However,withthelimited informationprovidedin1990ChinaPopulationCensus,Ifocusonlyontheextensivemarginin thisstudy. kids 2 ict isadummyvariableequalto1ifthewomanhastwoormorechildren; 10 9 AstheOCPgottightenednationallyin1980,andthefi1-boy-2-girlflappearedasearlyas 1982,Idonotexpectthe1984amendmenttogeneratetwoseparatecohortswithlargedifferences. MotherswithgirlswhowerenotallowedtohaveasecondbirthundertheoriginalOCP cancontinuetohavethesecondchildaftertheimplementationoffi1-boy-2-girlflrule. 10 Allwomeninoursamplehaveatleastonechild;therefore, kids 2 ict = 0meansthewomanhas onlyonechild. 7 X ict isavectorofwoman i 'scharacteristics,includingherage,ageatbirth,ethnicity,gender ofthechild,andeducationlevelsforbothherandherhusband; 11 g t istheagecohorted effect,and y c denotesthecountyedeffect.Weusethedummyvariable kids 2 ict tomeasure fertilityratherthannumberofchildren,sinceourDIDestimatesonlyapplytothediscretechange fromonechildtotwoormorechildren.AngristandImbens(1995)indicatesthattheresulting estimatedeffectswillbebiggerthanthetrueaverageper-uniteffectwhenthetreatmentvariableis incorrectlyparameterizedasbinary,whilethesignoftheAverageCausalEffectisstillconsistently estimated.Astheeffectsofchildrenonfemalelaborsupplyarelikelytobenon-linear(FigureG.1), wetakecautionininterpretingourestimatesofthecoefon kids 2 ict . Women'slaborforceparticipationmayaffectfertilitydecisionsandtheremightbeunobserved factors(e.g.health)thataffect LFP ict and kids 2 ict simultaneously.Forthesereasonsandothers,we believethatthecondition cov ( kids 2 ict ; e ict )= 0doesnothold.Asaresult,theOLSestimatorof b isnotconsistent.Toaddressthisendogeneityproblem,Iusetwosetsofdifferences-in-differences (DID)strategiestoconstructexogenousvariationinfertility,andthenusethisvariationtoinstru- ment kids 2 ict inEquation(1).TheDIDstrategyexploitthedifferencesintheprobability ofhavingtwoormorechildrenbetweenHanandnon-HanChinese,forwomenaffectedandun- affectedbytheOCP;whilethesecondstrategycomparesthedifferencesbetweenwomenwith girlsandwomenwithboys,foraffectedandunaffectedcohorts. 1.3.1DIDusingethnicity AsdescribedinSection2,before1988onlyHanChineseweresubjecttothestrictOne-ChildPol- icy,whilenon-Hancoupleswereallowedtohavetwochildren.Oneattractiveestimationstrategy, therefore,istouseethnicitytocapturetheexogenousvariationsinnumberofchildren.Unfor- tunately,knowledgeaboutChinaaswellassummarystatisticsshowninSection4suggestthat therearemanysystematicdifferencesbetweenHanandnon-HanChinese.Therefore,onewould 11 TableA.1reportstheofthekeyvariables(dependentvariables,covariates,and instruments). 8 worrythatethnicitymightdirectlyaffectwomen'slaborforceparticipationdecisions,andthusthe exclusionrestrictionwillnotholdifweuseethnicityaloneasaninstrument. UsingDIDcanhelpremovetheorderdifferencesbetweenHanandnon-HanChinese.The DIDmethodcanbesimplyexpressedas ( nonHan ; After Han ; After ) ( nonHan ; Before Han ; Before ) .UndertheassumptionthatwithouttheOCP,thedifferenceinfemalelaborforce participationbetweenaffectedandunaffectedcohortswouldbethesameforHanandnon-Han individuals,DIDestimateswillcapturetheexogenousvariationsof kids 2 ict duetothevariationin theOCP,whichaffectsdifferentethnicgroupsdifferently. Here,weuse After torepresentfemalesrestrictedbytheOCP,thatis,theyoungcohortsin oursample. Before denotesthecohortnotconstrainedbytheOCP,thatis,theoldcohortswho probablyhadtwoormorechildrenalreadybefore1979/1980.ThoughtheOne-ChildPolicywas announcedin1979,thereisnosimpledistinctionbetweenexposed/treatedandnon-exposed/non- treatedindividuals,aspeoplechoosetheirfertilitytimingdifferently.FigureG.3adepictsthe cumulativedistributionofageatsecondbirthfortherestrictedprovincessample.Itshowsthat over90%ofwomenwithtwoormorechildrengavebirthtotheirsecondchildbeforetheage of30.Considering1980astheyearthattheOCPbecamenationallyimplemented,Iassume thatwomenolderthan30in1980wererelativelyless-constrainedbytheOCP.Thecutoffage of30in1980meansthatthecutoffageis40in1990.Thus,womenolderthan40inthe1990 censuswillberegardedas Before cohorts,whilewomenage40oryoungerwillbethe After cohorts.Inregressions,ratherthanarbitrarilydividingthewomenintothesetwocohorts,Iuseage cohortdummiestoallowforthemostxibilityintheeffectoftheOCPonfertility.UsingDID tocapturetheexogenousvariationinfertilitycausedbyexogenousvariationinpolicychangeis fundamentallyequivalenttousinginteractionterms.Hence,thestageregressionof kids 2 ict ontheinteractiontermswillhavethefollowingform: kids 2 ict = 44 å l = 16 ( nonHan ict d l ) r l + X 0 ict k + a 2 + d t + q c + u ict (1.2) 9 where d l ; l = 16 to 44areagedummiesfrom16to44yearsofage. 12 1.3.2DIDusinggenderof In1984,topreventfemaleinfanticide,forcedabortionandforcedsterilization,anamendmentto theoriginalOCPwasimplementedin19provincesin1984.Undertheamendment,aruralcouple isallowedtohaveasecondchildiftheirchildisagirl,alsoknownasthefi1-boy-2-girlflrule. ThisamendmentprovidesuswithanotherDIDstrategytoconstructtheexogenousvariationin fertility. CrucesandGaliani(2007)thatthegenderofchildrendirectlyaffectswomen'slaborsupply inMexicoandArgentina.ThismightbetrueinAsia,especiallyinChina,aswell.InruralChina, wherethepreferenceforsonsisstrong,boysarevaluedmorehighlythangirls.Itmightbeexpected thatthemotherswithsonswouldspendmoretimeonchildcarethanmotherswithdaughters.Inthis case,thegenderofchildrenwilldirectlyaffectamother'slaborsupply.However,wecanremove thedirecteffectofgenderonfemalelaborsupplyusingDID.TheDIDmethodcan beexpressedas ( First - BornGirl ; After First - BornBoy ; After ) ( First - BornGirl ; Before First - BornBoy ; Before ) .Asabove, After representsthetreatedoryoungcohort,while Before denotesthecontroloroldcohort.Inallregressions,Iincludeagecohortdummies.Thekey assumptioninthisDIDstrategyisthatwithoutthevariationintheOCP,thedifferenceinfemale laborforceparticipationdecisionbetweentreatmentandcontrolcohortswouldbethesamefor womenwhosechildisasonandthosewhosechildisadaughter.Mystageregression usinginteractiontermsofagecohortsandgenderofbirthisthus: kids 2 ict = 44 å l = 22 ( First - BornGirl ict d l ) f l + X 0 ict m + a 3 + d t + p c + v ict (1.3) where d l ; l = 16 to 44areagedummiesfrom16to44yearsofage. 12 ReasonsforusingthesecohortsarediscussedinSection4. 10 1.3.3OtherresearchusingtheOCPtoconstructinstrumentsforfertility Althoughthereexistsonlyonestudyexploringtheeffectsoffertilityonfemalelaborforcepartic- ipationinChina,thestrategyofusingthevariationinimplementationoftheOCPtoinstrument forfertilityhasbeenwidelyadoptedinseveralstudiesexploringtheeffectsoffertilityonother outcomesinChina.ShortandZhai(1998)showthatinpractice,theimplementationoftheOCP variedgeographically.Someoftheresearchtriestoexploitthesespatialvariationintheimple- mentationoftheOne-ChildPolicy.Forexample,inherstudyoftheeffectsofnumberofchildren onschoolenrollmentoftheQian(2009)usedtheimplementationofthefi1-boy-2-girlfl ruleatcountylevelinteractedwithyearandgenderofbirthasinstrumentsfornumberof children.However,whenLee(2002)triestouselocalfamilyplanningrulesasinstrumentsfortwo ormorechildren,hethattheimplementationofthefi1-boy-2-girlflrulesatthecountylevel iscorrelatedwithcommunitylocationandinfrastructure.Communitieslocatedfar awayfromcitiesand/orwithpoorinfrastructurearemorelikelytoimplementthisrule.Therefore, theselocalrulesmaybeendogenoustolocallabormarket,sowedonotusethemasaninstrument inourstudy. 13 Inotherrelatedstudies,researchersusetheyearofthebirthtoinstrumentforfertility.To estimatetheeffectofnumberofchildrenonelderlyparents'health,IslamandSmyth(2010)use theinteractiontermsofarural/urbanindicatorandthreeperioddummiescorrespondingtotheyear 13 Lee(2002)mainlyexploitedthreelocalrules,i.e.,underthreedifferentscenarios,whethera communityallowsacoupletohaveasecondchildornot.Thethreescenariosare:1)thetborn childisagirl;2)bothhusbandandwifearefromaone-childfamily;3)oneparenthasfidangerousfl job.Onlyrulefirelevanceflasaninstrument.However,whenusingboththerule andgenderofchildasinstruments,Leedifferentresultsfromusinggenderofchild astheonlyinstrument.Asaresult,overtestisrejected.Assuminggenderof birthisexogenous,Leeclaimedthattheimplementationoftherule"atthecountylevelisnot avalidIV.Healsotriedtoputrule"directlyinOLSregressionofnumberofchildren,and generatedanegativecoefInaddition,hefoundthatcounty-levelrule" implementationiscantlycorrelatedwithcommunitylocationandinfrastructure.Communi- tieslocatedfarawayfromcitiesand/orwithpoorinfrastructuresaremorelikelytoimplementthis rule.Putallthesetogether,heclaimedlocalfi1-boy-2-girlflruleatthecountylevelisendogenous tothelocallabormarket. 11 ofthebirth.Wheninvestigatingtheeffectsoffertilityonparent'ssavingbehavior,Banerjee, MengandQian(2010)useadummyforwhetherthebirthwasbefore1972tocapturethe effectofthefiLater,Longer,Fewerflpolicyonthenumberofchildren.Thepotentialproblemwith thisIVstrategyisthattheyearofthebirthforagivenwomanisdeterminedbythecouple, andthuscanbeendogenous. 14 Toexaminetheimpactoffamilysizeonmothers'health,WuandLi(2012)usetheinteractions ofanon-HandummyandameasureoftimeexposuretotheOCP.Thisstrategyissimilartothe oneimplementedinthispaper.Inparticular,theyassumethattheeffectsofexposuretothepolicy islinear 15 andtheychooseanarbitraryyearwhenthepolicystartstotakeeffect.Usingourdata,I canshowthatallowingformorexibleeffectsoftheOCPoneachagecohortsgeneratessimilar butdifferentresultsfromusingtheirstrategy(TableG.1indicatethatusingstrategy inWuandLi(2012)generateslesspreciseestimatefortheeffectofchildrenonfemaleLFP,and TableG.2showstheregressionresultswhenusingastrategybasedongenderofthe withtheintervalfor ‹ b 2 SLS aswideas(-1.70,-0.019),surpassingthelimitof -1). OurDIDmethodusingethnicityismotivatedbyastudybyLietal.(2005).Inordertomeasure theeffectoftheOCPonfertility, 16 Lietal.(2005)usetheinteractionsofamother'sbirthcohorts andaHandummytoidentifytheexogenousvariationinnumberofchildrenduetotheOCP.Their robustnesstestsshowthatchangesinotherhouseholddecisions,suchasmarriageandthedecisions whethertohaveanychildren,arenotsystematicallydifferentbetweenHanandnon-Hanpeople duringthisperiod. 14 Inthesampleweuseinthispaper,IVresultsareverydifferentwhenusingyearofbirth dummiesinsteadofagecohortdummies,evenaftercontrollingforthemother'sageatbirth. Regressionresultsavailableuponrequests. 15 FigureB.5ashowsthisisnottrue,especiallywhenincludingveryyoungagecohorts. 16 TheyuseDIDdirectly,ratherthanapplyitasIV. 12 1.4DataandDescriptiveStatistics Thedatausedinthispapercomesfroma1%sampleofthe1990ChinaPopulationCensus,the fourthcensusinChina.Forstudiesonfertilityandlaborsupply,censusdatahasthedistinctad- vantagesoflargesamplesizeandnationalrepresentation.Forthisstudy,whichreliesonexposure totheOne-ChildPolicy,the1990Censushasthreeparticularadvantages.First,ifweusethe2000 censustodotheanalysis,ourcontrolcohortswouldbe50yearsoldatthetimeofthesurvey. Though55istheofretirementageforwomeninChina,manywomenretireasearlyas50 (Maurer-Fazioetal.,2011).Asaresult,thedecisionofwhetherornottoworkandhowitrelates tofertilityisnotlikelytoberelevantforthecontrolgroup. 17 Second,thoughtherewasa1982 Census,someaspectsoftheOCPandimportantexceptions(includingthe1-boy-2-girlrule)were notimplementeduntilafter1982.Third,beforethe1990s,thehouseholdmobilityinChinawas almostzeroduetotheverystricthouseholdregistration(Hukou)system.Thishelpstoreducethe concernthatfamiliesendogenouslymigrateinresponsetotheOCP. The1990ChinaPopulationCensuscontainslimitedbutessentialinformationatthehousehold andtheindividuallevel,including:age,sex,ethnicity,relationshiptothehouseholdhead,geo- graphiclocation(atcountylevel),education,employmentstatus,maritalstatus,andchildbearing status.DetailedinformationaboutthecensuscanbefoundinWang(2000).Onlythreelaborsup- plyrelatedquestionsareincludedin1990censuses:employmentstatus,industryandoccupation. IfollowMaurer-Fazioetal.(2005)tomydependentvariable,thefemalelaborforcepartic- ipation( LFP ). LFP ict = 1ifawomanhasajobonthedayofthecensusorifsheisunemployed butiscodedasfiwaitingtobeemployedfl. 18 Twotypesofchildbearingquestionswereaskedfor 17 TheeffectsofnumberofchildrenonfemaleLFPdependheavilyontheagegroup.Angrist andEvans(1998)showsthatthenegativeeffectsofchildrenonfemalelaborsupplywilldisappear forchildrenage13orolder. 18 Forindividualsaged15andabove,1990PopulationCensusaskedtheirindustriesand occupations.Forthosewhodidnotanswerthesetwo,theywereaskedquestionsfurtherabouttheir non-employmentstatus,withchoiceslistedas:1.currentlyenrolled;2.student;3.housework; 4.waitingforschooling;5.waitingtobeemployed;6.retired/resigned;7.lostabilitytowork; 8.other. 13 womenaged15to64:thebirthhistoryinthepreviousyearandthenumberofsonsanddaughters everbornandthenumberofsurviving.Nootherretrospectivefertilityinformationiscollected inthe1990Census.SimilartoAngristandEvans(1998),Imatchchildrentomotherswithinthe householdstogetdetailedinformationforthechildren. ThedifferencesbetweenurbanandruralareasinChinaarehuge.AsnotedinSection2,the implementationoftheOne-ChildPolicyinurbanareasmayhavedirectimpactsonfemalelabor demand(asthemostcommonly-usedpenaltyforabove-quotabirthisthejobloss),whichmight confoundthe2SLSestimates.Therefore,thispaperwillfocusontheeffectsoffamilysizeon femalelaborforceparticipationin rural Chinaonly.Iincludethehouseholdsthatareregisteredas agriculturalhouseholdsandalsoresidedinthecountrysideinthestudysample. 19 Inordertolink thechildrentotheirparents,thesampleisrestrictedtowomenwhoareheadsofthehouseholdsor spousesofthehouseholdheads,withatleastonechild.Idiscardasmallnumberofobservations forwhichtheageofthemotheratbirthwasunder15,aswellastheobservationsforwhichthe ageofthechildislessthan1(AngristandEvans,1998;CrucesandGaliani,2007).As olderchildrenaremorelikelytoleavethehousehold,existingliteratureusuallyrestrictsthesample towomenlessthanorequalto35yearsold(AngristandEvans,1998;CrucesandGaliani,2007). Unfortunately,Icannotfollowthatrulehere.Inordertoimplementdifferences-in-differencesI havetoincludecontrolcohortmotherswhowereolderthan29in1980.Thiscontrolcohortisaged 40orolderin1990.Therefore,insteadof35,thisstudyextendstheupperboundformother'sage to45.AngristandEvans(1998)showthattheirresultsarenotsensitivetoincreasingtheupper boundagefrom35yearsoldto45yearsold,andtheyfoundtheirresultsarenotsensitivetothat sampleselectionrule.Inoursample,FigureG.2showsthatinthe1990censusforruralChina,the averagenumberofchildrenis3.3forwomenaged45.Thisisconsistentwiththetrendsoftotal fertilityratesshowninFigureB.3,implyingthatthemoving-outofchildrenformothersyounger than45shouldnotbeabigconcern.Inaddition,mothersforwhomthenumberoflinkedchildren 19 SeetwocriteriaforclassifyingthepopulationintoruralversusurbaninWang(2000).HereI useboth. 14 didnotmatchthereportednumberofsurvivingchildrenareexcluded. 20 Withtherestrictionsabove,Iobtainedasampleof824,609womenin29provinces.A.1reports theofthekeyvariables(dependentvariables,covariates,andinstruments)andtheir summarystatisticsforthewholesample.ThedatashowthatatypicalruralChinesemotheraged between16and45hadanaverage2.22childrenin1990.Morethan75%ofmothersinrural Chinaatthattimehadtwoormorechildren.Amongthem,9%areminorityethnicgroups.And about48%ofthechildrenaregirls.About78%oftheseruralwomenhavenomorethan primaryschooleducation,andtheaverageeducationleveloftheruralwomenislowerthanthatof theirhusbands.Thelaborforceparticipationofruralwomenisover92%,whichishigherthanthe inMaurer-Fazioetal.(2005),wheretheruralfemaleLFPforwomenover15yearsoldin 1990isfoundtobe80.3%.ThereasonforthehigherfemaleLFPinthisstudyisthatwefocuson women45oryoungerwhohavearelativelyhigherLFPrate.Chinaisobservedtohavethehighest femalelaborforceparticipationintheworld.Peoplemaythinkthisistheresultofthehighly restrictedpopulationpolicy,however,asshowninFigureB.2,thefemalelaborforceparticipation wasevenhigherinthe1970sbeforetheOne-ChildPolicy.Thereareatleasttwopossiblereasons forthehistoricallyhighleveloffemalelaborforceparticipationinChina. 21 Onepossibilityisthat undertheCommunistruleinChina,womenweregrantedequalstatustomenthroughaseriesof laws.Theotherpossibilityisthatindividuallaborallocationwasnotanindividualchoicebefore 1978.Forexample,inruralareas,peoplewereassignedtoworkincollectiveagriculturebythe villagecollectives;thusalmosteveryadulthadtowork.Formoredetailsaboutthelaborforce participationinChina,refertoMaurer-Fazioetal.(2005). 20 Thismightcauseasampleselectionproblem.Toavoidsampleselectionissues,wecanusethe totalreportedfertility(surveyquestionaskedofallwomen)regardlessofwhetherthechildrenstill liveathome.Thecostofthisstrategywhichdoesnotneedtomatchmotherswiththeirchildren isthatIcannotgetinformationforgenderoftheItriedthisstrategyforethnicityDID, andtheregressionresultsareverysimilartoourmainresultswhenusingDIDbasedonethnicity. Regressionresultsavailableuponrequests. 21 In2010,thefemalelaborforceparticipationrateinChinawas67%,rankedoutof35 countries(HildaL.Solis,2012). 15 AsdescribedinSection2,differentprovincesinChinaenforcedtheOne-ChildPolicyindif- ferentways;thusIfurtherdividethiswholesampleintothreesubsamplessubjecttodifferent policyimplementations.Thesubsampleisthefi2-childrenprovincesfl,whichconsistofve provinces(Hainan,Yunan,Qinghai,Ningxia,Xinjiang)whereallruralcouplesareallowedtohave twochildren.Intheseprovinces,thereshouldbenodifferencesbetweenHanandnon-Hancouples andboyandgirlcouplesinprobabilityofhavingtwochildren.Thisallowsus toperformplacebotestsusingobservationsfromthissubsampletovalidateourinstruments.Sec- ond,fi1-boy-2-girlprovincesflarethe19provinceswithanamendedOCP,allowingruralcouples tohaveasecondchildifthechildisagirl.WecanemploytheDIDstrategyalongthedimen- sionofthegenderofthechildforthissubsample.Third,fi1-childprovincesflinclude threemunicipalities(Beijing,Shanghai,andTianjin)andtheremainingtwoprovinces(Jiangsu andSichuan).Intheseprovincesofhighpopulationdensity,allnon-minorityruralfamiliesare allowedtohaveonlyonechild,withoutexception.Inboththesecondandthirdgroups(restricted provinces,hereafter),regardlessofthefi1-boy-2-girlsflrule,theOCPismorestrictlyappliedto ethnicallyHanChinese,soIperformtheDIDaccordingtoethnicityonthiscombinedsubsample. TableA.2givesthesummarystatisticsofmajorvariablesforeachsample,andTableA.3further comparesthecharacteristicsofHanandnon-Hanmothersinrestrictedprovinces(fi1-boy-2-girl provincesflandfi1-childprovincesfl),aswellascharacteristicsofmotherswithgirlsand boysinfi1-son-2-girlprovincesfl. Ifpeopleareallowedtomove,coupleswithastrongerpreferenceforabiggerfamilysize mightmovetofi1-boy-2-girlprovincesflorevenfi2-children-provincesfl,andthiswillcontaminate ourestimatesoftheeffectsofOCPonfertilityandthereforetheestimatesoftheeffectsoffamily sizeonfemaleLFP.Thisisnotlikelyaconcerninoursample.Duetothestricthouseholdregis- tration(Hukou)system,peoplewerepreventedfrommovingingeneralbefore1990.Inthe1990 Census,peoplewereaskedabouttheirpermanentresidencein1985and99.13%oftheruralsam- plereportedtoliveinthesameprovinceasin1990(97.76%reportedtoliveinthesamecounty). Therefore,wedonotbelieveouranalysissuffersfromendogeneitycausedbypeople'spreference 16 andselectivemigration. TableA.2showsthat,comparedtothe1-boy-2-girlprovinces,therearemoreminoritiesin the2-childrenprovinces,andfewerminoritiesinthe1-childprovinces.Intermsoftheaverage numberofchildrenandtheprobabilityofhavingtwoormorechildren,2-childrenprovinceshave ahigheraverageandprobabilitythan1-boy-2-girlprovinces,whichhaveahigheraverageand probabilitythanthe1-childprovinces.1-childprovinces,meanwhile,havehigherfemalelabor forceparticipationthan2-childrenprovinces(thoughnotand2-childrenprovinces havehigherfemalelaborforceparticipationthan1-boy-2-girlprovinces. TableA.3showsthatnon-HanChinesehavemorechildrenandahigherprobabilityofhaving twoormorechildrenthanHanChineseinthe1-childand1-boy-2-girlprovinces,whilemothers withgirlsaremorelikelytohaveadditionalchildrenandbiggerfamiliesin1-boy-2- girlprovinces.Therearenodifferencesintheagepatternsofgivingbirthformothers inthesegroups,especiallyforthetimingofthesecondchild,thustheinteractiontermsinthe regressionwillmainlycapturethevariationsduetothepolicychangeratherthanthedifferences inbirthtimingpreferences(alsoseeninFigureG.3a).Ingeneral,however,Hanandnon-Han womenhavesomedifferentfeatures.Forexample,theaverageeducationlevelforHanpeople ishigherthanfornon-Hanpeople.DIDisneeded,therefore,toremovetheselevel differences. TableA.4andTableA.5showourbasicDIDestimatesofhavingtwoormorechildrenand laborforceparticipation,basedonethnicityandgenderofrespectively.ThoughtheOne- ChildPolicyhadanexplicitimplementationdateof1980,thereisnosimpledistinctionbetween notreatmentandtreatmentforeachindividual,aspeoplechoosetheirfertilitytimingdifferently. AsdiscussedinSection3,Icategorizemothersintotreatedandpre-treatmentcohortsbasedon theirages.Thecutoffageof30in1980meansthatthecutoffageis40in1990.Womenolderthan 40butyoungerthan46in1990censuswillberegardedasfiOld(Pre-treatment)Cohortsfl.Onthe otherside,thefiYoung(Treatment)CohortsflinTableA.4andTableA.5includemothers40years oldoryoungerin1990,i.e.,agecohorts16to40.Thelowerboundis16,becauseweexclude 17 womenwhohadchildrenwhenlessthan15yearsoldandwomenwhosechildwaslessthan 1.Thisdistinctionoftreatmentstatusisconsistentwiththeresultsfromtheregressionswithafull setofagecohortdummiesinSection5. TableA.4showsthattheprobabilityofhavingtwoormorechildrendecreasesforbothHan andnon-HanChineseaftertheOne-ChildPolicy,butdecreasesmoreforHanpeople by3.7percentagepoints.Althoughthefertilityofthenon-Hanpopulationwasnotofcon- strainedtoonlyonechild,thereareotherfactorssuchasincreasedincome,amonetarybonusfor one-childfamilies,andeasieraccesstofamilyplanningservices,thatmighthavereducedoverall fertilityinthisperiod,regardlessofethnicity.TherightpanelofTableA.4showsthattheDID estimateofthereducedformeffectoftheOCPonthelaborforceparticipationrateis-0.029. Similarly,TableA.5showsthatforbothmothersofgirlsandboys,fertilityde- creasedaftertheOne-ChildPolicytookeffect.Inaddition,thedeclineintheprobabilityofhaving twoormorechildrenislargerforcoupleswithboys.Ontheotherhand,theincreasein laborforceparticipationduringthisperiodissmallerformotherswithgirls. OnethingtonoticefromTableA.4andTableA.5isthattheOne-ChildPolicydidnotlead toalargenumberoffamilieswithonlyonechildinruralChina.Evenfortheyoungcohorts withboys,morethan70%stillhavemorethanonechild.Thisfactmaymakeexposure totheOCPaweakinstrumentforfertility,yetthehugesamplesizecanresolvethisproblemto someextent.Ontheotherside,weakenforcementoftheOCPdoesnotaffectthevalidityofthe instrument. 22 ThereareseveralpossiblereasonsforthelowcompliancerateinruralChina.First, withtherelaxationoftheOCPin1984,manyruralhouseholdswithgirlswereallowed tohaveasecondchild. 23 Second,itmaybediftofullyenforcetheOne-ChildPolicyin ruralChina,asthetheonlyseverepunishmentinruralareasforabove-quotabirthsisaone-off Moreover,eventhemaynotbeveryeffectiveinruralareas,becausemanypoorfarmers cannotaffordtopay(Lietal.,2005).Third,ruralhouseholdshavestrongincentivestodisobey 22 Allour2SLSregressionspasstheCragg-Donaldweakinstrumenttestatatleast5%level. 23 Thisonedoesnotexplainwhyhouseholdswithboyshavemorethanonechild. 18 theOne-ChildPolicy,aschildrenarevaluedinputstofarmlabor(SchultzandZeng,1995)and forprovidingoldagesupport,sincesocialsecurityandpensionsystemsinruralChinaarevery limited.However,intheurbanareaswheresocialsecuritysystemarebetterdeveloped,nofarm laborisneeded,andmoreseverepunishmentsareimplemented,sotheprobabilityofhavingtwo ormorechildrenismuchlower.Fromthedatainthe1990PopulationCensus,onlyabout30%of theurbancoupleshavemorethanonechild. 1.5RegressionResults 1.5.1OLSestimatesoftheeffectofadditionalchildrenonfemaleLFP PanelAofTableA.6showtheresultsofestimatingequation(1)usingOLS.Allregressionsinclude agecohortdummies,mother'sageatbirth,educationlevelsforbothparents,andcountyed effects.Standarderrorsareclusteredatthecountylevel.Columns(1)and(2)arebasedonsam- plesofobservationsfromrestrictedprovinces(1-childprovincesand1-boy-2-girlprovinces)and 1-boy-2-girlprovincesrespectively.Bothregressionsshownoeffectsofadditionalchil- drenonwomen'slaborforceparticipation.Column(3)andColumn(4)arebasedonobservations inrestrictedprovincesand1-boy-2-girlprovinces,exceptformotherswhosebirthoccurred laterthan1981.Chen,LiandMeng(2013)showthattherewasajumpinboththeabortionrate andthesexratioin1982(FigureG.4),thoughthesexratioforlooksstableoveryears. Tobeconservativehere,Idropallsampleswithpossiblynon-exogenousgender.More detailsaboutthissampleselectionrulearediscussedinSection6.Inthishighlyskewedsample, bothwomenandtheirchildrenareolder,soweexpecttheeffectsofchildrenonmother'slabor supplytobesmaller(AngristandEvans,1998).TheOLSestimatesinColumn(3)and(4)suggest thatcomparedtomotherswithonlyonechild,motherswithtwoormorechildrenhave1.3-1.5 percentagepointshigherprobabilitytowork.TheseOLSestimatesareverydifferentfromthe researchindevelopedcountriessuchastheU.S.,whereAngristandEvans(1998)report theirOLSestimatesoftheeffectofthreeormorechildrenonfemaleLFPtobearound-15%. 19 1.5.2FirstStage:theeffectsoftheOneChildPolicyonfertility AsdiscussedinSection3,OLSestimatorof b inEquation(1)islikelytobeinconsistentduetothe endogeneityoffertilitywithrespecttofemalelaborforceparticipation.Theimplementationofthe One-ChildPolicyallowsustoisolatetheexogenousvariationinfertility.DoingDIDtocapturethe exogenousvariationsinfertilitycausedbyexogenousvariationsinpolicychangeisfundamentally equivalenttousingcorrespondinginteractiontermsinaregressionfortheprobabilityofhavingtwo ormorechildren.Column(1)inTableAdisplaysourstageregressioncoefoffiMore thanonechildfl( kids 2 ict )ontheinteractiontermsofagecohortsandethnicity( r l inEquation(2)). SincetheOCPwasnationallyimplementedin1980,andmostwomenwithtwoormorechildren completedtheirsecondbirthatorbeforeage30,onlyfiTreatment(Young)Cohortsflwhowere atorunderage30in1980wouldberestrictedbytheOCPandthusmostlikelytochangetheir familysizebecauseoftherestrictionfromtheOCP.Therefore,weexpecttheinteractionstobe positiveinthestageforcohortsyoungerthan40,butnotforolderage cohorts.Thisisexactlywhatweonlyinteractionsforagecohorts40oryoungerarepositive, andonlyinteractionsforagecohorts38oryoungerarepositive.Forwomenaround 30yearsoldin1990theOne-ChildPolicydecreasedtheprobabilityofhavingadditionalchildren byabout8percentagepointsmoreforHanwomenthanitdidfornon-Hanwomen.Noticehere thatsomeveryyoungcohortdummiesarenotTherearetwopossiblereasonsforthis: thesmallnumberofobservationsinthosecohorts,andsecond,theyaretooyoungtohave theirlifetimefertility.ThecoeffromColumn(1)inTableAareplottedinFigure B.5a(theblueline). OursecondsetsofDIDestimatesarebasedonthedifferentpoliciesonsecondbirthsforHan coupleswithst-borngirlsandboysinthe1-boy-2-girlprovinces.Column(1)inTa- bleA.8displaysourstageregressioncoefontheinteractiontermsofagecohortsand whetherthewasagirl( f l inEquation(3)).SimilartoourinTableA.8,interac- tionsareonlypositiveforcohortsyoungerthan40.Forwomenaround30yearsold,theOne-Child 20 Policydecreasedfertilityformotherswithaboybyabout5percentagepointsmorethan motherswithagirl.ThecoeffromColumn(1)inTableA.8areplottedinFigure B.5b(theblueline). Toincreasethepoweroftheinstruments,Ifocusonmothersage30andabovewhoaremore likelytocompletetheirsecondbirthsinTableG.1.Usingindividualsage40andaboveasthe controlgroup,thereportslargerF-statistics. 1.5.3Reducedform:theeffectsoftheOne-ChildPolicyonfemaleFLP Thereducedformregressionscanbeexpressedas: LFP ict = 44 å l = 16 ( nonHan ict d l ) t l + X 0 ict z + a 4 + d t + w c + u ict (1.4) LFP ict = 44 å l = 16 ( First - BornGirl ict d l ) h l + X 0 ict n + a 5 + d t + s c + v ict (1.5) where d l ; l = 16 to 44areagedummiesfrom16to44yearsofage.Individualsaged45in1990 formthecontrolgroup,andareomittedfromtheregression.Oneusefulcheckofinstrument validityistoseeitseffectontheuntreatedgroup.Asthefertilityofthefipre-treatmentcohortsfl arenotaffectedbyOCP,thecoef t l and h l areexpectedtobe0for l > 40.Ourin Column(1)ofTableAandTableA.10areclosetothis.ForColumn(1)inTableA,theinteraction termsofagecohortsandnon-Hanareonlynegativeforwomenaged41oryounger. AndforColumn(1)inTableA.10,theinteractiontermsofagecohortsandFirst-BornGirlare rarelynegativeforwomenover40.Forwomenaroundage30,comparedtowomen atage45,thedeclineofHanmothers'LFPisabout3percentagepointslargerthanthatfornon-Han mothers,andthedeclineinrst-bornboymothers'LFPisabout2.3percentagepointslargerfor motherswithboysthanformotherswithgirls.ThecoeffromColumn (1)inTableAandTableA.10areplottedinFigureB.5a(theorangeline)andFigureB.5b(the orangeline)respectively.(Columns(2)and(4)inTableG.1showsimilarresults.) 21 1.5.42SLSestimatesoftheeffectofadditionalchildrenonfemaleLFP PanelBinTableA.6reportsthe2SLSestimatesoftheeffectofadditionalchildrenonfemale laborforceparticipationinruralChina.UsingDIDbasedonethnicity(genderofbirth)as instruments,theresultsshowthat,otherthingsequal,formothersunder46yearsoldandwithone child,havingadditionalchildrenwilldecreasethepossibilityofworkingby15.3%(8.4%)inrural Chinain1990.Thesetwoestimatesareclosetotheestimatedeffectsofthreeormorechildren onfemaleLFPintheU.S.(between-9.2%and-12%inAngristandEvans(1998)),Mexicoand Argentina(between-6.31%and-9.58%inCrucesandGaliani(2007)),andTaiwan(-12.6%in Ebenstein(2009)).(Whenfocusingonmothersage30andabove,andusingindividualsage40and aboveasthecontrolgroup,TableG.2displaysimilarresults,whichsuggestthat,havingadditional childrenwilldecreasemother'slaborforceparticipationby12.1%-12.9%.) Whilethetwoinstrumentsyielddifferenteffectsizes,thetwocoefarenotstatistically differentfromeachother.WhydothetwosetsofIVgeneratesuchdifferentestimates?The differentpowersofthetwoIVsetsmightbeonereason.Whenusingson-preferencetoestimatethe effectsoffamilysizeonfemalelaborsupplyinTaiwan,Ebenstein(2009)showsthatIVestimates willdropwhentheinstrumentisweaker.Basedonsimulateddatafromthestructuralmodel, 24 heshowsthattheestimatesbasedonweakerinstrumentsaresmallerthanthetrueaveragecausal effect,andthestrongertheinstruments,theclosertotheaveragecausaleffecttheestimatesare. TheCragg-DonaldWaldFStatisticshowthatourinstrumentfromDIDbasedongenderofthe isweakerthantheinstrumentfromDIDbasedonethnicity.Differentlocalaverage treatmenteffectsestimatedbythetwosetsofIVanddifferentsampleforestimationmightalsobe thereason. Hausmantestsshowthatthe2SLSestimatesusingDIDbasedonethnicity(genderof birth)arestatisticallydifferentfromOLSestimatesat1%(5%)level.Our2SLSestimatessuggest alargernegativeeffectofchildrenonfemalelaborsupplythanOLSestimates.Thisdiffersfrom 24 Ebenstein(2009)usesamodelofamother'sjointdeterminationoffertilityandlaborsupply allowingforunobservedheterogeneityinboththeandcostsofchildren. 22 mostofthereportedinthepreviousliterature.Thereareatleasttwopossiblereasons forthisFirst,asinmanyotherdevelopingcountries,thefertilityofruralhouseholdsin Chinamightbedeterminedthroughacollectivebargainingprocess,inwhichonlyindividualswith morebargainingpowercandecidethetotalnumberofchildren(Rasul,2008).Inthiscase,if womenwithlessbargainingpowerareforcedtoworkmoreaswellashavemorechildren,then theOLSestimateswillunderestimatethenegativeimpactsofnumberofchildrenonlaborforce participation.Second,controllingfortheendogeneityoffertilityremovesthepossiblefactors thatpromotefertilityandfemaleLFPsimultaneously.Forexample,inruralChina,ifpeople withahigherearningcapacitycanaffordtohavemorechildrenandalsohavemore opportunitiestowork,thenthissimultaneitywillbiastheOLSestimatesup(Fangetal.,2010). 1.5.5HeterogeneousEffects InthissectionIexplorewhetherthefemalelaborforceparticipationresponsetofertilityisuniform orvariesbymothers'education.ExtendingGronau's(1977)modelofmarketandhomeproduc- tion,AngristandEvans(1998)incorporatechildqualityeffectstothemodeloffertilityandfemale laborsupply.Theirmodelpredictsthatthelaborsupplyofmoreeducatedwomenismoresensitive tofertility,andthusthenegativeeffectsofadditionalchildrenonLFPislargerforwomenwith highereducationlevels.HereweuseexogenousvariationinfertilitybroughtbytheOCPasin- strumentstoexplorehowthelabormarketconsequencesofchildbearingwouldvarywithmothers' education.Thisisdonebyrunningseparateregressionsonwomenwithatmostprimaryschooled- ucation(73.48%oftherestrictedprovincessample,73.24%ofthe1-boy-2-girlprovincessample), andwomenwithatleastjuniorhighschooleducation(26.52%oftherestrictedprovincessample, 26.76%ofthe1-boy-2-girlprovincessample).FigureB.6adepictsthestageandreducedform coefofinteractiontermswhenusingDIDonethnicityasinstruments.Itshowstheeffect oftheOCPonfertilityislargerforwomenwithatleastajuniorhighschooleducation.Onthe otherhand,thereducedformregressioncoefarecloseforwomenwithdifferenteducation levels.FigureB.6brepresentsthecoefofinteractiontermswhenusingDIDongenderof 23 birthasinstruments,anditshowssimilartrendasFigureB.6a.SinceWaldestimatesequals totheratioofreducedformcoeftotstagecoeftheseresultssuggestthatwomen withlowereducationtohavelargernegative2SLSestimatesoftheeffectoffertility.The2SLS estimatesinTableA.14this,thoughthedifferencebetweenestimatesforlowerandhigher educatedwomenisnotstatistically Theseresultscontradictthepredictionsofthetheoreticalmodel,yetareconsistentwiththeem- piricalinAngristandEvans(1998),whichsuggestthatthelaborsupplyconsequencesof childbearingaresmallerformoreeducatedwomen.Theresultspresentedherearemerelydescrip- tiveandshouldnotbeover-interpretedbecausemanyestimatesareandeducationis correlatedwithotherindividualpreferencesthatmayaffectthelaborsupplydecisions. 1.6ValidityofInstruments ThekeyassumptionunderlyingmyDIDestimationframeworkisthattheinstrumentsdonotaffect femalelaborsupplythroughchannelsotherthanfertility.Inotherwords,Iassumethatdifferences infemaleLFPbetweenpre-treatmentandtreatednon-Hanminorities(motherswithgirls) shouldbethesameasthedifferencesbetweenthepre-treatmentandtreatedHanmothers(mothers withboys)intheabsenceoftheOCP.ThevalidityofourDIDinstrumentsareextensively discussedandtestedinthissection. 1.6.1DIDusingethnicity First,ifthedifferentialimplementationoftheOne-ChildPolicybetweenHanandnon-HanChi- neseisendogenous,thentheDIDstrategyintermsofethnicityisnotvalid.Drawingonresearch insciencestudiesandearlydocumentsinChina,Greenhalgh(2003)concludedthatthedecision toexcludenon-HanpeoplefromtheOne-ChildPolicywasdrivenbypurepoliticalconsidera- tionsratherthanbydifferentialfertilityratesorothereconomicfactors.Therefore,weregardthe exemptionofnon-Hanpeopletobeexogenous. 24 Second,ifaroundthesametimeastheOCPthereareotherpoliciesthatalteredthepreferences and/orlabordemandsforHanandnon-Hanpeopledifferently,thentheDIDinstrumentswillpick upthoseeffectsandviolatetheexclusionrestriction.Iamawareofonlyoneethnically-divided policyfrom1978-1984thatmayleadtosuchaconcern.InMarch1981,theStateCouncilreleased thefiReportonthe1981NationalCollegeEnrollmentConferencefl.Accordingtothereport,ethnic minoritieswereabletoentercollegewithlowergradesandlowertuitionfees.Thispreferential policywasaccompaniedbyotherlocaleducationpoliciesfavoringnon-Hanpopulations.These policychangesmighthavepromotedtheeducationlevelsofnon-HanChineserelativetotheHan Chinese.Thisinturnmightreducethelaborforceparticipationforyoungnon-HanChinese,who mightchoosetogotoschoolratherthanworkafterthepolicychange.Totestthispossibility,I examinedthechangeineducationlevelsfornon-HanpeoplerelativetoHanpeoplethroughthe followingregression: Educ ict = 44 å l = 16 ( nonHan ict d l ) l l + a 6 + d t + w c + u ict (1.6) Where Educ ict istheyearsofeducationforwomen i ,incounty c ,agecohort t .Column(1)inTable A.11showstheestimatesof l l .Theestimatesindicatethattheeducationlevelofnon-Hanwasnot promotedbythepreferentialpolicies.Onthecontrary,comparedtowomenaged45,theeducation disadvantagefornon-HanrelativetoHanpeoplewasevenbiggerfortheveryyoungcohorts. ThoughIcannotexplainwhythatisthecaseinthispaper,Itriedtoexcludetheyoungcohorts fromthemainregressionandtesttherobustnessofmyTableG.3reportstheeffectsof fertilityonfemaleLFPusingDIDusingethnicityasIVwhenexcludingmothersyoungerthan27 (fortheyoungercohorts,non-Hanmothershavewideneddisadvantagesineducation). TheresultsaresimilartothemainresultsinTableAandTableA,with2SLSestimatesdroppinga littlefrom-0.153to-0.139,andthesetwoestimatesarenotstatisticallydifferent. Thethirdconcernisthatinordertohavemorethanonechild,someHancouplesmayhave changedtheirethnicitytonon-HanaftertheimplementationoftheOne-ChildPolicy.Although thereissomeanecdotalevidenceofpeoplechangingtheirethnicity(Scharping,2003),suchre- 25 isnotpopularinChina(Lietal.,2005).Infact,beforetheyear1981,whentheState CouncilannouncedtheCircularofRestoringandCorrectingEthnicity,itwasalmostimpossible forpeopletochangetheirregisteredethnicity.Therefore,wecandoarobustnesscheckexcluding motherswithbirthafter1981. 25 Column(3)and(4)inTableA.6,andColumn(2)inTable A-TableA.10aretheregressionresultsbasedonthisrestrictedsample.Thistruncatedsample hasolderchildrenthanthatusedinmainregressions,andthuswemayexpectsmallernegative effectsoffertilityonmother'slaborforceparticipation. 26 However,the2SLSestimatesinTable A.6shownodifferencesbetweenthetruncatedandthemainregressionsamples. 1.6.2DIDusinggenderofthe ThefimissinggirlflproblemisabigchallengeforChina'spopulation.Sincethe1980s,afterthe implementationoftheOCP,theseximbalanceofchildrenhasincreasinglyfavoredboys(Lietal., 2011).Giventhis,wemightbeconcernedthatthesexofchildrenisendogenousdueto sexselection.Thoughendogeneitymightbetrueforthehighparitybirths,forthegenderofthe birththethreatismuchlesssevere.Chenetal.(2013)thataccesstotheB-ultrasound isnotassociatedwithanychangeinthesexratiooftbirths,whiletheincreased localaccesstoultrasoundtechnologyisfoundtosubstantiallyincreasethesexratio.Theirdata fromthe1992ChineseChildrenSurveyalsoimpliesthattheabortionrateisreallylowforthe birth,andthesexratioofthebirthisratherstablebothbeforeandaftertheimplementationof theOne-ChildPolicy.FigureB.4depictsthepercentageofgirlbirthsbytheageofthe childreninthe1990Census.Theverticallineisforchildrenage6,whowerebornin1984, theyearoftherelaxationoftheOCP.Thesecondverticallinedenoteschildrenage11,whowere 25 Ideally,weshouldtestwithmothersolderthan30in1981,butthatwouldleadtoavery smalltreatmentgroup,sowecompromisetoexcludewomenwithafter1981only.The ideaisthat,itwouldbemuchmorediftochangeethnicityforbothparentsandchildinthe household. 26 Ichecktheageofyoungestchildinthissample,thattheaverageageofyoungestchildis 6.6,andover50%ofthemarelessthan6yearsold. 26 bornwhentheOCPwasannouncedin1979.Forallthreesubsamplesofprovinces,wedonot seeanychangeinsexratiobeforeandaftertheOCPorthetherelaxationoftheOCP. AnotherwaytotesttheeffectsoftheOCPonsexratioistorunDIDregressionsofrst-born genderonethnicity.Asnon-Hanfamiliesareallowedtohavetwochildren,theydonothavean additionalincentivetohavethebeaboyaftertheimplementationoftheOCP.Givenno sexselectionofthefornon-HanfamiliesduetotheOCP,wecancomparethe genderbetweenHanandnon-Hanpeople.Ifthereisnodifferencebetweenthem,then wearemorethatthegenderofthein1-boy-2-girlprovincesisnotendogenous aftertheimplementationofOCP.TheDIDregressionscanbeexpressedas: First - BornGirl ict = 44 å l = 16 ( nonHan ict d l ) x l + a 7 + d t + w c + u ict (1.7) Column(2)inTableA.11istheestimatesof x l ,whicharenotstatisticallyforallage cohorts.Thisprovidesevidencefortheexogeneityofthegenderofbirth. Chen,LiandMeng(2013)showthattherewasajumpinbothabortionrateandsexratioin 1982(FigureG.4),thoughthesexratioforlooksstableoveryears.Tobeconservative here,wecandropallsampleswithpossiblynon-exogenousgender,andrunarobustness checkonthetruncatedsampleexcludingmotherswithafter1981.Theregressionresults arereportedinColumn(3)and(4)inTableA.6,andColumn(2)inTableA-TableA.10.Asex- plainedabove,therobustnesschecksontherestrictedsamplegenerateresultsthatareverysimilar totheresultsfromthemainregression. 1.6.3Placebotestson2-Childrenprovinces Iftherearepolicyshocksorchangesinsocial-economicvariablesotherthantheOCPinthesame periodthathaveaffectedthefemalelaborforceparticipation,thentheDIDmethodmayconfound theeffectofthesepoliciesorchanges.Asaftest,wecanrunthestageandreduced formregressionsforobservationsinthefi2-childrenprovincesfl,whereallcouplesareallowed tohave2children.Asthereisnovariationintermsoftheeligibilityofhavingasecondchild 27 betweenHanandnon-Hanmothers,ormotherswithgirlsandboys,weexpect theinteractiontermsofnon-Han( First - BornGirl )andagecohortstobezeroforbothstage andreducedformregressions. TableAandTableAshowtheregressionresultsinthefi2-childrenprovincesfl.TableAdisplays theandreduced-formcoefwhenusingDIDbasedonethnicityasIV,andTable AshowsthecoefofinteractiontermswhenusingDIDbasedongenderofbirthas instruments.Wecanthat,interactiontermsofagecohortsandnon-Han/genderofbirth arerarely 27 forbothstageandreducedform.Thuswenoevidencethatthe trendsinfertilityandfemalelaborforceparticipationforHangirl)andnon-Han bornboy)mothersaredifferentintheseplaceboprovinces. FigureB.5ashowsthecoefofinteractionterms, nonHan ict d t ,byagecohortsint stageandreducedformregressions.Thesolidlinesarefromstages,whiledashedonesare fromreducedforms.Fortherestrictedprovincesdenotedbythickerlines,bothand reduced-formcoefarearoundzeroforcohortsagedolderthan41,andthentheydepart fromeachotherastheagebecomesyounger.For2-childrenprovincesrepresentedbythinner lines,bothandreducedformcoefarearoundzeroforallcohorts,especiallyfor thereducedforms.Similarly,thecoefofinteractionterms, First - BornGirl ict d t ,byage cohorts,forbothstageandreducedformregressionsaredepictedbyFigureB.5b.Whileit showssimilarpatternstoFigureB.5ainthestagecoefthedeclineofcoefof reduced-forminteractionsforyoungercohortsseemtobesmaller.That'swhyour2SLSestimate ofeffectsonfemaleLFPbasedonthissetofIVsisinsmallermagnitude. 27 Theinteractionsofagecohortandnon-Hanarepositiveforagecohortsyounger than28.Doingfurtherinvestigationsintothedata,Ithatmightduetotheearlierbirthsfor non-Hanin2-Childrenprovinces.FigureG.3bshowsthedifferenttimepatternsofsecondbirthfor Hanandnon-Hanin2-Childrenprovinces,indicatingnon-Hanusuallyhaveasecondbirthearlier intheseplaceboprovinces.Meanwhile,FigureG.3ashowsthat'snotaconcernforrestricted provinces. 28 1.7RobustnessCheck 1.7.1Usingtripledifferencesasinstrument IfthetrendsofHanandnon-HanChinese(ormotherswithboysversusgirls) intermsoffemalelaborsupplyaredifferentintheabsenceoftheOCP,thentheinstrumentsbased onethnicity(orgender)willnotbevalid.However,wecanstillusethetripledifferences, [( fb _ girl ; After ; Han fb _ boy ; Before ; Han ) ( fb _ boy ; After ; Han fb _ boy ; Before ; Han )] [( fb _ girl ; After ; nonHan fb _ girl ; Before ; nonHan ) ( fb _ boy ; After ; nonHan fb _ boy ; Before ; nonHan )] tocapturetheexogenousvariationinfertilitycausedbyexogenousvariation inthepolicychange.Inthistripledifferencestrategy,theassumptionweneedisthedifference intrendsofHanandnon-HanChineseisthesameformotherswithgirlsand boys.ThisassumptionisweakerastheeffectoftheOCPisoffofdifferencesacross ethnicities,genderoftherstbornchild,andcohort.Thestageregressioncanbeexpressedas: kids 2 ict = 44 å l = 15 ( nonHan ict First - BornGirl ict d l ) z l + 44 å l = 15 ( nonHan ict d l ) J l + 44 å l = 15 ( First - BornGirl ict d l ) s l + nonHan ict First - BornGirl ict t + X 0 ict n + a 8 + d t + j c + w ict (1.8) Theinstrumentsherearethetripleinteractionsofdummyvariablesforwhetherthehousehold membersareHanChinese,whetherthebornisagirlandagecohort.Wecanonlyrunthis onthesampleof1-boy-2-girlprovinces.Column(2)inTableA.15isthe2SLSregressionresults usingtripledifferenceasinstruments.TableG.4impliesthatthetripledifferenceisnotvery powerfulinthestage.Asaresult,wedonnothavea2SLSestimatesinTable A.15.Thoughtheestimateisnotprecise,thepointestimateisstillnegativeandlargerthanthe OLSestimatesinmagnitude. 29 1.7.2Usetwinningasinstrument Someresearchusetwinningasaninstrumenttodealwithendogenousfamilysize.Forexample, AngristandEvans(1998)usebothsame-sexandmultiplebirthsasinstrumentsforhavingthree ormorechildren.Theysmallernegativeeffectsoffertilityonmaternallaborsupplyusing multi-birthasinstrument,whichtheyattributetothefactthatathirdchildduetotwinningisolder thanthirdchildreninothercases.Additionally,fitwinningflitselfmayaffectmother'slaborforce participation(RosenzweigandWolpin,2000). Toprovideanadditionalrobustnesscheck,Ialsotrytoestimatetheeffectsofadditionalchil- drenonfemalelaborforceparticipationusingmulti-birthasinstrument.Monthofbirthisreported inthe1990ChinaPopulationCensus,soIusethisinformationtoidentifymulti-births.Inthe wholesample,0.38percentofbirthsaremulti-births.Inthe1-boy-2-girlsubsample,thein- cidenceofmulti-birthinbirthissimilar,0.39percent.Thisoccurrencerateislowerthanthe inAngristandEvans(1998), 28 whichimpliesthatthemanipulationofmulti-birthwas relativelylowinruralChinaatthattime.Column(3)inTableA.15displaystheregressionresults whenusingmulti-birthasinstrument.The2SLSestimateisstatisticallyntat5%level, butsmallerthanthecoeffromthemainregressionusingDIDestimatorsasinstruments.It impliesthatformotherswithonechild,theadditionalchildrenduetofitwinningflwilldecrease theirlaborforceparticipationby6.9percentagepoints.Thevalidityofthefitwiningflinstruments andthedifferencesinthelocalaveragetreatmenteffectsmaycontributetothesmallereffectsI here. 1.8Conclusion Theimportanceofchildreninfemalelaborsupplydecisionshaslongbeenrecognizedbyeconomists. Thispaperexaminestheeffectofhavingtwoofmorechildrenonmother'slaborforceparticipation 28 Intheir1980marriedsample,byquarterofbirth,theprobabilityofmulti-birthis 0.83percent. 30 inruralChina.Itresolvestheendogeneityproblembyinstrumentingfertilitywithexogenousvaria- tionscausedbytheOne-ChildPolicy.Byexploitingvariationinthepolicy'simplementationacross ethnicitiesandgenderofornchildren,Iconstructtwosetsofdifferences-in-differences.The DIDestimatesindicatethattheOne-ChildPolicyhasnegativeeffectsonfertilityforthetargeted populations.UsingthesetwosetsofDIDestimatesasinstruments,Ithathavingtwoormore childrendecreasesthemother'slaborforceparticipationby8-15percentagepointsinruralChina in1990.ComparingdatafromthreeChinaPopulationCensuses(1982,1990,and2000),Maurer- Fazioetal.(2011)suggestthatduetoincreasedlevelsofincomeandmorefreedomofchoicefor laborallocation,thenegativeeffectsofyoungchildrenonfemalelaborsupplyhaveincreasedin China.Ifthatisthecase,thediscouragingeffectsofchildrenonfemalelaborsupplymaybeeven largernow. RecentlytherehasbeenacallfortherelaxationoftheOne-ChildPolicy(Feng,2010).In November2013,theChineseCommunistPartyreleasedfiDecisiononMajorIssuesConcerning ComprehensivelyDeepeningReformsfl,statingthatfiChinawillstarttoimplementthetwo-child policyforthecoupleswhereeitherthehusbandorwifeisfromasinglechildfamilyfl.Thispaper providesaperspectiveforthepotentialeffectsofsuchpolicyrelaxationsonfemalelaborsupply. Withtwoormorechildren,womenwillbemorelikelytostayathome,ratherthanwork,atleast inruralareas. 31 CHAPTER2 HOWDOJOBDISPLACEMENTSAFFECTFERTILITYINTHEU.S. 2.1Introduction Economicrecessionsaffectpeople'sbehaviorinmanyways,mainlythroughreducingconsump- tion.Recessionsmightaffectpeople'sfertilitydecisionsaswell,asshownbyFigureD.1.The GreatDepressionisoneexample,wherethefertilityratedroppedfrom4.11childrenperwomen in1921to3.14childrenperwomanin1933. 1 Thedeclineofthefertilityrateduringtheenergy crisisin1970sandtherecentrecessionof2008aretwootherexamples.Eventhemediareports thistrend.Forinstance,toshowtheirconcernaboutpotentialbabybust,anarticleintheLosAn- gelesTimesinDecember2008stated:fiBirthratestypicallydeclineduringeconomicdownturns. Would-beparentsstrugglewiththewisdomofwaiting.fl FigureD.2depictsbothunemploymentrateandbirthrateintheU.S.from1973to2012. 2 Let'stakeacloserlookatthethreepeaksofbirthratesinFigureD.2.First,in1990,thebirthrate increasedfrombelow16inthe1980sto16.7per1000,whiletheunemploymentratewas5.4%for women,and5.2%formenin1989,muchlowerthantheaverageunemploymentratesin1980s. Lateron,in2000,boththemaleandfemaleunemploymentratewasaslowasaround4%,andat thesametime,thebirthratereverseditsdecliningtrendsandreachedalocalpeakat14.4per1000. Beforetherecentrecession,theunemploymentratewas4.5%in2007,andthebirthratewas14.3 per1000,higherthantheratesafter2009,whicharelessthan13. Inspiredbytheoppositetrendsinfertilityandunemploymentrate,scholarsinvestigatethe effectsofrecessionsandunemploymentonfertilitydecisionsusingmacrodata.Manyresearchers 1 Hereisthetotalfertilityrate(TFR)fortheU.S.TFRisthesumofage-specibirthrates forwomenwhoare15to44yearsold.TheformulaforTFRatyear t canbewrittenas TFR t = S 44 a = 15 BirthRate t ; a 1000. 2 Birthrateisthetotalnumberofbirthsper1,000ofapopulationinayear. 32 indemographicsandeconomics(Becker(1960),Ben-Porath(1973),Adsera(2005),Adseraand Menendez(2009),CurrieandSchwandt(2014),tonamebutafew)recordedaprocyclicalpattern infertilitythatis,duringtheperiodswhenunemploymentratesarehigher,birthratesareusually lower,andmaternityissomewhatpostponed. 3 Thereareseveralconcernsaboutanalyzingtheeffectsofaggregateunemploymentlevelon fertility.Thechallengeisthat,withthegeneralequilibriumeffects,itisalmostimpossibleto getsomereallyexogenouschangeinunemploymentlevelthatisunrelatedtochangeinfertility. Theendogeneityproblemofunemploymentleadstodoubtoncausality.Forinstance,usingdataof womencohortsbystateandyearofbirth,CurrieandSchwandt(2014)analyzetheeffects ofunemploymentrateexperiencedbyeachcohortatdifferentagesontheirfertility.Theircausal estimationmightbebiased,ifthereexistssomestatetrendsinwomen'srightsmovement,which raisefemalelaborforceparticipation,increasingunemploymentrate,andreducefertilityatthe sametime.Oriftherearesomeexogenousimprovementineducationalattainmentsatthestate level,whichleadtoreductioninbothunemploymentrateandfertilityrate.Inaddition,fertility responsetounemploymentmightdifferfordifferentgroupsofindividuals.Forexample,Dehejia andLleras-Muney(2004)thatwhenunemploymentincreases,thenegativeeffectsonfertility forblackwomenislargerthanthatforwhitewomen,whileHoynes(2002)showsthatlowskilled womenexperiencegreatercyclicaleffectsthanhighskilledmen.Withtheseheterogeneouseffects, theinterpretationoftheprocyclicaltrendsoffertilitywarrantsmoreattention.Tosolvethesetwo problems,Ananatetal.(2011)estimatedtheeffectofcounty-levelforcedjoblossduetoalayoffor closingratherthanunemployment.Shearguesthatsuchjoblossescanberegardedasanexogenous shocktotheworkers,sotheestimatescancapturethecausaleffects.Additionally,sheconsiders theeffectsseparatelyforwomenwithdifferentdemographics. TherearestillsomelimitationstoAnanatetal.(2011).First,wemaywanttoknowwhich 3 Therearesomeexceptions:ButzandWard(1979)noticedthatin1970sintheU.S.fertility trendswerelikelytobecomecounter-cyclical;Ermisch(1980)andErmisch(1988)showedthat counter-cyclicalfertilitytrendsemergeintheFederalRepublicofGermanyandBritainduringthe 1970saswell. 33 groupchangestheirfertilityfacingthelocaleconomicdownturn:aretheythepeoplewholose theirjobs,orthepeoplewhofacetheriskofdisplacementbutdonotactuallylosetheirjob? Adsera(2011)arguethatthestrongfeelingofeconomicsinstabilityratherthancurrentincomeloss mighthavestrongimpactsoncurrentfertilitydecisions.Theaboveresearchcannotdistinguishthe effectsforthesetwogroups.Second,thestaticmodelcannottelluswhethertheobservedchanges infertilityaretemporaryorpermanent.Thatis,willpeople'slifetimefertilitybechangeddueto thejobloss?Usingindividual-levelpaneldatacanhelpanswerthesetwoquestions. UsingdatafromthePanelStudyofIncomeDynamics(PSID),Lindo(2010)examinedhowa man'sjoblossaffectshisfertility.HisOLSestimatesshowthatmalejobdisplacementincreases fertilityintheyearimmediatelyafterdisplacement,buttheeffectbecomesnegativeafterthesecond year.Thetotaleffectonfertilitybytheeighthyearafterjobdisplacementisslightlynegative,and thelifetimefertilityislowerforthedisplacedgroup.However,whenusingaedeffectmodel tocontrolindividualunobservablesmanycoefsbecomeinsigniAsthemainideaof Lindo'sresearchistoshowtheeffectsofincomeonfertility,hedoesnotlookattheeffectof femalejobdisplacement,whichwouldbeaffectedbysubstitutioneffectsaswell 4 . TherearesomeresearchstudyingthefertilityeffectsofjobdisplacementinEuropeusingin- dividualpaneldata.DelBonoetal.(2012)examinedtheeffectsofawoman'sownjoblossusing Austrianadministrativedatafrom1972to2002.Comparingthebirthratesofdisplacedwomen withthoseunaffectedbyjoblosses,theythatjobdisplacementreducesaveragefertilityby 5%to10%.Thestrongaverageresponseismainlyexplainedbythebehaviorofwhitecollar women.UsingFinnishdata,HuttunenandKellokumpu(2012)showsimilarnegativeeffectsof women'sjoblossonfertility,especiallyforhighly-educatedwomen.Foreveryonehundreddis- placedwomenthereareapproximatelyfourfewerchildrenborn.Theyalsofoundthatmalejob losshasnoimpactonlifetimefertility.Theseresearchersusehigh-qualityadminis- trativedatatoidentifytheeffectsofexogenousjobdisplacementonfertilityinEurope,butitis 4 Thechannelsthroughwhichfertilitycanbeaffectedbyemploymentarediscussedinmore detailsinSection2. 34 diftogeneralizetheirtotheU.S.,whichhasadifferentcultureandsociologyas wellaslabormarketpolicythanEurope.Inparticular,maternityleaveintheU.S.ismuchless generousthanthatinAustriaandFinland,andthechildcaresystemforchildrenundertwoyears oldislesswidespreadandlessdevelopedintheU.S.Withfewermaternityforemployed workers,theincomeeffectofdisplacementwillbesmaller.Withlimitedaccesstochildcare,the timecostofchildren-andthusthesubstitutioneffect-wouldbegreaterforwomenintheU.S. Combiningthesmallerincomeeffectswhichtendtobenegativeandthelargersubstitutioneffects whichareusuallypositive,wemayexpecttheeffectoffemalejobdisplacementonfertilitytobe morepositiveintheU.S.Totestthisconjectureandinvestigatetheeffectsofjobdisplacementin theU.S.,wehavetolookatthedataintheU.S. Inthispaper,inordertoexploretheeffectsofjoblossonfertility,Ifocusonexogenousjobloss generatedbyjobdisplacementforbothmenandwomenintheU.S.Inparticular,controllingthe individualunobservablesbyaedeffectmodel,Icomparefertilityforpeoplewithandwithout anexperienceofdisplacement,beforeandafterthedisplacement.DatafromtheNationalLongi- tudinalSurveyofYouth1979showsthat,intheimmediateyearsfollowingdisplacement,thereis nochangeinfertilityformenorwomen.Inthelateryears,however,maledisplacement hasanegativeeffectonfertility,whilefemaledisplacementhasheterogeneouseffectsforwomen withdifferentlevelsofeducation.Thepositiveeffectsareobservedforwomenwith nocollegeeducation,andnoeffectsareobservedforwomenwithatleastsomecollege education.Resultsofusingsomeotherincludingnarrofidisplacementfl criteriaandedeffectpropensityscorematchingtherobustnessoftheestimates. Thispapermakesseveralcontributionstotheliteratureonanalyzingtheeffectsofdisplacement onfertility.First,tomyknowledge,thispaperistheonetolookattheeffectsoffemale displacementonfertilityintheU.Swithindividualleveldata.UsingPSIDdata,Lindo(2010)and Amialchuk(2013)showedthatmen'sjobdisplacementhadnegativeeffectsonfertility,buttheydid notinvestigatethefertilityconsequencesofwomen'sdisplacement.Duetothedifferentsocially acceptedgenderexpectationsinraisingchildren,Ibelieveitisnecessarytoexploretheeffectsof 35 displacementforwomenseparately.Second,basedontheconceptualmodel,Itheimplications ofheterogeneouseffectsforwomenwithdifferentlevelsofeducation,andthentestand thoseimplicationsintheempiricalpart.Inaddition,Icheckthatmymainresultsarerobustto usingadisplacedgroupofindividualswholosttheirjobsduetoworkplaceclosings,andalso robusttoedeffectpropensityscorematching.Workplaceclosingsgeneratedisplacementsthat aremorelikelytobeexogenoustoindividualcharacteristics.Propensityscorematchinggenerates amoresimilarnon-displacedgroupforfurtheredeffectanalysis. Theremainderofthepaperislaidoutasfollows:Section2discussesthemechanismsthrough whichdisplacementsaffectfertility,includingbothreviewsoftherelatedliteratureandasimple conceptualmodelofanalyzingfertilitydecision.Section3describesthedataandcomparisons betweendisplacedandnon-displacedgroup.Section4introducestheestimationstrategy,andthe regressionresults.Section5andSection6presenttherobustnesscheckresultsusingnarrower ofdisplacementandedeffectpropensityscorematchingrespectively.Section7con- cludesthepaper. 2.2WhyWouldDisplacementAffectFertility 2.2.1RelatedLiterature Researchershaveacontinuinginterestintheproblemsofdisplacedworkers,andhaveproduceda substantialliteratureonthistopic.Theyhavefoundmanyadverseandlong-lastingconsequences ofjobdisplacement.Themostprominentconsequenceisthedecreasesinlifetime earnings(Ruhm(1991),Jacobsonetal.(1993),KletzerandFairlie(2003)andCouchandPlaczek (2010)).EvenforyoungworkersintheNLSYsamplewithlesscapital,Kletzerand Fairlie(2003)foundsizablelong-termearninglosses. Ontheotherhand,theexistingliteratureshowsthatchangesinearningsmightchangepeople's fertilitydecisions.HeckmanandWalker(1990)estimatetherelationshipbetweenearningsand fertilityusingretrospectivefertilitydatafromSweden.TheyarguethatbecausemostSwedish 36 earningsaresetbycollectivebargainingagreements,theyareexogenoustothefertilityprocess. Theysupportforapositiveeffectofmaleearningsonfertility,whiletheeffectoffemale earningsisfoundtobenegative.MerriganandPierre(1998)similarresultsusingthesame methodologywithdatafromCanada.Schultz(1985)usesaninstrumentalvariablestrategyto identifythecausalrelationship.Heusesworldpricesofgrain,whichisamalelabor-intensive product,astheinstrumentformalewages,andpricesofbutter,whichisafemalelabor-intensive product,astheinstrumentforfemalewagesinhisanalysisofcounty-levelfertilityratesinSweden from1860to1910.Heonequarterofthedeclineinfertilityduringthatperiodcanbe explainedbytheincreaseinthefemale-to-malewageratio.Thedoublingofrealmalewageshadno effectonlifetimefertility,butitdidinduceearliermarriageandexpeditedfertility.More recently,usingwidevariationinenergypricesinthe1970'sastheinstrument,Blacketal.(2013) detectedthepositiveeffectsofmen'sincomeoncompletedfertility.Tosumup,itisgenerally foundthatmaleearningshaveapositiveeffectonfertility,whilefemaleearningsreducefertility. Combiningthesubstantialearninglossescausedbydisplacementandtheeffectsofearnings onfertility,itisreasonabletopredictthatjobdisplacementwillhavesomeonfertil- itydecisions.Besideschangesinearnings,thereareadditionalmechanismsthroughwhichjob displacementmayaffectfertilitydecisions.Forexample,displacementisalsoshowntoreduce jobstability(Stevens(1997)),increasethehazardofdivorce(CharlesandStephens(2004),Elia- son(2012))andhavenegativeimpactsonhealth,education,andlabormarketoutcomesforthe childrenofthedisplacedworkers(StevensandSchaller(2011),Oreopoulosetal.(2008)).Inthe conceptualmodelbelow,focuswillcenterontheimpactofinducedchangeinearningsonly. 2.2.2ConceptualModel TheclassicmodelsofBecker(1960)andMincer(1963)pioneeredtheassociationoffertilitywith parent'swagesandhouseholdincome.Afterthatthemodelhasbeenexpandedgreatlybyother theorists(Willis(1973),Hotzetal.(1997)).DrawnfromJonesetal.(2010),astandardstatic modeloffertilitydecisionislikethis:Parentstrytomaximizetheirutilityfromconsumption 37 andquantityofchildren,subjecttobudgetandtimeconstraints.Childrengenerateutilityforthe parentswiththecostoftimeandmoney.Conventionally,womensubsumethetotaltimecostsof thechildren,andmenprovidesupport.Inthismodel,anincreaseinmen'swageswill onlyraisehouseholdincome.Forwomen,conversely,bothhouseholdincomeandthepriceof childrengoesupwithanincreaseoftheirwages,resultinginoffsettingincomeandsubstitution effectsonthedemandofchildren.Thiscouldproduceanambiguousneteffect.Therefore,even ifthemagnitudeofearninglosscausedbyjobdisplacementissimilarformenandwomen,the effectsonfertilitymightbedifferent. 5 Withtheintroductionofmarketchildcare,thefertilityeffectsforwomenwithdifferentearning levelswilldifferaswell.InspiredbySinghetal.(1986)andPerry(2003),Ianalyzefertility decisionsinahousehold(individual)productionmodel.Theindividualgeneratesutilityfrom consumption( C )andquantityofchildren( n ),soamother'sutilityfunctioncanbewrittenas U ( C ; n ) .Foreachchild, l unitsofchildcareisrequired,andfor n children,the n l unitsofchild carecanbeobtainedeitherfromhomeproduction l h ,orfrommarketpurchase l m .Thehome productionfunctionforchildcare l h = f ( K ) isassumedtobeincreasingandconcave,where K is thetimeinputforhomeproduction.Marketchildcarecanbepurchasedatanexogenousprice p l . Foreachmother,therearetwoconstraints.First,isthebudgetconstraint 6 ,andthesecondisthe timeconstraint.Insummary,with w denotingmarketwagefortheindividual,and L asthelabor 5 Theconcernthatemployerswillinvestlessonwomen'scapitalwithanticipations ofchild-birthinterruptionsimpliesthatearningslossfollowingjobdisplacementshouldbesmaller forwomen.However,empiricalstudiesdonotstronglysupportthisconjecture(Jacobsonetal. (1993),KletzerandFairlie(2003),CouchandPlaczek(2010)). 6 Husbandearnings y h isnotincluded,sinceIregardthatasexogenousunearnedincome,which willnotaffectthemainconclusion. 38 supplyinthemarket,themaximizationproblemcanbeexpressedas: max : U ( c ; n ) s : t : : C + p l l m = wL l m + l h = n l l h = f ( K ) K + L = T (2.1) Fourconstraintscanbecombinedasthefullincomeconstraint: C + p l n l = wT + p p = p l f ( K ) wK (2.2) Now,themaximizationproblemcanbesolvedintwosteps.First,choose K tomaximize p ( w = p l f 0 ( K ) forinnersolution);second,given p ,choose C and n tomaximizeutility.Inthis framework,achangeinthewomen'swageratemightleadtodifferenteffectsonthedemandfor children,dependingonthelevelofthewagerateandhow K isdetermined. First,if w issohighthat w > p l f 0 ( K ) atthecornersolution,marginalchangein w willnot changethelevelof K .Womenwillnotproduceanychildcareathomebothbeforeandafterthe change.Soadeclinein w willonlyreducetheincomelevels,andlowerincomewillleadtoa lowerdemandforchildren.FigureD.2depictsthiscasegraphically.InFigureD.2,thewagerate decreasesfrom w 0 to w 1 ,andthenumberofchildrendropsfrom n 0 to n 1 correspondingly. Second,if w issolowthat w < p l f 0 ( K ) atthecornersolution.Womenwillproducethe entiretyofchildcareathome,andusetheremainderofavailabletimetoworkonthemarket.In thiscase,adecreasein w willnotchangetheamountof l m ,whichisalwaysequaltozero,but K willbechanged,aswomenhavetotradeofftheandcostofhavingchildren.Now,the orderconditionforoptimal K is f 0 ( K )= U 0 c ( C ; n ) w l U 0 n ( C ; n ) .Declinein w willleadtodeclinein f 0 ( K ) , withconcavity, K willrise.Intuitively,alowerwageratereducestherelativecostforchildren,so thedemandforchildrenwillbeincreased. Third,if w = p l f 0 ( K ) wehaveaninnersolutionandonlypartofthechildcarewillbe producedathome.Nowifwageratedecreases, K willbeincreased,astherelativecostofhome 39 productionislower.However,smaller wT willexertanegativeeffectonthepurchaseof l m .Given thesetwofactors,itisunclearwhetherthetotaleffectonfertilitywillbepositiveornegative.For low-wagewomen,astheyarepurchasingverylittlechildcarefromthemarket,thelatterimpacts aresmaller,whiletheformereffectsarelarger.Theirtotaleffects,therefore,willmoreprobably bepositive.Forhighwagewomen,theoppositeshouldbetrue. Perry(2003)alsoarguesthatthehigh-wagewomenhaveastrongerincomeeffectwhilelow wagewomenhaveastrongersubstitutioneffect.Therefore,thedemandforchildrenwillbere- ducedforhigh-wagewomenbutincreasedforlow-wagewomenwhenthewageratedecreases. Usingindustryandlocationwagevariationastheinstrument,Perryfoundthata10%in- creaseinearningsreducestotalfertilitybyage35by.09childrenforlow-educationwomen,who mightfacealowerwageoffer.Inthecaseofhigh-educationwomen,meanwhile,a10%increase inearningswillincreasethetotalnumberofchildrentheyhavebyage35byabout.03children. Basedonmyconceptualmode,withthelargeearninglosscausedbydisplacement,wecanex- pecthigh-educatedwomenwhoexperienceajobdisplacementtodecreasetheirfertility,while low-educateddisplacedwomenwillincreasetheirfertilityafterexperiencingajobdisplacement. Inthesimplesettingabove,weuseastaticmodelassumingthattherelevantunitoftimefor maximizationistheindividuals'lifetime.Whilestaticmodelscansimplifytheanalysisoffertility decisions,dynamicmodelsareattractiveastheyemphasizetheinherentlysequentialnatureof fertilitydecisions.Agreatdealofliteraturesuggeststhatitisimportanttoinvestigatethetiming andspacingofbirthsoverthelifetime(WardandButz(1980),HotzandMiller(1993),Amialchuk (2013)). Inadynamicframework,individualsmaximizethediscountedsumofutilitybyjointlychoos- ingtheallocationoftime,consumption,andthetimingandnumberofchildren.Theoptimal choicefromsuchamodelgenerallyentailstheconsumptionsmoothingandearlybirths,dueto theincentivetoenjoytheutilityfromchildrenearlierwithalowerdiscount.Howdoearnings affectfertilityinadynamicmodel?Thisusuallydependsonwhethertheeffectistransitoryor permanent,andwhethertheindividualsarecredit-constrainedornot(seee.g.Hotz,Klerman,and 40 Willis,1993).Forourstudyofdisplacement,muchoftheliteraturehasshownthatthelossofearn- ingsispermanent,therefore,wecanexpectthelifetimefertilitytobechangedbytheincidenceof displacement.Forlower-educatedwomen,thestaticmodelsuggeststheymightincreasefertility afterbeingdisplaced,butifweconsiderthecreditconstrainttheymightfaceafterdisplacement, wewouldexpectthatincreaseinfertilitytooccurnotimmediatelyafterdisplacement,butlaterin life.Inaddition,itmighttaketimeforindividualstolearntheinformationaboutincomelossfrom ajobdisplacement,sothedisplacedindividualsmightadjusttheirfertilityinresponsetoincome lossseveralyearslaterafterthedisplacement. 2.3DataandDescriptiveStatistics Forthisstudy,IusepublicdatafromtheNationalLongitudinalSurveyofYouth1979(NLSY79), anationallyrepresentativesampleofyoungmenandwomenaged14to22wheninterviewed in1979.After1979,theywereinterviewedannuallyuntil1994,andthenbiennially.Irestrict thesampletoobservationstoperson-yearsbelowtheageof42afterwhichbirthisrare(Lindo, 2010).ThedatafromNLSY79iswellsuitedtothisstudyasitincludesbothdetailedlabormarket informationandchildbirthhistories.Forstudiesofjobdisplacement,NLSY79hasseveraldistinct advantagesoverotherwidely-useddatasets.First,incomparisontoCPSDisplacedWorkerSurvey (DWS),theNLSY79islongitudinalandincludesbothdisplacedandnon-displacedindividuals. Thispermitscomparisonsbetweenworkerswhodoanddonotsufferajobdisplacementboth beforeandafterthedisplacements.Second,comparedtothePanelStudyofIncomeDynamics (PSID),theNLSY79recordsthereasonsforjoblossinmoredetailsbydistinguishinglayoffs frombeing(PSIDputsthesecategoriestogetherasonechoice).Formyfertilityanalysis,the NLSY79hastwoparticularadvantages.Itsamplesbothmenandwomen(PSIDincludesonlythe headofthehousehold,whoaremostlymen),enablingmetotestthedifferenteffectsforthem,and theheterogeneouseffectsforwomenwithdifferenteducationallevels.Also,theNLSY79follows peoplefromaveryyoungage,whentheyjuststartedtoenterthejobmarketandhavechildren,so 41 wecanseeacompletehistoryofdisplacementandfertility. Eachyear,theNLSY79gathersdetailedemploymentinformation,includinginformationon uptovejobsheldduringtheinterviewperiod(approximatelyoneyear).Ifanindividualwas nolongerworkingattheirpreviouslyreportedjobandthereasonforthejobendingwasfilayofffl orfiplantclosurefl,thenIcategorizethisasajobdisplacement(Fortherobustnesscheck,Iuse narrowlydisplacementgroupwhichconsistsofpeoplewholosttheirjobsduetoplant closureonly).Foreachjob,thereisinformationtolinkacrossinterviews.Usingthisinformation, wecandetermineiftherespondentreportedbeingre-employedwithsameemployer.Iftherewas amatchofemployer,thereportedjoblosswillnotbecountedasadisplacement.Followingthe literature,tolimittheanalysistoworkerswithareasonablystrongattachmenttothelabormarket, anotherrestrictionforfidisplacementflisthatthelostjobshouldbeafulltimeone 7 ,thatis,the individualmusthaveworkedanaverageof25ormorehoursperweekwhenworkingatthejob. Inthisstudy,Ionlyconsidertheeffectsofjobdisplacementthatoccurredfrom1984to1994. Idonotincludedisplacementinformationfrom1979to1983becauseofchangesinthepossi- bleresponsesforthefireasonleftjobflquestion.Priorto1984,temporaryandpermanentlayoffs weregroupedtogether.Topel(1990)arguesthatthePSIDmightinaccuratelymeasuredisplace- mentssincethequestionfocusesonanindividual'slastjob.Ifarespondenthasheldandleft anotherjobafteraninitialdisplacementandbeforesurveyed,theindividualwillbecategorized asnotdisplaced.BecausetheNLSY79includesinformationforupto5jobs,thisproblemwill belesssevere.However,after1994,whenthesurveywasdonebiennially,thismightbecomea concern.Inaddition,therecallerrorswillrisewhenthefrequencyofsurveyingdrops 8 .There- fore,Ionlyincludedisplacementinformationuntil1994intheempiricalanalysis.(Ialsotried todropobservationswiththeirdisplacementsafter1994,aswellasusingdatafrom1984to 7 Lindo(2010)arguesthatthestrongattachmentprovidessomereasontothinkthatthesework- ershavefisomethingfltolosewithjobloss. 8 Forwomani'sjobj,thetimewhenwasthequestionofdisplacementansweredmightormight notbethetimeofdisplacement.Further,Jacobsonetal.(1993)suggestthatworkerstendtoreport remoteinstancesofdisplacement. 42 1992onlyasrobustnesschecks.ResultsarereportedinSection3.)Toreducethemeasurement errorsindisplacement,therespondentswhohadnotbeeninterviewedfortwoormoreconsecutive yearsfrom1984to1994arealsoexcluded.Iincludeonlytheobservedjobdisplacementfor eachindividual(ifoneexists)duringthesurveyperiod,andIincludeitonlyifitmeetsthework experiencerestriction.Additionaldisplacementsfortheseindividualsarenotincludedseparately, asIviewfuturedisplacementsasapotentialcostoftheinitialdisplacement(Stevens,1997). Atotalof7,659individualsmeetthescreeningcriteriaforthesample.Ofthose,2,685(35%) sufferedatleastonedisplacementfrom1984to1994. 9 FigureD.4depictsthedisplacementrate intheNLSY79samplebyyear.Noticehere,Iincludealldisplacementsnomatterwhetheritisthe displacementfortheindividualorasubsequentdisplacement.Thedashedlinerepresentsthe ofannualunemploymentrateintheU.S.Itshowsasimilartrendtothecalculateddisplace- mentrate. 10 FigureD.4suggeststhatdisplacementrateforoursampledropsfrom1980sto1990s. Twopossiblereasonscanleadtothisdeclinetrend,oneofwhichistheimprovementinthelabor marketintheU.S.overthatperiod.Increaseintheageofthefollowedindividualsinthesample mightalsocontributetothedecliningtrendofobserveddisplacementrate.FigureD.5showsthe displacementrateintheNLSY79samplebyage.Atrendofdecliningdisplacementratecanbe foundwhenagegettingold. TableC.1comparesthemeansofkeyvariablesforthedisplacedgroupandthenon-displaced group.Regardingtime-variantvariables,followingLindo(2010),Icalculatedthemeansofthe displacedgroupbyusingvaluesthreeormoreyearspriortothetimeofdisplacement.For thenon-displacedgroup,allperson-yearvaluesareused.Asthistreatmentreducesthe ageofthedisplacedgroup,Icalculatedthedifferenceinmeansafteradjustingtocontrolageand 9 TherateofjobdisplacementcalculatedishigherthaninKletzerandFairlie(2003)(24%). Partofthereasonforthedifferenceistheexclusionofnon-consecutivelyinterviewedindividuals inoursample.Iftheseindividualsareincluded,thetotaldisplacementpercentagewilldropto 30%.Iemailedauthorsfortheircode,buttheycannotprovideasitwaswrotelongtimeago,soI cannotoutwhatelsemightbethereasonforthedifference. 10 FigureH.1dividesthetotalsampleintothreeagecohorts.Amongthesecohortsasimilar trendcanbeseen. 43 yearedeffects.Itwouldbeidealifthepre-displacementcharacteristicsofbothgroupswere similar.Asinthatcase,theexogeneityofdisplacementswillbesupported.Thatisnotwhatwe here,however.TableC.1indicatesthatsomedifferencesinthemeancharacteristicsbetween thedisplacedandnon-displacedgroupsaree.g.thedisplacedgroupismorelikelyto beblack,Hispanicandlesseducated,andtheirfamilyincomeinthepreviousyearyearisabout6- 8%lowerthanthenever-displacedgroup.Thissuggeststhatthesetwogroupsmaybedifferentin termsoftheirunobservedcharacteristicsaswell.Iftheunobserveddifferencesarecorrelatedwith fertility,theestimatedeffectsofdisplacementwillbebiased.Thisfactunderscoresthepotential importanceofcontrollingforindividualunobservableswhenestimatingchangeinfertilitydueto jobdisplacement.Therefore,wewillfocusonedeffectmodelsinourregressionestimation, andtryvarioustoverifytheresults. Beforeestimatingtheeffectsofjobdisplacementonfertility,wegraphicallyrepresentthe dynamicoffertilityanditsassociationwithjobdisplacement.FigureD.6showsbirthrates amongmenintwogroups:thosesufferingatleastonejobdisplacementandnever-displaced. Fordisplacedmen,thex-axisdenotestimebeforeandafterjobdisplacement.Fornon-displaced men,theydonothaveayearofdisplacementreference,andthusIgeneratedarandomyear-of- displacementforeachofthem.Afterconductingtherandomizationwiththeprobabilityforeach yearbasedontheconditionaldistributionofoccurrenceratesofdisplacementforthedisplaced group,Igraphedbothgroups.ThefifakeeventanalysisflinFigureD.6impliesthatdisplacedmen havehigherfertilitybeforedisplacementcomparedtothenever-displacedmen.However,about5 yearsafterthefidisplacementfl,displacedmendisplaylowerfertility.FigureD.7showsthedynam- icsforwomenwhoaredisplacedatleastonceandneverdisplaced.Nodifferencescan bedetectedforthetwogroups,bothbeforeandafterthefidisplacementfl.Accordingtotheanalysis inSection3,womenwithdifferentlevelsofeducationmighthaveheterogeneouseffectsonfer- tility.Therefore,Itrytoinvestigatethedynamicsoffertilityforwomenwithandwithoutcollege educationseparatelyinFigureD.8andFigureD.9.Forwomenwithcollegeeducationwhonever experiencedajobdisplacement,fertilityishigherthanfortheirdisplacedcounterparts,bothbe- 44 foreandafterthefidisplacementfl.Forwomenwithoutcollegeeducationwhoareneverdisplaced, meanwhile,fertilitywashigherthantheirdisplacedcounterpartsbeforethefidisplacementfl,but thedifferencesgraduallydisappearafterthefidisplacementfl. 2.4RegressionResults 2.4.1FixedEffectsModel AlthoughFigures6-9areinformative,Iammoreinterestedinestimatingtheeffectsofdisplace- mentonfertilitythroughcontrollingindividual'sunobservedheterogeneities.Followingthelit- eratureaboutdisplacementeffects(Jacobsonetal.(1993),KletzerandFairlie(2003),Couchand Placzek(2010),etc.),especiallystudiesaboutthedisplacementeffectsonfertility(Lindo(2010), DelBonoetal.(2012),HuttunenandKellokumpu(2012)),Iusethefollowingedeffectlinear probabilityforregressionanalysis: Birth it = G D it + X 0 it D + a 1 + g t + f i + e it (2.3) where D it =( d it 2 ; d it 1 ; d it ; d it + 1 ; d it + 3 ; d it + 5 ; d it + 7 ; d it + 8 + ) . Birth it isanindicatorforwhether ornotindividual i hasanyadditionalchildreninyeart. g t areyearedeffectsthatcapturethe generaltimepatternoffertilityinthesociety. f i areindividualedeffectstocaptureindividual time-invariantunobservablesandotherheterogeneity. X it canincludeavectorofobserved,time- varyingindividualvariables,andherelimitstoagedummies. 11 D it isavectorofdummyvariables indicatingtheindividual'sdisplacementinafuture,currentorpreviousyear,whichcanhelpusto capturethetimingoftheeffects.Tobe D it includesindicatorsfortwoyearspriortodis- placement,oneyearpriortodisplacement,theyearofdisplacement,andindicatorsforsubsequent 11 Idonotincludewages,earningorotherlabormarketoutcomevariablesdirectlyinthemodel becausetheyareknowntobeendogenoustobirthtiming(Walker,2002).Marriagestatusisim- portanttofertilitydecisionandshouldbecontrolledifitisexogenous.However,asshownby CharlesandStephens(2004)andEliason(2012),jobdisplacementgenerallyincreasesthehazard ofdivorce,whichhasadetrimentaleffectonfertility.Therefore,includingmarriagevariablemay causebiasedestimationondisplacementimpact.Robustnesscheckusinganalternative tionincludingmarriagevariablesisdoneandreportedinTableH.2. 45 yearsfollowingadisplacement(oneyearafterthedisplacement,2-3yearsafterthedisplacement, 4-5yearsafterthedisplacement,6-7yearsafterthedisplacement,andeightormoreyearsafter thedisplacement).Theomittedindicatorsareforthreeormoreyearspriortodisplacementand forthenever-displaced.Asitusuallytakes9monthsfromaconceptiontoabirth,wecanregard thecoefof d it + 1 astheeffectsofdisplacementinyear t ontheconceptionsinyear t .Sim- ilarly,thecoefof d it + 3 representstheeffectsofdisplacementinyear t ontheconceptions inyear t + 1andyear t + 2.Becausewecontrolthetimetrend g t andindividualheterogeneity f i , thisframeworknowcompareschangesindisplacedworkers'fertilitytothoseofthenon-displaced worker.Thisisessentiallythesameasthefidifference-indifferencesfltechnique,whichusesacon- trolgrouptocapturethefertilitychangesthatwouldhaveoccurredintheabsenceofdisplacement. Basedontheassumptionthatwithoutdisplacement,changeinfertilityforthepeopleindisplaced groupwouldbethesameasthatforthepeopleinthenon-displacedgroup,nomatterhowworkers' permanentcharacteristicsarerelatedtotheirdisplacementstatus,theestimatesofthedisplacement effectsareconsistent. TableC.2istheimpactofdisplacementsfortheentiresample,menandwomenrespectively. Formen,theeffectsarenegativenegativeat10%levelfor6ormoreyearsfollowing thedisplacement).Theannualbirthratewillbedecreasedby1.5-1.6percentagepoints6years afterthedisplacement.UsingPSIDdataforhouseholdheadonly,Lindo(2010)foundahusband's displacementhasaverysmallpositiveeffectintheyearsimmediatelyfollowingthedisplacement, andalargernegativeeffectmanyyearslater.Themagnitudeofnegativeeffectafterfouryearsis about1.4-2.3percentagepoints,similartoourestimateshere.Further,theonlyeffect inhisestimatesistheonefor8+yearsafterthedisplacement.Forwomen,theeffectsreportedin Column3arepositive,butallcoefarenot TableC.3showsheterogeneouseffectsformen/womenwithdifferentlevelsofeducation. Columns1and3arepeoplewithhighschooleducationsorless.Columns2and4areforpeo- plewithatleastsomecollege.Formen,itseemstheeffectsofjobdisplacementonfertilityare morenegativeforlesseducatedpeopleexceptfortheeffectsduring6-7yearsafterthedisplace- 46 ment,yetthedifferencesbetweenmoreandlesseducatedmenarenotsubstantial.Forwomen, myestimatesimplythatonlylower-educatedwomenshowpositiveeffectsfouryears afterthedisplacement.Highlyeducatedwomenwillalsoincreasetheirfertilityiftheysufferajob displacement,yetthepositiveeffectsarenotForwomenwhodidnotgotocollege,if theyloseajob,theprobabilityofhavingadditionalchildrenforeachyearwillincreaseby2.2-2.3 percentagepointsfourormoreyearslater.ThisisdifferentfromtheinEuropebyboth DelBonoetal.(2012)andHuttunenandKellokumpu(2012).Theirstudiesthatjobdisplace- mentshavenegativeeffectsonfertilityforwomeninAustriaandFinland,andthenegativeeffects areconcentratedinhigheducated,white-collarwomen.Comparedtoourhere,itisim- pliedthatthetotaleffectsofjobdisplacementintheU.S.aremorepositive.Thiscanbecausedby eitherlargersubstitutioneffects(limitedaccesstothemarketchildcare)orsmallerincomeeffects (fewermaternityforemployedworkers). Totestthedifferentialeffectsofdisplacementforwomenwithdifferentlevelsofeducation, Iincludetheinteractiontermsofadummyvariableofcollegeeducationandindicatorsofyears afterthedisplacementintheregression.ThecoefoftheinteractionsarereportedinTable C.4.Itisfoundthattheeffectsofdisplacementaremorenegativeforcollegeeducatedwomen, andthedifferencesarestatisticallyfortheeffectsduring2-5yearsand8+yearsafter thedisplacement. Oneoftheadvantagesofhavingpaneldataisthatwecananalyzetheeffectsontotalfertil- ity.Follow(Lindo,2010),Icalculatethetotalfertilityeffectforthetreatedgroupinthreesteps. First,getthesumofthepredictedpost-displacementprobabilityofbirthsforthedisplacedpeople. Second,bysettingtheindicatorsofyearsafterdisplacementequaltozero,Igetthecounterfac- tualprobabilityofbirthforthedisplacedpeopleineachyearafterthedisplacementandaddthem together.Third,byaveragingthedifferencesbetweenthesetwosumsforeachpeopleinthedis- placedgroup,wecangettheaveragetotaleffectsonthetreated.Thesecondtothelastrowin TableC.2andTableC.3reportstheeffectsontotalfertility.Standarderrorsarecalculatedthrough bootstrap.Theresultssuggestahusband'sjobdisplacementhasatotalnegativeeffectsonfertility, 47 reducingthetotalnumberofchildrenby0.11 12 .Forwomenwithoutcollegeeducation,jobdis- placementwillleadtoa0.08increaseintotalnumberofchildren.Similarto(Lindo,2010),the estimatedtotalfertilityeffectsarenotstatistically 2.4.2TimeTrendModel Byintroducing f i ,edeffectmodelcangiveusconsistentestimateseveniftherearesometime- invariantdifferencesbetweenthedisplacedandnon-displacedgroup.Now,supposedisplacedand non-displacedworkersarealsodifferinginbirthtimingpattern.Forexample,displacedworkers systematicallyhavechildrenearlierthanthenon-displacedones.Inthiscase,inordertoget consistentestimates,wecanapplythefollowingwhichallowsheterogeneityintime trendbyintroducing w i t and h i t 2 : Birth it = G D it + X 0 it D + a 1 + g t + f i + w i t + h i t 2 + e it Iusethequadraticformheretoaccommodatetheinverseu-shapepatternweobservedinFigures6 through9.Toestimatethismodel,Iuseagenericquasi-differencetechnique(Wooldridge,2010). First,foreachindividual i ,regress Birth t , D t and X t on t and t 2 ,andgettheresidualsas ¨ Birth it , ¨ X it and ¨ D it .Then,wecanapplyOLStoregress ¨ Birth it on ¨ X it , ¨ D it andyeardummies.TableC.5 showstheregressionresults.Theestimatedeffectsbecomelargerwhencontrollingforworker- timepatterns 13 ,andthuswehavemorethatourpreviousestimatesarenot causedbythesystematicdifferencesinbirthtimingbetweendisplacedandnon-displacedworkers. TheresultsinTableC.5suggestthatahusband'sjobdisplacementwillreduceannualbirthrateby 2.4to3.8percentagepoints4yearsafterthedisplacement.Forwomenwithoutcollegeeducation, theirjobdisplacementwillleadto3-4.8percentagepointsincreaseinannualbirthratesoneyear afterthedisplacement. 12 (Lindo,2010)thathusbandjobdisplacementwillreducethetotalnumberofchildrenby 0.098. 13 Duetothesimilaritybetweenquasi-differenceandstdifference,wegetcloseresultsin TableC.5andTableH.1,whichreportsregressionresultsfordifferencemodel. 48 2.4.3RobustnessCheck Consideringthebinarynatureof Birth it ,Ialsotrytousenon-linearforregression analysis.Themodelcanbeexpressedas: Pr ( Birth it = 1 )= F ( G D it + X 0 it D + a 1 + g t + f i + e it ) (2.4) Unfortunately,becauseoftheincidentalparameterproblems,withoutfurtherassumptions,Ican- notgetaconsistentestimationforthismodel(Wooldridge,2010). 14 Therefore,theMundlak- Chamberlainapproach(Mundlak,1978)isintroducedhere,basedontheadditionalassumption that: f i = y + ¯ X i x + a i , a i j X i ˘ Normal ( 0 ; s 2 a ) . where ¯ X i isthemeanoftimevariantvariables X it .Thisallowscorrelationbetweenindividual time-invariantcharacteristicsandthemeansof X it .Thismethodologyhasbeenusedsuccessfully toestimateacorrelatedrandomeffectmodel(Wooldridge,2010).Now,theresultingmodelcanbe writtenas: Pr ( Birth it = 1 )= F ( Birth it = G D it + X 0 it D + a 1 + g t + y + ¯ x i x + u it ) (2.5) Iusetherandomeffectprobitmodeltoestimate(5).TableC.6reportstheaveragepartial effectsofdisplacement,whicharesimilartothecoefreportedinTableC.2andTableC.3. However,theeffectsofwomen'sjobdisplacementaresmallerandnotanymore.For men,Column1inTableC.6suggeststhattheprobabilityofhavingadditionalchildrenwillbe reducedby1.4to1.6percentagepointssixyearsafterahusband'sjobdisplacement.Forwomen withoutcollegeeducation,Column3inTableC.6impliesthat,iftheysufferajobdisplacement, theirbirthratewillincreasebyaround1percentagepointsannually4yearslater,yettheseeffects arenot 14 Fixedeffectlogitmodelcangenerateconsistentestimators,however,marginaleffectscannot beestimatedsincetheindividualedeffectsarenotactuallyestimated.AndtheSTATAprogram failstoconvergehere. 49 Inregressionsabove,Ionlyconsidertheeffectsofjobdisplacementthatoccurredfrom1984to 1994.Sincenon-displacedpeoplemayencountertheirdisplacementafter1994thatwillnotbe measuredinourestimation,theeffectsestimatedabovewillbeunderestimated.Inordertocheck themagnitudeofthepossiblebias,twoothersampleselectionsareimplementedasrobustness check.Oneistodropallnon-displacedpeoplewhoreportjobdisplacementsafter1994;theotheris touseperson-yearsfrom1984to1992only,asthoseobservationsincludecompleteinformationof bothbirthanddisplacement.ResultsarereportedinTableC.7andTableC.8separately.Estimated effectsinTableC.7arelargerthantheresultsreportedinTableC.2andTableC.3forbothmen andwomenwithoutcollege.Whenfocusingonperson-yearsfrom1984to1992,thesamplesize issubstantiallyreducedandleadtoimpreciseestimatesforwomen.Formen,theestimatedeffects aremuchlargernow,andnegativeeffectsareobservedforallperiodsincludingyearspriortothe displacement. 2.5PlantClosure Theaboveestimateswillbebiasedifselectivelylayoffemployeeswhoseperformancewas poorbeforethetimeofseparation,andatthesametimetheperformanceiscorrelatedwithindi- vidual'sfertilitypreferencesforthefuture.Onewaytosubstantiallyreducethiskindofselection biasofdisplacedworkersistorestrictanalysistoworkerswholosejobsasaresultofworkplace closings,asthistypeofjobdisplacementisregardedtobemoreexogenous(CouchandPlaczek (2010),Lindo(2010)andDelBonoetal.(2012)).TableC.9displaysthecomparisonsbetweenthe displacedandnon-displacedgroupswhenwerestrictthedisplacedgrouptopeoplewholosetheir jobsduetobusinessclosure.ComparedwithTableC.1,twogroupsaremoresimilar,especially forwomen.Moreimportantly,thereisnodifferencesinfertilitybetweentwogroups. TableC.10showstheregressionresultsbasedonthisnarrowlyfidisplacementfl.The estimatesherearesimilartothemainregressionresultsinTableC.2andTableC.3insignand magnitude,butarerarelystatisticallyasaresultofthesmallsamplesizeofthedisplaced 50 group.Fortheworkplace-closingtypeofjobdisplacement,itimpliesthatwomenwithnocollege educationwillincreasetheirfertility4yearsafterthedisplacement.Theeffectofmaleworkplace- closingjobdisplacementonfertility6yearsafterthedisplacementisnegative,yetthenegative effectsarenot CharlesandStephens(2004)thatthedivorcehazardincreasesafteraspouse'sjobdis- placement,butthatriseisfoundfordisplacementduetolayoffsonlybutnotclosure.Asour resultsimplythatbothtypesofjobdisplacementcanleadtotheobservedchangesinfertility,we believethatthechangeinstabilityofmarriageafterjobdisplacementisnotthemainreasonfor ourobservedchangesinfertilitybehavior. 2.6PropensityScoreEstimates RegressionresultsfromSection5generallyshowthatfertilitywillbedecreasedifahusband suffersajoblossnomatterhiseducationlevel.Incontrasttothis,womenwithlowereducationwill increasetheirfertilityfollowingajobdisplacement.Wedrawtheseconclusionsfrombothed effectandtimetrendmodel.Further,thisoverallpatternisalsorobusttoacorrelatedrandomeffect probitmodel,differentsampleselectionrules,aswellasusinganarrowofdisplacement thatonlyincludesworkplaceclosings.Therearestillsomeotherconcernsworthyofinvestigation, however.Asthereisanon-randomselectionofwhichestablishmentsaregoingoutofbusiness (DelBonoetal.,2012),workersmightbeselectedintotheclosingIfthisselectionis alsocorrelatedwiththepreferencesforfertility,thentheestimatedeffectsfromnarro displacementstillwillbeinconsistent.Therefore,weneedsomealternativewaystofurthercontrol theheterogeneityamongindividuals.Inordertodoso,Iintroducematchingestimatorstocheck therobustnessofmyFirst,basedonthepropensityscore,foreachdisplacedworker, wechooseanon-displacedindividualwhoresembleshim/her,andthenwecanusethosepairsto calculatethefertilitychanges.CouchandPlaczek(2010)useasimilarmethodtocalculatethe earningslossofdisplacementandgetsimilarresultstotheiredeffectmodels. 51 Theideainusingamatchingestimatoristoaverysimilarcontrolindividualforeach treatedindividual.Toreducethedimensionalityofthisproblem,RosenbaumandRubin(1983) suggestthatmatchescanbebasedonthepredictedprobabilityoftheevent.Iincludeageand racedummies,aswellaseducation,maritalstatus,weeksworkedandnumberofchildren2years beforedisplacementasthepredictorsforfidisplacementfl.Fortime-variantdummies,Couchand Placzek(2010)usethelevelattheyearforallobservationsinestimatingpropensityscores, astheyarguethattheyearinformationprovidessubstantialexplainingpower.Thismethod isnotappropriatehere,asoursamplecommencesatarelativelyyoungage,of14-22in1979. Thesampledoesnothavemuchinvariationeducation,maritalstatus,weeksworkedandnumber ofchildrenin1979,andthuscannotprovideenoughinformationforpredictingtheprobabilityof displacement.Therefore,Ifollowthe inflated methodinLechner(1999)togetthepropensity scores.The inflated methodexpandsthenon-displacedgroupbytreatingeachnon- displacedindividualineachyearduring1984to1994asaseparateobservationwiththerespective displacementyear.Nowallnon-displacedindividualshaveadisplacementyear,sowecaneasily usetime-variantvariablestocalculatethepropensityscores.FigureD.10isthecalculatedpropen- sitiesofdisplacementforwomendisplacedin1979andtheir counterparts . 15 Thedistributionof probabilitiesisquitebalanced,sowecanproceedwithfurtheranalysis. Afterretrievingthepropensityscores,Icanthebestmatch(withthesamepropensityscore ortheclosetpropensityscore)foreachdisplacedpersonandcalculatethedifferenceinthechange infertilityforeachpair.Inthiscalculation,thedisplacedpersonandtheirmatchedpaireach havetheirowndemeanedbirthrate.Then,Icancomparethedifferenceforeachpair,andaverage theDIDacrossthesamplefordisplacedpeople.ThisisreferredtointhetextasFixedEffects PropensityScoreMatchingEstimator(FEPSME).TheformulaforFEPSMEcanbeexpressedas 15 Theestimationisbasedonalogitmodel. 52 follows: FEPSME = E f [ E [ Y 1 it j D i = 1 ; p ( x i )] E [ ¯ Y 1 i j D i = 1 ; p ( x i )]] [ E [ Y 0 it j D i = 0 ; p ( x i )] E [ ¯ Y 0 i j D i = 0 ; p ( x i )]] j D i = 1 g Whilematchinghelpustoeliminatetheleveldifferences,demeaningcanfurtherhelptoremove anysystematictrendbiasremainingacrossdisplacedandnon-displacedindividuals. TableC.11showsthepreliminaryresultsfortheestimationofFEPSME.Thestandarderrors areobtainedthroughbootstrap.ThesmallereffectsonfertilityreportedinTableC.11areconsistent withtheideathatthosewhoexperiencejobdisplacementsaresystematicallyselected.Menarestill showntobelesslikelytohaveanyadditionalchildrenafterjobdisplacement,buttheseeffectsare onlyduring6-7yearsafterdisplacement.Less-educatedwomenreduce theirprobabilityofbirth4yearsfollowingthedisplacement,yettheeffectswehereis smallerthanwefoundinSection4. 2.7Conclusion Theaimofthispaperistoexplorehowfertilitydecisionsareaffectedbyjobdisplacementinthe U.S.byusingmicrodata.Historicalmacrodatashowsanegativeassociationbetweenunemploy- mentandbirthrate,butthereisalackofcasualanalysisatmicrolevel,especiallyfortheeffects ofwomen'sjoblossintheU.S. Themajorempiricalresultsofthispaperareasfollows:Displacementofmenwillleadto reductioninfertilityinthefollowingyears,whiletheeffectsofdisplacementforwomendepends onthewomen'seducationlevels.Forwomenwithnocollegeeducation,theirfertilitywillincrease afterdisplacement;forwomenwithcollegeeducation,thereisnoeffectsonfertility afterthejobdisplacement.Byintroducingmarketchildcareandhouseholdproductionmodel,the conceptualmodelcanhelptoexplaintheheterogeneouseffectsforwomen. Theempiricalareobtainedthroughaedeffectmodelandatimetrendmodelwhich controltheindividualtime-invariantheterogeneityandtimetrendheterogeneity,andarerobust 53 toseveraldifferentincludingacorrelatedrandomeffectmodel,differentsample selectionrules,aedeffectpropensityscorematchingmodelandanarrowofjob displacement.Onethingtobenoticedisthatjobdisplacement(orjoblossduetoclosure)is assumedtobeexogenousinourestimation.Torelaxthisassumption,wecantrytousestateby yearunemploymentrateasinstrumentsanddo2SLSregressionsinfutureresearch. 16 Ourresultssuggestthattheshort-run(1-3yearsafterthedisplacement)effectsarequitesmall forbothmenandwomen.Therefore,joblossitselfcannotfullyexplainthepro-cyclicaltrend infertilitythatisobservedwithmacrodata.Someothermechanismsneedtobeexploredto explainthecausalrelationshipsbehindthattrend.Forinstance,Adsera(2011)arguethatthe feelingofeconomicsinstabilityandtheriskofjoblossmighthavestrongerimpactsoncurrent fertilitydecisions. 16 ThepublicNLSY79dataIuseinthispaperdoesnotprovidestateinformation. 54 CHAPTER3 THEMORETHEMERRIER?THEEFFECTOFFAMILYSIZEONPARENT'S MENTALHEALTHINRURALCHINA 3.1Introduction ThereisanoldChinesesaying,fiMorechildren,moreblessingsfl.Thisrepresentsthethoughts ofmostChinesepeoplethatmorechildrencangeneratemorehappinessfortheparents.Infact, therearenotmanyempiricalstudiesofwhetherparenthoodornumberofchildrenhaveeffectson parentalwell-being,especiallyonmentalhealth,inChinaorworldwide. Mentalhealthproblemarecostlytosocietybothintermsofdirectspendingontreatmentand throughindirectcostssuchasthelossofproductivity(Pengetal.,2013).AccordingtotheWHO, depressioncanresultindisability,prematuredeath,andseveresufferingofthoseaffectedandtheir families(Demyttenaereetal.,2004).Huetal.(2007)showthatthetotalannualcostsofdepres- sioninChinaisatleastUS$6.26billion(at2002prices).Inspiteofthehugecosts,mentalhealth problemswereneglectedforalongtimeinChina.In2007,about130millionpeopleinChinahad mentalillnesses,butamongthemonlyabout20%arediagnosedandtreated(Luetal.,2009).And thementalhealthprobleminChinaisgettingmorewidespreadnow.Datafromthe2011China HealthandRetirementLongitudinalStudyshowsthatover40%ofpeopleage45andabove,or about140million,showobvioussymptomsofdepressionin2011.Comparingtotheestimated totalnumberofpeoplewithmentalillnessin2007,whichisaround130millionforpeopleatall ages,weseeahugeincrease.Iftheoldsayingistrue,thedecreasingfertilityduetotheOne-Child Policymaycontributetotheprevalenceofdepressiontosomeextentbecausethesepeopledonot haveasmanychildrentoblessthemastheyage.However,weneedempiricalevidencesto theeffectsoffertilityonparent'smentalhealth. Howwouldadditionalchildrenaffectthementalhealthofparentsinthelongrun?Theunderlying 55 mechanismsarecomplex,andatleastthreedifferentchannelshavebeensuggestedbythelitera- ture.Thechanelisthesupporteffects,whichincludeeffectsofbothemotionalandphysical supportfromadultchildren.Deanetal.(1990)arguethatexpressivesupportfromone'sspouse andfriendscanreducedepression.Similarly,havingchildrencangenerateasenseofgratitudeand feelingsofmeaninginlife(EvensonandSimon,2005),whichreducestheprobabilityofexperi- encingdepressivesymptoms(BuberandEngelhardt,2008).InChina,havingmorechildrenhas beenfoundtobeassociatedwithbettersupportofagingparentsreceive(PeiandPillai,1999;Zim- merandKwong,2003),meanwhile,receivingmoreinstrumentalandsupportwilllead tobettermentalhealth(CongandSilverstein,2008).Thesecondchannelisthebudgeteffects. Additionalchildrenbringbothdirectcosts(theconsumptionoftheadditionalchildren)andoppor- tunitycosts(thereductionofparent'searningpotential)tothefamily.UmbersonandGove(1989) indicatethat,duetotheeconomicscostsandbindingconstraints,additionalchildrencanmake theirparentsbevulnerabletomentaldiseases.Opportunitycostsarisesincehavingmorechildren mayincreasethetimespentonchildcareanddecreasematernallaborsupply.Sinceemployment historycandirectlyaffectone'shealth(GoveandGeerken,1977),theeffectoffertilityonmaternal healthcouldworkthroughthechanneloflaborsupply.Thethirdchannelisthebiologicaleffects. Childbearingandnursingcanhavebothnegativeandpositiveeffectsonamother'shealth,which willconsequentlyaffectthemother'smentalhealth(Kendigetal.,2007;Hurtetal.,2006).As thesethreeeffectshavedifferentsigns,thetotaleffectoffertilityonparent'smentalhealthisan empiricalquestion. Theempiricalofthecausaleffectoffertilityonparent'smentalhealthiscompli- catedbytheendogeneityprobleminvolvedwiththefertilitydecision.Forinstance,peoplewith poormentalhealthmayitdifforthemtokeepastablemarriageandtohavemorechildren (BuberandEngelhardt,2008).Inaddition,peoplewithdifferentlevelsofmentalhealthmayhave differentpreferencesregardingfertility.Aseachindividualchoosestheiroptimalleveloffertility, thenumberofchildrenmightbedeterminedbythementalconditionoftheparentratherthanthe otherwayaround(KrukandReinhold,2014).Withtheseendogeneityproblems,theOLSestima- 56 torforcoefoffinumberofchildrenflwillnotbeaconsistentestimatorofcausaleffects. Researchontheeconomicsofhappinessisalwaysinterestedintheeffectsofchildrenonparental happiness.Mostofstudiesinthisstrandtypicallyshownegativeornulleffectsofnumberofchil- drenonlifesatisfactionand/orsubjectivewell-being(DiTellaetal.,2003;Alesinaetal.,2004; wer,2009;Gilbert,2009).Basedontheseeconomistsclaimthathavingmore childrendoesnotmakeushappier(Angeles,2010). Comparingtotheeffectsonhappiness,existingliteratureabouttheeffectoffertilityonmental healtharerelativelythin,andmostofthemareprovidedbyresearchersintheofpublichealth, psychologyanddemographyratherthaneconomics.Theiranalysesusuallyignoretheendogene- ityoffertilityandgeneratesurprisinglyinconsistentresults.Usingdatafromthe1988National SurveyofFamiliesandHouseholds,Koropeckyj-Cox(1998)foundintheU.S.,childlesswomen suffergreaterratesofdepressioninmiddleandoldage.Ontheotherhand,usingdatafromthe U.S.,GoveandGeerken(1977)andBurton(1998)bothrecordthenegativeassociationbetween havingchildrenandmentalhealthintheU.S. 1 Usingmulti-birthandsexcompositionofchildren asinstrumentsforfertility,KrukandReinhold(2014)identifythenegativeeffectsofnumberof biologicalchildrenonmentalhealthformothersinEurope.Therearealsoanumberofstudies suggestingthattheeffectsofnumberofchildrenonparents'mentalhealthare(Buber andEngelhardt,2008;MirowskyandRoss,2002).Forinstance,Hank(2010)statesnodifferences inmentalhealthamongmiddle-agedpeoplewithvariousnumbersofchildreninGermany,and KrukandReinhold(2014)alsonoeffectsonfather'smentalhealth.Tothebest ofmyknowledge,KrukandReinhold(2014)istheonlystudyaccountingforendogeneitywhen lookingatthementalhealtheffectsoffertility.Theinconsistentinexistingstudiesmay inpartbeduetodifferencesincharacteristicsofstudygroupandvariationininstitutionalcontext. Moreimportantly,withoutcarefullydealingwiththeendogeneityproblem,differencesinthese- lectionofcontrolvariableswillleadtodifferentresultsaswell. 1 GoveandGeerken(1977)analyzeddatafromasurveyconductedinChicago,whileBurton (1998)useanationalprobabilitysample. 57 TheeffectsoffertilityonmentalhealthinChinahavenotbeenfullyexplored.Silversteinetal. (2006)thatfewerchildrenisassociatedwithmoredepressivesymptomsinChina.Onthe otherhand,CongandSilverstein(2008)reportnoeffectsoffinumberofadultchildrenfl onparents'depressionsymptomsinChinathroughtheirclusteredregressionanalysis.Bothtreat fertilityasanexogenousvariable.Inthispaper,Iwillfocusontheeffectsofadditionalchildrenon mentalhealthforpeopleage45andaboveinruralChina.Asfertilityisendogenouslydetermined, IusethevariationintheimplementationoftheOne-ChildPolicytoconstructexogenousvariation infamilysizeanddo2SLS. Inordertocurbtherapidpopulationgrowth,theChinesegovernmentbegantoimplementthe One-ChildPolicy(OCP)in1979.Underthispolicy,amarriedcouplecanonlyhaveonechildin mostareas.AftertheimplementationoftheOCP,femaleinfanticide,forcedabortion,andforced sterilizationemergedinsomeplaces.Topreventtheseextremecases,19provincesadoptedthe fi1-boy-2-girlflrulein1984,whichmeansruralcouplesinthese19provincescanhaveasecond childifthechildisagirl(Qian,2009). Inthispaper,IuseexogenousvariationsinfertilitygeneratedbythevariationintheOne-Child PolicyinChina.More,Iusethedifferencesbetweencoupleswithgirlsand boys.Basedondatafromthe2011ChinaHealthandRetirementLongitudinalStudy wave),resultsshowthat,formothersage45andaboveinruralChina,havingmorechildren hasanegativeeffectontheirmentalhealth.Theeffectonfather'smentalhealthisalsonegative, but(Theeffectsarenotstatisticallydifferentformenandwomen.)Byinvestigating theheterogeneouseffectsforpeoplewithdifferentlevelsofeducation,Ialsothatthenegative effectsoffertilityarestrongerformotherwithmoreeducation.Thisisconsistentwiththe heterogeneouseffectsfoundinthestchapterofthedissertation,whichindicatesthenegative effectsonlaborsupplyformotherswithmoreeducationislarger. Afteridentifyingtheeffectsonmentalhealth,Ilookatonepossiblepathwayofgeneratingsuch effects.WuandLi(2012)showthatinChina,havingmorechildrenwoulddecreasetheresources allocatedtomothersandthusaffectstheirhealthoutcomes,andpoorphysicalhealthmayleadto 58 depressionsymptoms(Berkmanetal.,1986).Twomethodsareappliedtotesttheroleofphysical health.First,IuseSelf-ReportedHealth(SRH)astheindicatorofphysicalhealthandapplythe sameinstrumentsinthe2SLS.Second,SRHisaddedtotheoriginal2SLSonmentalhealthas controlvariables.Bothstrategiesdonotprovidesalientevidenceonthepathwaythroughphysical health. Inthispaper,IusescoresonCES-Dtomeasureanindividual'smentalhealth.Asdifferentpeople mayhavedifferentscalesofdepressivefeelings,theymaygivedifferentanswersforCES-Deven underthesamementalhealthstatus.TocorrectthepossiblebiasinreportedCES-D,Itrytodetect systematicscalebiasusingvignettesquestionsaskedinCHARLS. Theremainderofthepaperislaidoutasfollows.Section2providesbackgroundinformationon China'sOne-ChildPolicyandintroducestheestimationstrategy.Section3describesthedataand summarystatistics.Section4showsthemainregressionresultsincludinganalysisofheterogenous effectsforindividualswithdifferentlevelsofeducation.Section5discussesthepossiblereasons fortheseeffectsandteststheroleofphysicalhealthusingself-reportedhealth.Section6presents arobustnesscheckonthemeasurementbiasonmentalhealth.Section7concludes. 3.2TheOne-ChildPolicyinChinaandEstimationStrategy Duetoveryhighfertilityrates,thepopulationgrowthrateinChinareached27 : 5 h peryearduring 1962-1970,andthetotalpopulationwas816millionin1970(Yang,2004).Toalleviatesocial, economic,andenvironmentalproblemscausedbythehugepopulation,theChinesegovernment begantocurbpopulationgrowthasearlyas1972.ThepolicywassummarizedasfiLater(late marriageandchildbearing),Longer(birthspacingshouldbeatleastthreeyears),andFewer(two childrenshouldbeenough)fl(Qian,2009).Implementationinthatperiodreliedprimarilyonpro- paganda,persuasion,andsocialpressure(McElroyandYang,2000). Andlateron,in1979,Chinabegantoimplementamorerestrictivepolicy,thefiOne-ChildPol- icyfl(OCP).Underthispolicy,amarriedcouplecanonlyhaveonechildinmostareas,exceptfor 59 coupleslivingintheruralareainveprovinces(Hainan,Yunan,Qinghai,Ningxia,Xinjiang) 2 , whoareallowedtohavetwochildren(Peng,1996).Inpractice,implementationofthispolicyin someregionsbeganasearlyas1978,andtheenforcementbecamenationallytightenedin1980. InareassubjecttotheOCP,asecondbirthwasonlypermittedifonechildwouldcauseahouse- holdfirealdiftiesfl,e.g.,verybadhealthconditionofthechild.Coupleswhohadan abovequotabirthwithoutpermissionwouldbeheavily 3 .Localcadresweregiveneconomic andpromotionincentivestoimplementthepolicy.Intheearly1980s,fipartsofthecountrywere sweptbycampaignsofforcedabortionandsterilizationandreportsoffemaleinfanticidebecame widespreadfl(Greenhalgh,1986). Topreventfemaleinfanticide,forcedabortionandforcedsterilization,andtobetteraddressregion- conditions,theCentralPartyCommitteeissuedfiDocument7flinApril,1984.fiDocument 7flallowsregionalvariationinfamilyplanningpolices.Themainrelaxationpolicyfollowing fiDocument7flisthefi1-boy-2-girlflrulein19provinces,whichallowsruralcouplesinthese19 provincestohaveasecondchildifthebornisagirl(Qian,2009).ButaccordingtoWhite (1991),thesekindofpermissionsbegantobeissuedasearlyas1982.Thedifferenttreatmentof coupleswithgirlsversuscoupleswithboysallowsustoconstructexogenous variationinfertilitygeneratedbytheOne-ChildPolicy. Followingtheliterature,themainregressionmodelweareinterestedincanbewrittenas: CESD ict = b kids 2 ict + X 0 ict d + a 1 + g t + y c + e ict (3.1) where CESD ict ismentalhealthindicatorforwoman i incounty c ,agecohort t 4 . kids 2 ict isa dummyvariablethatequalsto1iftheindividualhastwoormorechildren.Insome wealsousenumberofchildren( nkids ict )tomeasurefertility. X ict isavectorofindividual i 's characteristics,includinggender 5 ,age,agewhengivingthebirth,genderofchild,ed- 2 TherearenorestrictionsonnumberofchildrenforruralcouplesinTibet. 3 Therearelocalvariationsin(WeiandZhang,2011). 4 Themeasurementof CESD ict willbeexplainedindetailinthenextsection. 5 Toexploredifferencesintheeffectsbygender,weconductanalysesforwholesample,aswell asformenandwomenseparately. 60 ucationlevels,numberofsiblingsandself-reportedhealthstatusduringchildhood; g t istheage cohortedeffect,and y c isthecountyedeffect.Sincementalhealthmayaffectindividual's fertilitydecision,andpeoplewithpoormentalhealthmayhavedifinmaintainingastable marriageandhavingmorechildren, cov ( kids 2 ict ; e ict ) 6 = 0.Therefore,theOLSestimatorof b is notconsistent.Toaddresstheendogeneityproblem,inthispaper,Iuseasetofdifferences-in- differences(DID)estimatestoconstructexogenouschangeinfertility,andthenusethesechanges asinstrumentsto kids 2 ict ( nkids ict )inEquation(1).TheDIDestimateswillexploitdifferencesin familysizebetweenpeoplewithgirlsandboys,beforeandaftertheOCP.The detailedexplanationofthisestimationstrategyisasfollows. Supposewehavefourcouples.BothCouple1andCouple2havetheirbirthsintheyear1984. Couple1hasagirl,andCouple2hasaboy.BecauseoftheamendedOne-ChildPolicy,Couple1 canhaveasecondchild,whileCouple2wasnotallowedto.In2011,wemayobservethesetwo coupleshavedifferentmentalhealthlevels.However,weshouldnotattributeallthedifferencesto thevariationinthenumberofchildren,asthegenderofthebirthmightdirectlyaffectpeople's mentalhealth(Kohleretal.,2005).However,wecanremovethisgenderdifferencebyusingthe othertwocouples.BothCouple3andCouple4havetheirbirthsin1974,Couple3hasagirl andCouple4hasaboy.AstherewasnoOne-ChildPolicyinimplementationuntil1979,both couplescanhaveasecondchildiftheywantto.Wealsohavetheirmentalhealthstatusin2011. ThedifferencesinmentalhealthbetweenCouple3andCouple4nowcanbeconsideredasthefiin- trinsicdifferencesbetweenpeoplewithgirlsandboysfl.Nowifweassumethe fiintrinsicdifferencesbetweenparentswithgirlsandboysflarethesamefortwo coupleswithbirthsin1974andtwocoupleswithbirthsin1984,thenwecanuseadiffer- encesindifferencesmethodtoremovethefiintrinsicdifferencesflandgettheremainingexogenous variationinfertilitygeneratedbytheOne-ChildPolicy.Thisestimationwillworkaslongaswe havethesefourcouples,butintherealdataset,wehavethousandsofcouples,withbirthseither beforeoraftertheimplementationofthepolicy.Thelargesamplewillmakeourestimationmore convincingandmoreprecise.TheDIDmethodcanbeexpressedas ( First - BornGirl ; After 61 First - BornBoy ; After ) ( First - BornGirl ; Before First - BornBoy ; Before ) . After represents theyoungcohort,whoisaffectedbytheOne-ChildPolicy,while Before representstheoldcohort whoisnotaffected.Consideringtheendogeneityprobleminvolvedwiththetimingofbirth,I decidedtouseagecohortsratherthantheyearofthebirthinregressions.Thestagefor kids 2canbeexpressedasequation(2),theinteractiontermsofwhetherisagirlandage cohortsareourinstruments. kids 2 ict = 44 å l = 22 ( First - BornGirl ict d l ) f l + X 0 ict m + a 3 + d t + p c + v ict (3.2) WhenusingEquation(2)asstageregressionwithrespecttoeither kids 2or nkids ,fairlysmall F-statisticsarereported,whichraisetheconcernofweakinstruments.(Detailsarediscussedin Section4.)Tosolvethisproblem,accordingtothecutoffageobservedfromregressionequation (2)aswellasTableE.8inthechapterofthedissertation,Idivideallparentsintotwoage groups:parentsage62andaboveasthe Before group,andparentsbelow62asthe After group. SincetheOCPwasnationallyimplementedin1980,andmostwomenwithtwoormorechildren completedtheirsecondbirthatorbeforeage30,only After groupwhowereatorunderage30in 1980wouldberestrictedbytheOCPandthusmostlikelytochangetheirfamilysizebecauseofthe relaxationoftheOCP.Inlinewiththisargument,Ifoundonlyinteractionsforcohortsbornafter 1950arepositiveinthestageinchapter1.Asaresult,Iuseage62asthecutoffageandthe stagefor kids 2iswrittenasequation(3).Now,theinteractiontermoftwodummyvariables, whetherisagirlandwhetherinthe After group(Age < 62)isthesingleinstrument. kids 2 ict = First - BornGirl ict After ict + X 0 ict m + a 3 + d t + p c + v ict (3.3) Fortheexclusionrestrictiontobetrue,weneedtoassumethatwithouttheOCP,thedifferencein mentalhealthbetweenparentswithgirlsandboyswouldbethesameinboth agecohorts. Withitsrelaxationofthehukou(householdregistration)systemandotherrestrictiveregulations,as wellasitsrapideconomicdevelopment,Chinahasbeenexperiencingahugescaleoflabormigra- tionsince1990s.Accordingtotherecentpopulationcensus,morethan261millionruralresidents 62 inChinalivedinplacesotherthantheirbirthplacesin2010(NBSC2012).Duetothehighmobil- ityofrecentyears,basedongeographicalvariationsintheOCPimplementationhas asmallpower.Thatis,whenItriedtorunstageregressionoffertilityforOne-boy-two-girl provinces,One-childprovincesandTwo-childrenprovincesseparately,thecoefoninterac- tionterm( First - BornGirl ict After ict )areallpositive.Therefore,Ididnotdistinguishrespondents fromdifferentprovincesandfocusedonagedummiesandgenderofonlytoconstruct exogenousvariationinfertility. 3.3DataandSummaryStatistics Thedatausedinthispapercomefromthe2011ChinaHealthandRetirementLongitudinalStudy (CHARLS),whichisapartofasetoflongitudinalagingsurveysincludingtheHealthandRe- tirementStudy(HRS)intheUnitedStatesandsimilarsurveysin20othercountries.CHARLS isanationalrepresentativedatasetoftheresidentsinChinaage45andabove,withnoupper agelimit(Zhaoetal.,2012).150countieswererandomlychosenfromeightgeographicregions acrossChina.CHARLScontainsawiderangeofinformationondemographics,familystruc- ture/transfer,healthstatusandfunctioning,etc.DetailedinformationofCHARLScanbefound at http://charls.ccer.edu.cn/en .Inthispaper,wewillonlyincludeindividualswithagricultural householdregistrationandresidingincountryside.Thesampleisfurtherrestrictedtopeoplewith birthaftertheageof15.Withtheserestrictions,weobtainasampleof9,657individualsfrom 28provinces,amongthem,4,517(46.8%)aremen 6 . Inthispaper,thementalhealthofparentsismeasuredbydepressivesymptomsbasedonaChinese versionof10itemCES-D(CenterforEpidemiologicStudiesDepressionScale).Afulllistofthe 10-itemCES-DisprovidedinTableI.1.The20itemCES-Disoneofthemostcommonscreen- ingtestsforidentifyingdepressivesymptomsinthegeneralpopulation.The10-itemscaleisthe shorterversion,whichalsoprovidesaself-reportedmeasureofanindividual'sdepressivefeelings 6 Whenusing kids 2tomeasurefertility,werestrictthesampletoparentswithatleastonechild. 63 andbehaviorsinthepastweek.Itisdesignedforstudiesinvestigatingtherelationshipbetween depressionandothervariables(Kohoutetal.,1993),anditsreliabilityandvalidityhavebeencon- (Boey,1999).Thecriterionfortheassessmentofmentalhealthisthesumofindividual symptoms 7 .TheCES-Disacontinuousmeasureofdepressivesymptoms,withscorerangingfrom 0to30.Higherscoresindicatinghigherlevelsofdepression.ThemeanvalueofCES-Dinthe wholesampleis8.87,withameanof7.76formenand9.85forwomen.Someliteratureregards10 inCES-Dasthethresholdvaluefordepression(Irwinetal.,1999).Ifwefollowthiscriteria,then 39.8%ofpeopleinoursamplehavedepression.FigureF.1plotstheaverageCES-Dscoresagainst thenumberofchildren,suggestinganonlineareffectoffertility.Comparingtochildlesspeople, parentshavebettermentalhealthwhentheyhavenomorethan4children.Ontheotherhand, conditionalonthepresenceofchildren,additionalchildrenwillraiseCES-Dscoreingeneral. TableE.1givesthesummarystatisticsofCES-D,numberofchildren,aswellasothercontrolvari- ablesthatarestandardintheliteratureexaminingthedeterminantsofmentalhealth:age,gender, ageatbirth,genderofthebirth,educationlevel,numberofsiblings,andself-reported healthstatusduringchildhood.Fornumberofchildren,Ionlyincludebiologicalchildren,andI includebothchildrenaliveandthosealreadydeceased(KrukandReinhold,2014).Theaverage numberofchildrenis2.96inoursample,andover88.6%oftherespondentshavemorethanone child.Theaveragenumberofsiblingsthattherespondentshaveis3.9,whichimplies4.9isthe averagenumberofchildrenoftheirparents.Comparingtotheaveragenumberofchildrenthat theythemselveshave,weseeshrinkageinfamilysizeoverageneration.Theaverageageofthe respondentsis58.6,andtheaverageageatbirthis24.Thisindicatesthatmostofthepeople inmysamplehadchildrenseveraldecadesagoandthattheresultsarethelong-termconsequences offertility.Onaverage,peopleinmysamplehave4.7yearsofeducation,onlyabouthalfofthem havecompletedprimaryschool.74.5%ofthepeoplereporttheyhadgood,verygoodorexcellent healthwhentheywereyoung. 7 CES-Dscoreequalstothesumofeightfinegativeflindicatorsplustheabsenceoftwofiposi- tiveflindicators.DetailedformulaisreportedinTableI.1 64 TableE.2showsthebasicDIDestimatesofprobabilityofhavingtwoormorechildren,number ofchildrenandmentalhealth,basedongenderofThefiYoungCohortflinTableE.2in- cludesindividualsborninorafter1950.Considering1980istheyearthattheOCPwasnationally implemented,andmostpeoplewithtwoormorechildrengavebirthtotheirsecondchildbefore theageof30,itisreasonabletobelievethatpeopleborninorafter1950(whowereabove30in 1980)willbeconstrainedbytheOCP.TherelaxationofOCPwillthereforeleadtodifferences infertilityforindividualsinthiscohort.Theabovetwopanelssuggestthatfertilitydecreasesfor bothpeoplewithgirlsand-bornboys,butthereductionisgreaterfor peoplewithboys.Intermsofnumberofchildren,thedifferencebetweenyoungandold cohortsforparentswithaboyis0.05biggerthanthatforparentswithagirl. Meanwhile,thebottompanelshowsthat,forparentsofbothgirlandboy,the youngcohorthasabettermentalhealththantheoldcohort,buttheimprovementinmentalhealth isbiggerfortheboyparents,whohavegreaterdecreaseinfertility. 3.4RegressionResults Somepreviousstudiesshowthatduetodifferentgenderroleswithinafamily,theeffectsoffertil- ityonmentalhealtharedifferentformothersandfathers(IslamandSmyth,2010).Astraditional caregivers,womenmaysuffermorefromhavingadditionalchildren.Inaddition,someresearch thattheeffectsofmarriageondepressivesymptomsaredifferentformenandwomen(Earle etal.,1997).Inordertoinvestigatethedifferenteffectsforeachgroup,besidestheregressions basedonthewholesample,Ialsoruntheregressionsformenandwomenseparatelyforallspeci- TableI.2andTableI.3displayourstageregressionsof kids 2 ict and nkids ict ontheinteraction termsofagecohortdummiesandgenderofbirth( f l inEquation(2)).Oneusefulcheckof instrumentvalidityistoseeitseffectontheuntreatedgroup,whichistheoldcohortsinthiscase. InTableI.2,wetheinteractionsareinthestageforyoungcohorts,butnot 65 foroldcohorts.BecausetheamendedOCPallowedparentstohaveasecondbirthif theonewasagirl,wewouldexpecttheinstrumentstohavelargerpowerinexplainingthe discretechangeinnumberofchildrenfrom1to2.ComparingTableI.2andTableI.3,we TableI.3reportsfewertermsfortheyoungcohorts. Animportantconcernwiththe2SLSistheweakinstrumentsproblem(StaigerandStock,1994). TheCragg-DonaldWaldFstatisticsreportedinTableI.2andTableI.3suggestthattheinteraction termsofagecohortdummiesandgenderofbirthtendtobeprettyweakinstruments,with F-statisticsbetween1.44to6.79(TableI.4andTableI.5showthatusinginteractiontermsforco- hortsage61toage70giveshigherF-statistics,butstillcannotpasstheweakIVtestatthe10% level.).Therefore,inthefollowingwork,Iwillnotusethemasmyinstruments.Thepurposeof TableI.2andTableI.3istoprovideabasisforchoosingcutoffage.FromAppendixTables2-5, wecanthattheinteractiontermsaremainlypositiveforcohortsyoungerthan62(peopleborn inorbefore1950).ThisisconsistentwithmyinChapter1ofthedissertation.Therefore, Iamgoingtouse62asthecutoffagetodividepeopleintotwogroups,andusetheinteractionsof groupdummyandgenderofasthesingleinstrument.Ialsoshiftthecutoffageto61, whichyieldssimilarresults. TableE.3andTableE.4showthetstageregressionsforfiTwoorMoreChildrenflandfiNum- berofChildrenflrespectively.Thecoefontheinstrumentvariable,theinteractiontermsof fiyoungerthan62flandisagirlflareforbothTheF-statistics suggestthatthisinstrumentpassestheweakinstrumenttestsatthe1%levelforall TableE.3(TableE.4)indicatesthatcomparedtopeopleage62andabove,thedifferenceinprob- abilityofhavingtwoormorechildren(numberofchildren)betweenparentswithgirls andboysforpeopleyoungerthan62is10.2%(0.263)larger. TableE.5reportstheregressionresultsofEquation(1)usingOLS.TheleftpanelusesfiTwoor MoreChildrenfl( kids 2)asameasureoffertility,whiletherightpanelusesfiNumberofChildrenfl ( nkids ).Bothsuggestnoeffectsoffertilityonparents'mentalhealth.Childbirth decisionisendogenoustomentalhealth,andthereforeOLSestimatesmightbebiased.TableE.6 66 showstheregressionresultsof2SLSusingtheinteractionofagegroupdummyandgenderof birthastheinstrument.Again,theleftpanelisfor kids 2,whiletherightpanelisfor nkids .After controllingendogeneityinfertility,nowthecoefonfertilityindicatethatmotherswithmore childrenhavehigherriskofexperiencingdepressionsymptomsinruralChina,whiletheeffects onfathersaregenerallynotHowever,thedifferenteffectsformenandwomenare notstatisticallyasthecoefontheinteractiontermoffertilityandgenderare whenIuseafullmodelwithallcontrolvariablesinteractedwithgender.Comparing TableE.5andTableE.6,wecanthatwithoutcontrollingendogeneity,OLSunderestimatesthe negativeeffectsoffertilityonparents'mentalhealth.Theunderestimationmightbecausedbyei- therthenegativeeffectsofdepressiononmarriageandfertility,orthenegativecorrelationbetween poormentalhealthandpreferenceforchildren.Let'stakeacloserlookatthecoefThe coefon kids 2inColumn(3)impliesthathavingadditionalchildrenwillraiseCES-Dby 9.83forwomenwithonechild,andthecoefonnumberofchildreninColumn(6)suggests thathavingonemorechildwillincreaseCES-Dby3.33,that's33.8%oftheaveragelevelforall womeninoursample.Inmysample,formotherswithmorethan1child,theaveragenumberof childrenis3.31.Foramotherwith1child,usingthecoefinColumn(6),wecancalculate thatthetotaleffectsondepressionfromadditional2.31childrenis7.69,similartotheestimated effectson kids 2. 8 Theestimatedeffectsofothercontrolvariablesarelargelyasexpected.Earlychildbearing,sepa- ration,divorce,andwidowingwillraisethesymptomsofdepression,whilemoreeducationleads tobettermentalhealth.Thearealsoconsistentwiththeextensiveliteratureshowingthat menhavebettermentalhealththanwomen. Aspoormentalhealthisfoundtobepositivelyassociatedwithmortalityrate,onemayworryabout theproblemofsampleselection.IndividualswithhigherCES-Dtendtodieearly(Demyttenaere etal.,2004),whichmeansthattheexclusionofthisgroupofpeoplefromoursampletendstobias 8 Thestandarderrorsof2SLSestimatesaresomewhatlarge,wetriedtouseTS2SLStosolve this.Unfortunately,thepopulationinthecensusdataarenotthesameasthepopulationhere,and theregressionresultsarenotsatisfying. 67 theestimatesupwards.Therefore,theestimatednegativeeffectsinTableE.6canbeviewedasa lowerbound. Sofar,Ihavefocusedontheaverageeffectsoffertilityonmen'sandwomen'smentalhealth.For individual,however,theeffectsonmentalhealthmayvaryacrosspeoplewithdifferentcharacter- istics.Therearedifferentreasonstoexpecttheeffectstobeheterogeneous(Cáceres-Delpianoand Simonsen,2012).Forexample,Ifoundthenegativeeffectsofchildbirthonfemalelaborforce participationtobestrongerformoreeducatedwomen(womenwithatleastprimaryschooledu- cation)inruralChinainChapter1.Iftheemploymentisthemainchannelthroughwhichfertility affectsmentalhealth,thenwewouldexpecttoseelargernegativeeffectsonmentalhealthfor moreeducatedwomenwhenrunregressionsforwomenwithdifferenteducationlevelsseparately. Tocheckforthis,Idividethesampleintosubgroupsbasedonpeople'seducationlevelandrun regressionsonsubgroupsseparatelytochecktheheterogeneouseffects.Ifanindividualhasless than6yearsofschooling,he/sheisputintothelesseducatedgroup;otherwisehe/shebelongsto themoreeducatedgroup. TableE.7andTableE.8presentthe2SLSresultsofheterogeneouseffectsonmentalhealthfor peoplewithdifferentlevelsofeducation(TheOSLestimatesofheterogeneouseffectsarereported inTableI.7andTableI.8).Thesamplesizeismuchsmallerforeachsubgroup,asaresult,many coefarenolongerIntermsofmagnitude,thereseemsnoheterogeneityinef- fectsonmentalhealthformenwithdifferenteducationlevels.Meanwhile,theresultsfromboth TableE.7andTableE.8revealalargernegativeeffectsoffertilityonwomen'smentalhealthforthe moreeducatedgroup.Forexample,estimatesfromcolumns(3)and(6)inTableE.8showthatone morechildrenwillraiseCES-Dby1.5pointsfortheless-educatedgroup,and5.3pointsforthe more-educatedgroup.Ifwetakethisastrue,thenonepossibilitytoexplainthenegativeeffectson mentalhealthmightbethefactorsoftimeallocationandemployment.Womenreducetheirlabor forceparticipationwhentheyhavemorechildren(asfoundinchapter1),andfeweremployment attachmentswillleadtopoorermentalhealth.Asbothofthesecoefsarehow- ever,wecannotcometoaconclusionthatthenegativeeffectsonmentalhealtharestronger 68 formotherswithmoreeducation. 3.5PhysicalHealthandLivingArrangements Giventheevidenceofnegativeeffectsofnumberofchildrenonmother'smentalhealth,wemay wanttofurtherlookatthepossiblepathwaysofgeneratingsucheffects.Asdiscussedinintroduc- tionpart,threedifferentchannelshavebeensuggestedbytheliteraturethatcanaffectthemental healthofparentsinthelongrun.Inthispaper,we'regoingtoinvestigatetwoofthem,biological effectsandsupporteffects. 3.5.1Self-ReportedHealthandChronicDiseases Ithasbeenshownthatpoorphysicalhealthmayleadtodepressionsymptoms(Berkmanetal., 1986),andmanystudiessuggestthatfertilitymayleadtosomephysicalhealthproblemsforthe parentsinthelongrun.UsingdatafromNHISduring1982-2003,Cáceres-DelpianoandSimon- sen(2012)concludethathavingmorechildrenwillincreasethelikelihoodofhavinghighblood pressureandbecomingobeseforthemothers.InChina,WuandLi(2012)showthathavingmore childrenwoulddecreasetheresourcesallocatedtomothersandthusleadstobothparentsbeing underweight.BasedonasampleofwomenfromShanghai,Zhangetal.(2009)provideevidence thathavingmorechildrenincreasestheriskofstrokelaterinlife.Morerecently,us- ingpilotdatafromtheChinaHealthandRetirementLongitudinalSurvey(CHARLS),Islamand Smyth(2010)thathavingfewerchildrenhasapositiveeffectonself-reportedhealth 9 .Inthis paper,inordertotestthehypothesisthatchildbearingnegativelyimpactsthemother'sphysical health,whichinturnexertslong-termeffectsonmentalhealth,Iusetwodifferentmethods.First, followIslamandSmyth(2010),Iuse2SLStoestimatetheeffectsoffertilityontheself-reported health(SRH),whichisanindicatorofphysicalhealth.TheSRHquestioninCHARLSisphrased 9 Duetothelimitednumberofobservation,theydonotlookatmenandwomenseparately. 69 asfollows:fiIngeneral,howwouldyourateyourhealth?flRespondentsareaskedtochoosea pointalongave-pointscale,andtwoscalesarerandomlyprovided.Oneisfi(1)excellent,(2) verygood,(3)good,(4)fair,and(5)poorfl;andtheotherisfi(1)verygood,(2)good,(3)fair,(4) poor,and(5)verypoorfl.Icombinetheresponsestogether,andrecodetheSRHas:excellent/very good=4,good=3,fair=2,andpoor/verypoor=1.Thehigherthescoreis,thebetterhealthisre- ported. 10 TableE.9showshighlysimilarresultstoTableE.6,whichrepresents2SLSresultsformental health.TherowinTableE.9indicatesthatwomenwithmorechildrenhavegenerallypoorer self-reportedhealth(onlyat10%level),whiletheeffectsoffertilityonmen'smental healthareAsself-reportedhealthmetricsaresometimearguedtobebiaseddueto subjectivityandmeasurementerror(Mu,2013),regressionsonthediagnosisofsomechronicdis- easesareusedastherobustnesscheck.TableE.10reports2SLSresultsonthediagnosisofthree mostcommonchronicdiseasesinoursample,i.e.arthritisorrheumatism(32.42%),hypertension (23.44%)andstomachorotherdigestivedisease(17.87%).Itsuggeststhatadditionalchildrenwill notleadtohigherprobabilityofdiagnosisofanyofthesechronicdiseases. Thesecondwaytotesttheroleofphysicalhealthistoaddphysicalhealthindicatortotheoriginal 2SLSonmentalhealth.Iftheeffectsoffertilityonmentalhealthareduetotheworseningof physicalhealthcausedbyadditionalchildren,weexpectthecoefonfertilitywillbereduced onceweconditiononindicatorofphysicalhealth. TableE.11showsthe2SLSresultswhenweincludetheSRHasacontrolvariable.Thesignif- icantlynegativecoefonSRHsuggestthatphysicalhealthhasapositiveeffectonmental health.Fortheothervariables,mostofthecoefdonotchange.Thecoefonfertility isreducedsomewhatbutstillpositive.Thisimpliesthattherearestillsomeother reasonsforthenegativeeffectsonmentalhealth. 10 Fororderedresponsemodels,wecanuseorderedprobit/logitaswell.Tocontrolforendo- geneity,canapplyRivers-Vuongmethod,whichgivesussimilarresultstoTableE.9. 70 3.5.2LivingArrangements Livingarrangementsaffectsupporteffectsdirectly,aslivingtogetherwithchildrenleadstohigher emotionalandphysicalsupportfromadultchildren(EvensonandSimon,2005).BuberandEngel- hardt(2008)showthatsupportsfromchildrensubstantiallyreducetheprobabilityofexperiencing depressivesymptoms.InChina,havingmorechildrenhasbeenfoundtobeassociatedwithhigher probabilityofcoresidencewithchildren(ZimmerandKwong,2003).Inthispaper,similartothe aboveanalysisonphysicalhealth,Itrytwodifferentmethodstotesttheintermediatingeffectsof livingarrangementsonmentalhealth.First,2SLSisappliedtoestimatetheeffectsoffertilityon livingarrangements,thenlivingarrangementsindicatorisaddedtotheoriginal2SLSonmental health. CHARLSaskedtherespondentthelivingsituationofeachoftheirchild,Iconstructedadummy variablecalledfilivingtogetherfltomeasurethelivingarrangement.filivingtogetherflequalsto one,iftherespondentsreporthavingatleastonechildcurrentlylivinginthesamehousehold,or thesameoradjacentdwelling/courtyard;equalstozeroforallothercases.Inoursample,51.27% oftherespondentsarelivingtogetherwiththeirchildren,andpeoplewithonechild(51.55%)re- portssimilarrateoffilivingtogetherfltopeoplewithmorethanonechildren(51.24%). TableE.12indicatesthatdifferentnumberofchildrendoesnotleadtodifferentlivingarrange- mentsinoursample.Theeffectsofco-residencewithchildrenonmentalhealtharereported inTableE.13.Itshowsthatlivingwithchildrencanhelpimprovemen'smentalhealthat10% level,whiletheeffectsofco-residenceonwomen'smentalhealthare positive.Aftercontrollinglivingarrangements,thechangeincoefoffertilityisverysmall comparingtothemainresultsinTableE.6.ThisisconsistentwiththeinTableE.12. Therefore,oursamplecannotprovideevidenceonlivingarrangementsasapathwayforfertilityto affectparents'mentalhealth. 71 3.6RobustnessCheck Inthispaper,IusescoresonCES-Dtomeasureindividual'smentalhealth.AsCES-Disaself- reportedmeasurement,itmightbebiased.Differentpeoplemayhavedifferentscalesofdepressive feelings,andthustheymaygivedifferentanswersforCES-Deveniftheyhavethesamemental healthstatus.Forexample,ifpeoplewithmorechildrentendtohavesystematicallydifferent scalesthanpeoplewithfewerchildren,thenthe2SLSestimateswillbethesumofthetrueeffects offertilityandthedifferencesinscales.Tocheckwhethertherearesomesystematicdifferences inscales,thebestwayistocomparepeoples'scalesdirectly.Onewaytogetanindividual'sscale istousevignettequestions.Ifthesamevignettequestionsareshowntodifferentrespondents,the onlyreasonforrespondentstogivedifferentresponsesistheirheterogeneousscales. InCHARLS,vignettequestionscoveringsixdomains(bodypain,sleepdisorder,difinmo- bility,cognitionproblems,shortnessofbreath,andmentalproblems)areincludedinthequestion- naire.Respondentsareaskedtoevaluatethehealthconditionsofthehypotheticalpersons.They aregiventworandomlyselecteddomainswiththreevignettequestionsforeach.Inoursample, only810people,451ofthemwomen,wereaskedaboutvignettequestionsonmentalproblems. Icalculatevignettescoresbysummingupthescoresfromthreequestions.Ahigherscoremeans peopletendtohavealowerthresholdforreportingdepressivesymptoms,andthusaremorelikely toover-reporttheirowndepression.Alowerscoremeanspeopletendtohaveahigherthreshold forreportingdepressivesymptoms,andthusaremorelikelytounder-reporttheirowndepression. FigureF.2andFigureF.3plotvignettequestionscoresagainstthetwokeyvariablesinthisstudy, numberofchildrenandtheCES-Dscore,respectively.Thereisnosalientevidenceofanysys- tematicbiasinvignettescoresinthesetwoTherefore,themeasurementerrorinCES-D duetosystematicdifferencesinrespondent'sscalesisnotabigconcerninthisstudy.Iattempted tocontrolvignettescoresin2SLSregressions,butduetotheverysmallsamplesize,mostofthe resultsarenotprecise. 72 3.7Conclusion Thetotaleffectoffertilityonparent'smentalhealthisanempiricalquestion,whichiscomplicated bytheendogeneityprobleminvolvedwiththefertilitydecisions.Mostoftheexistingliterature ignorestheendogeneityandgeneratesurprisinglyinconsistentresults.Thispaperistheoneto treatfertilityasendogenous,andtoidentifythecausaleffectoffertilityonparent'smentalhealth inChinausing2SLS.China'sOne-ChildPolicy,whichcameasasurprisetomanyfamiliesand hadvariationsinitsimplementation,issnaturalexperiment.Iconstructexogenousvariationin fertilitythroughaDIDstrategybasedontheOCPandusethatastheinstrumenttodo2SLS.Using datafromthe2011ChinaHealthandRetirementLongitudinalStudy,resultsshowthat,forwomen age45andaboveinruralChina,havingmorechildrenhasanegativeeffectontheirmentalhealth. Theeffectonmen'smentalhealthisalsonegative,but Afterestimatingtheeffectsonmentalhealth,Iconductfurtherinvestigationstotesttheroleof physicalhealthingeneratingsucheffectsusingtwomethods.First,IuseSelf-ReportedHealth (SRH)astheindicatorofphysicalhealthandapplythesameinstrumentsinthe2SLS.Second, IaddSRHtotheoriginal2SLSonmentalhealthascontrolvariables.Theresultssuggestthat physicalhealthcanhelptoexplainaverylimitedpartofthetotaleffectonmentalhealth. Ifoundstrongereffectsoffertilityonmentalhealthformoreeducatedwomenwheninvestigating theheterogeneouseffects.Thisisinlinewiththeinthechapterthatmoreeducated womenaremoreresponsivetofertilitychangeintermsoflaborforceparticipationdecisions. Thesefactorssuggestthatthechangeinemploymentattachmentmaycontributetothechange inmentalhealth.Inthefuture,withbetterinformationonemploymenthistory,wecanmore thoroughlytestthishypothesis. ChinaisnowinaprocessofrelaxingtheOne-ChildPolicy.Withmorechildren,thispapersuggests thatthementalhealthproblemsinChinamightgetevenworse.Moreresourceswillbeneededto adequatelyaddressthisproblem. 73 APPENDICES 74 APPENDIXA TABLESFORCHAPTER1 TableA.1:DescriptiveStatistics,Womenaged16-45withatleastonechild MeansS.D. #ofChildrenNumberofsurvivingchildren2.2230.0011 Kids2=1ifmotherhasmorethan1child,=0otherwise0.7560.0005 LFP=1ifthewomanhasajoborisfiwaitingtobeem- ployedflonthedayofthecensus,=0otherwise 0.9220.0003 AgeMother'sageinyearsonJuly1st,199032.4360.0067 Ageat1stBirthMother'sageinyearswhenchildwasborn22.7840.0030 Ageat2ndBirthMother'sageinyearswhensecondchildwasborn25.5270.0041 non-Han=1ifbothmotherandfatherareminority,=0other- wise 0.0900.0003 First-BornGirl=1ifthechildisagirl,=0otherwise0.4840.0006 Primary=1ifmother'shighesteducationachievementispri- maryschool,=0otherwise 0.4760.0006 Junior=1ifmother'shighesteducationachievementisju- niorhighschool,=0otherwise 0.2180.0005 Senior=1ifmother'shighesteducationachievementisse- niorhighschool,=0otherwise 0.0420.0002 Primary_Husband=1iffather'shighesteducationachievementispri- maryschool,=0otherwise 0.4150.0005 Junior_Husband=1iffather'shighesteducationachievementisju- niorhighschool,=0otherwise 0.3930.0005 Seniro_Husband=1iffather'shighesteducationachievementisse- niorhighschool,=0otherwise 0.1170.0004 Notes:Dataisfrom1990ChinaPopulationCensus.Sampleincludeswomenaged16-35withatleastone childwhoarehouseholdheadsorthespouseofthehouseholdheads.Womenwhosechildislessthan oneyearoldareexcluded. 75 TableA.2:SummaryStatisticsforEachSample 1-ChildProv 1 1-Boy-2-GirlProv 3 2-ChildrenProv 2 Obs.165,969612,78545,855 MeansS.D.MeansS.D.MeansS.D. #ofChildren1.8260.00202.2960.00132.6890.0061 Kids20.6080.00120.7900.00050.8430.0017 LFP0.9840.00030.9010.00040.9810.0006 Age32.9920.015332.3190.007831.9730.0289 Ageat1stBirth22.9710.006122.7890.003522.0470.0132 Ageat2ndBirth 4 26.0740.010325.4860.004624.6210.0158 non-Han0.0240.00040.0830.00040.4160.0023 First-BornGirl0.4820.00120.4840.00060.4860.0023 Primary0.5330.00120.4690.00060.3530.0022 Junior0.2180.00100.2230.00050.1420.0016 Senior0.0390.00050.0440.00030.0240.0007 Primary_Husband0.4690.00120.3970.00060.4500.0023 Junior_Husband0.3790.00120.4050.00060.2850.0021 Senior_Husband0.0940.00070.1280.00040.0640.0011 Notes:Dataisfrom1990ChinaPopulationCensus. 1 Listof1-ChildProvinces:Hainan,Yunnan,Qinghai,Ningxia,Xinjiang. 2 Listof2-ChildrenProvinces:Beijing,Shanghai,Tianjin,Jiangsu,Sichuan. 3 1-Boy-2-GirlProvinces:AllotherprovincesexceptTibet,19provincesintotal. 4 Thisisbasedonmotherswithatleasttwochildren.Thesamplesizesare100,478for 1-childprovinces,482,842for1-boy-2-girlprovinces,38,597for2-childrenprovinces. 76 TableA.3:HanVs.non-HanandFirst-BornGirlVs.First-BornBoy 1-ChildProv+1-Boy-2-GirlProv1-Boy-2-GirlProvinces Hannon-HanFirst-bornGirlFirst-BornBoy Obs.723,94954,805296,183315,777 MeansS.D.MeansS.D.MeansS.D.MeansS.D. #ofChildren2.1790.00112.4130.00482.4110.00192.1870.0017 Kids20.7480.00050.7970.00170.8100.00070.7700.0007 LFP0.9190.00030.9110.00120.9010.00050.9010.0005 Age32.4760.007232.2860.026832.2710.011132.3620.0109 Ageat1st Birth 22.8230.003122.8930.012322.8410.005122.7380.0048 Ageat2nd Birth 1 25.5810.004425.6660.015825.4760.006525.4940.0065 non-Hann.a.n.a.n.a.n.a.0.0830.00050.0830.0005 First-BornGirl0.4840.00060.4850.0021n.a.n.a.n.a.n.a. Primary0.4860.00060.4390.00210.4700.00090.4680.0009 Junior0.2240.00050.1990.00170.2230.00080.2240.0007 Senior0.0430.00020.0450.00090.0440.00040.0450.0004 Primary_Husband0.4120.00060.4160.00210.3970.00090.3980.0009 Junior_Husband0.4030.00060.3610.00210.4070.00090.4040.0009 Senior_Husband0.1210.00040.1160.00140.1280.00060.1280.0006 1 Thisisbasedonmotherswithatleasttwochildren.Thesamplesizesare539,737forHanand 43,583fornon-Hanrespectivelyinthe1-childprovincesandthe1-boy-2-girlprovinces,239,597for girland242,830forboyinthe1-boy-2-girlprovinces. 77 TableA.4:DIDEstimatesRegardingEthnicity Having2orMoreChildrenLaborForceParticipation Old Cohorts 1 Young Cohorts 2 DifferenceOld Cohorts 1 Young Cohorts 2 Difference Han0.9640.691-0.2730.9040.9200.016 (s.d./s.e.)(0.1851)(0.4620)(0.0018)(0.2943)(0.2706)(0.0011) non-Han0.9700.734-0.2360.9230.910-0.012 (s.d./s.e.)(0.1707)(0.4418)(0.0060)(0.2668)(0.2856)(0.0040) Difference0.0050.0430.0370.019-0.010-0.029 (s.e.)(0.0026)(0.0021)(0.0065)(0.0041)(0.0012)(0.0040) Notes:Thesampleismadeofobservationsfromtherestrictedprovinces(the1-childprovinces andthe1-boy-2-girlprovinces).Standarderrorsareinparenthesis. 1 OldCohortsareconsistedofmothersolderthan40butyoungerthan46in1990. 2 YoungCohortsareconsistedofmothersage40oryounger. 78 TableA.5:DIDEstimatesRegardingGenderofFirstBirth Having2orMoreChildrenLaborForceParticipation Old Cohorts 1 Young Cohorts 2 DifferenceOld Cohorts 1 Young Cohorts 2 Difference First-Born Boy 0.9670.713-0.2540.8780.9030.025 (s.d./s.e.)(0.1775)(0.4522)(0.0027)(0.3270)(0.2958)(0.0018) First-Born Girl 0.9710.756-0.2160.8930.9020.008 (s.d./s.e.)(0.1672)(0.4297)(0.0027)(0.3085)(0.2975)(0.0020) Difference0.0040.0420.0390.015-0.001-0.016 (s.e.)(0.0015)(0.0012)(0.0038)(0.0028)(0.0008)(0.0027) Notes:Thesampleismadeofobservationsfrom1-Boy-2-Girlprovinces.Standarderrorsare inparenthesis. 1 OldCohortsareconsistedofmothersolderthan40butyoungerthan46in1990. 2 YoungCohortsareconsistedofmothersage40oryounger. 79 TableA.6:OLSand2SLSEstimatesoftheEffectofAdditionalChildrenonFemale LFP (1)(2)(3)(4) Restricted1-Boy-2-GirlRestricted1-Boy-2-Girl A:OLS kids20.000-0.0020.0150.013 (0.001)(0.002)(0.002)***(0.004)*** non-Han0.0030.008 (0.008)(0.008) First-BornGirl-0.001-0.0010.0000.000 (0.000)(0.001)*(0.001)(0.001) Observations 778,754561,921396,607282,125 B:2SLS kids2-0.153-0.084-0.157-0.108 (0.046)***(0.036)**(0.054)***(0.041)*** non-Han0.0040.01 (0.008)(0.008) First-BornGirl0.0080.0030.010.005 (0.003)***(0.002)(0.003)***(0.002)*** Cragg-DonaldWaldF statistic 16.28611.73244.73519.708 HansenJstatitstic31.46631.00830.53926.566 Observations 778,710561,875396,580282,095 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.*at10% level,**at5%level,***at1%level.Allregressionscontrolsagecohort dummies,mother'sageatbirth,educationlevelsforbothparents,countyedeffects. Col(1)isonobservationsfromthe1-childprovincesandthe1-boy-2-girlprovinces;Col(2)is onHanpeopleonlyinthe1-boy-2-girlprovinces;Col(3)isonobservationsfromthe1-child provincesandthe1-boy-2-girlprovinces,exceptformotherswithbirthlaterthan1981;Col (4)isonHanpeopleonlyinthe1-boy-2-girlprovinces,exceptformotherswithbirthlater than1981. 80 TableA.7:CoefofInteractionTermsforthe1stStageRegressionsRegardingEthnicity (1)(2)(1)(2) age16*non-Han-0.028age31*non-Han0.0850.082 (0.070)(0.014)***(0.012)*** age17*non-Han0.001age32*non-Han0.0810.084 (0.054)(0.014)***(0.011)*** age18*non-Han0.059age33*non-Han0.0760.082 (0.071)(0.013)***(0.011)*** age19*non-Han0.072age34*non-Han0.0660.079 (0.039)*(0.013)***(0.011)*** age20*non-Han0.005age35*non-Han0.0530.065 (0.028)(0.012)***(0.010)*** age21*non-Han0.036age36*non-Han0.0480.061 (0.021)*(0.012)***(0.009)*** age22*non-Han0.039age37*non-Han0.0270.042 (0.020)**(0.012)**(0.009)*** age23*non-Han0.0680.712age38*non-Han0.0280.035 (0.019)***(0.263)***(0.011)**(0.009)*** age24*non-Han0.0490.096age39*non-Han0.0120.019 (0.019)***(0.042)**(0.011)(0.009)** age25*non-Han0.0520.087age40*non-Han0.0040.01 (0.019)***(0.020)***(0.011)(0.008) age26*non-Han0.0580.099age41*non-Han-0.0060.006 (0.019)***(0.022)***(0.011)(0.008) age27*non-Han0.0570.037age42*non-Han-0.012-0.005 (0.019)***(0.022)*(0.011)(0.008) age28*non-Han0.0550.062age43*non-Han-0.0060.005 (0.018)***(0.016)***(0.011)(0.008) age29*non-Han0.0810.051age44*non-Han-0.0030.005 (0.016)***(0.017)***(0.012)(0.009) age30*non-Han0.0790.053 (0.016)***(0.015)*** Observations 778,754396,607 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.*at10%level,** at5%level,***at1%level.Allregressionscontrolsagecohortdummies,ethnicity dummies,genderofbirth,mother'sageatbirth,educationlevelsforbothparents,countyed effects.Col(1)isonobservationsfromthe1-childprovincesandthe1-boy-2-girlprovinces;Col(2)ison observationsfromthe1-childprovincesandthe1-boy-2-girlprovinces,exceptformotherswithbirth laterthan1981. 81 TableA.8:CoefofInteractionTermsforthe1stStageRegressionsRegardingGenderof 1stBirth (1)(2)(1)(2) age16*First- BornGirl -0.043age31*First- BornGirl 0.050.06 (0.017)**(0.009)***(0.008)*** age17*First- BornGirl -0.034age32*First- BornGirl 0.0540.056 (0.064)(0.009)***(0.007)*** age18*First- BornGirl 0.042age33*First- BornGirl 0.0530.064 (0.045)(0.009)***(0.008)*** age19*First- BornGirl -0.014age34*First- BornGirl 0.0470.064 (0.025)(0.009)***(0.007)*** age20*First- BornGirl 0.004age35*First- BornGirl 0.040.053 (0.017)(0.008)***(0.007)*** age21*First- BornGirl -0.016age36*First- BornGirl 0.030.044 (0.012)(0.008)***(0.007)*** age22*First- BornGirl 0.008age37*First- BornGirl 0.0170.029 (0.01)(0.008)**(0.006)*** age23*First- BornGirl 0.0181.021age38*First- BornGirl 0.0110.024 (0.010)*(0.011)***(0.007)(0.006)*** age24*First- BornGirl -0.001-0.086age39*First- BornGirl 0.0080.016 (0.009)(0.079)(0.007)(0.006)*** age25*First- BornGirl 0.0120.037age40*First- BornGirl -0.0010.008 (0.009)(0.030)(0.007)(0.005) age26*First- BornGirl 0.0190.036age41*First- BornGirl -0.0020.002 (0.008)**(0.019)*(0.007)(0.005) age27*First- BornGirl 0.0160.047age42*First- BornGirl -0.0010.002 (0.008)**(0.012)***(0.007)(0.005) age28*First- BornGirl 0.0060.039age43*First- BornGirl 0.0090.009 (0.009)(0.011)***(0.008)(0.006) 82 TableA.8(cont'd) age29*First- BornGirl 0.0170.038age44*First- BornGirl 0.0090.007 (0.009)*(0.010)***(0.008)(0.006) age30*First- BornGirl 0.0380.045 (0.009)***(0.009)*** Observations 561,921282,125 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.*at 10%level,**at5%level,***at1%level.Allregressionscontrols agecohortdummies,ethnicitydummies,genderofbirth,mother'sageatbirth, educationlevelsforbothparents,countyedeffects.Col(1)isonHanpeopleonlyinthe 1-boy-2-girlprovinces;Col(2)isonHanpeopleonlyinthe1-boy-2-girlprovinces,except formotherswithbirthlaterthan1981. 83 TableA.9:CoefofInteractionTermsfortheReducedFormRegressionsRegardingEthnic- ity (1)(2)(1)(2) age16*non-Han0.025age31*non-Han-0.031-0.031 (0.062)(0.011)***(0.013)** age17*non-Han-0.097age32*non-Han-0.023-0.022 (0.089)(0.010)**(0.011)** age18*non-Han-0.025age33*non-Han-0.029-0.029 (0.018)(0.010)***(0.010)*** age19*non-Han-0.01age34*non-Han-0.03-0.029 (0.026)(0.010)***(0.010)*** age20*non-Han-0.022age35*non-Han-0.029-0.026 (0.016)(0.010)***(0.010)** age21*non-Han-0.028age36*non-Han-0.027-0.027 (0.013)**(0.010)***(0.010)** age22*non-Han-0.03age37*non-Han-0.02-0.017 (0.011)***(0.010)**(0.010)* age23*non-Han-0.038-0.104age38*non-Han-0.033-0.031 (0.011)***(0.023)***(0.010)***(0.010)*** age24*non-Han-0.033-0.038age39*non-Han-0.021-0.02 (0.010)***(0.033)(0.010)**(0.010)* age25*non-Han-0.032-0.063age40*non-Han-0.025-0.023 (0.010)***(0.021)***(0.010)**(0.011)** age26*non-Han-0.034-0.037age41*non-Han-0.027-0.028 (0.010)***(0.024)(0.010)***(0.010)*** age27*non-Han-0.035-0.025age42*non-Han-0.01-0.007 (0.010)***(0.019)(0.011)(0.011) age28*non-Han-0.032-0.049age43*non-Han-0.013-0.014 (0.010)***(0.015)***(0.010)(0.010) age29*non-Han-0.03-0.045age44*non-Han-0.011-0.011 (0.011)***(0.015)***(0.010)(0.010) age30*non-Han-0.029-0.022 (0.010)***(0.013) Observations 778,754396,607 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.*at10%level,** at5%level,***at1%level.Allregressionscontrolsagecohortdummies,ethnicity dummies,genderofbirth,mother'sageatbirth,educationlevelsforbothparents,countyedef- fects.Col(1)isonobservationsin1-ChildProvincesand1-Boy-2-GirlProvinces;Col(2)isonobservations in1-ChildProvincesand1-Boy-2-GirlProvinces;exceptformotherswithbirthlaterthan1981; 84 TableA.10:CoefofInteractionTermsfortheReducedFormRegressionsRegardingGen- derof1stBirth (1)(2)(1)(2) age16*First- BornGirl 0.049age31*First- BornGirl -0.023-0.017 (0.090)(0.008)***(0.009)* age17*First- BornGirl -0.018age32*First- BornGirl -0.021-0.016 (0.072)(0.008)**(0.008)* age18*First- BornGirl -0.05age33*First- BornGirl -0.023-0.021 (0.031)(0.008)***(0.008)*** age19*First- BornGirl -0.036age34*First- BornGirl -0.021-0.02 (0.018)**(0.008)***(0.008)** age20*First- BornGirl -0.03age35*First- BornGirl -0.02-0.018 (0.012)**(0.008)**(0.008)** age21*First- BornGirl -0.016age36*First- BornGirl -0.023-0.023 (0.009)*(0.008)***(0.008)*** age22*First- BornGirl -0.015age37*First- BornGirl -0.018-0.017 (0.008)*(0.008)**(0.008)** age23*First- BornGirl -0.02-0.098age38*First- BornGirl -0.019-0.017 (0.008)**(0.008)***(0.008)**(0.008)** age24*First- BornGirl -0.0220.061age39*First- BornGirl -0.018-0.018 (0.008)***(0.071)(0.008)**(0.008)** age25*First- BornGirl -0.022-0.045age40*First- BornGirl -0.015-0.016 (0.008)***(0.034)(0.009)*(0.008)* age26*First- BornGirl -0.023-0.034age41*First- BornGirl -0.013-0.013 (0.008)***(0.019)*(0.009)(0.009) age27*First- BornGirl -0.024-0.047age42*First- BornGirl -0.014-0.015 (0.008)***(0.014)***(0.009)(0.009)* age28*First- BornGirl -0.023-0.023age43*First- BornGirl -0.018-0.018 (0.008)***(0.012)*(0.009)*(0.009)** 85 TableA.10(cont'd) age29*First- BornGirl -0.015-0.003age44*First- BornGirl -0.013-0.014 (0.008)*(0.011)(0.010)(0.010) age30*First- BornGirl -0.026-0.038 (0.008)***(0.009)*** Observations 561,921282,125 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.*at10% level,**at5%level,***at1%level.Allregressionscontrolsage cohortdummies,ethnicitydummies,genderofbirth,mother'sageatbirth,education levelsforbothparents,countyedeffects.Col(1)isononHanpeopleonlyin1-Boy-2-Girl Provinces;Col(2)isonHanpeopleonlyin1-Boy-2-GirlProvinces,exceptformotherswith birthlaterthan1981. 86 TableA.11:HeterogeneousEffectofAdditionalChildrenonFemale LFP (1)(2)(3)(4) Primary Junior Primary Junior A:OLS kids20.005-0.010.004-0.012 (0.001)***(0.002)***(0.002)**(0.003)*** non-Han0.005-0.006 (0.008)(0.009) First-BornGirl0-0.002-0.001-0.002 (0.001)(0.001)*(0.001)(0.001)* Observations 572,222206,532426,719161,725 B:2SLS kids2-0.1850.06-0.059-0.035 (0.053)***(0.077)(0.045)(0.037) non-Han0.007-0.008 (0.008)(0.010) First-BornGirl0.01-0.0070.002-0.001 (0.003)***(0.006)(0.002)(0.002) Observations 572,196206,471426,690161,663 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.*signif- icantat10%level,**at5%level,***at1%level.Allre- gressionsareonobservationsfrom2-childrenprovinces.Allregressionscontrols foragecohortdummies,ethnicitydummies,genderofbirth,mother'sage atbirth,educationlevelsforbothparents,andcountyedeffects.Col(1) isonmotherswithatmostprimaryschooleducationintherestrictedprovinces; Col(2)isonmotherswithatleastjuniorhighschooleducationintherestricted provinces;Col(3)isonmotherswithatmostprimaryschooleducationinthe 1-boy-2-girlprovinces;Col(4)isonmotherswithatleastjuniorhighschooled- ucationinthe1-boy-2-girlprovinces;Col(1)and(2)inPanelBuseDIDbased onethnicityasinstruments.Col(3)and(4)inPanelBuseDIDbasedongender ofbirthasinstruments. 87 TableA.12:CoefofInteractionTermsfortheRegressionsofEducationandGenderofFirst Birth (1)(2)(1)(2) EducationFirst- born Girl EducationFirst- born Girl age16*non-Han-0.7640.559age31*non-Han-0.0870.013 (0.777)(0.187)***(0.152)(0.023) age17*non-Han-1.537-0.039age32*non-Han0.1550.009 (0.977)(0.182)(0.140)(0.022) age18*non-Han-0.274-0.074age33*non-Han0.0220.023 (0.445)(0.081)(0.139)(0.021) age19*non-Han-0.930.044age34*non-Han0.004-0.002 (0.319)***(0.053)(0.140)(0.022) age20*non-Han-0.52-0.019age35*non-Han0.0520.009 (0.192)***(0.033)(0.132)(0.022) age21*non-Han-0.3370age36*non-Han0.054-0.006 (0.163)**(0.026)(0.127)(0.022) age22*non-Han-0.3720.009age37*non-Han0.058-0.005 (0.150)**(0.024)(0.136)(0.022) age23*non-Han-0.3960.004age38*non-Han0.0430.001 (0.149)***(0.023)(0.132)(0.021) age24*non-Han-0.4340.014age39*non-Han0.0710.011 (0.147)***(0.022)(0.136)(0.022) age25*non-Han-0.413-0.005age40*non-Han0.1050.008 (0.140)***(0.020)(0.129)(0.022) age26*non-Han-0.3310.004age41*non-Han0.032-0.003 (0.147)**(0.021)(0.135)(0.023) age27*non-Han-0.2340.001age42*non-Han0.08-0.012 (0.145)(0.020)(0.138)(0.024) age28*non-Han-0.210.006age43*non-Han0.217-0.028 (0.146)(0.022)(0.153)(0.024) age29*non-Han-0.12-0.019age44*non-Han0.233-0.02 (0.159)(0.023)(0.152)(0.025) age30*non-Han-0.2150.004 (0.153)(0.022) Observations 778,754778,754 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.*at10% level,**at5%level,***at1%level.Bothregressionsareonob- servationsfromtherestrictedprovinces.Col(1)controlsforagecohortdummies,ethnicity dummies,andcountyedeffects;Col(2)controlsforagecohortdummies,ethnicitydum- mies,mother'sageatbirth,educationlevelsforbothparents,andcountyedeffects. 88 TableA.13:CoefofInteractionTermsforthe1stStageRegressionsRegardingEthnicity (2-ChildrenProvinces) (1)(2)(1)(2) First Stage Reduced Form First Stage Reduced Form age16*non-Han0.07-0.011age31*non-Han-0.013-0.002 (0.031)**(0.011)(0.034)(0.013) age17*non-Han-0.2110.005age32*non-Han-0.0150.009 (0.196)(0.012)(0.033)(0.011) age18*non-Han0.056-0.019age33*non-Han-0.020.005 (0.120)(0.029)(0.032)(0.011) age19*non-Han0.0070.006age34*non-Han-0.02-0.007 (0.095)(0.018)(0.029)(0.011) age20*non-Han0.0830.014age35*non-Han-0.0340.009 (0.055)(0.021)(0.026)(0.012) age21*non-Han0.2230.017age36*non-Han-0.015-0.006 (0.051)***(0.014)(0.026)(0.011) age22*non-Han0.1540.008age37*non-Han-0.022-0.003 (0.040)***(0.013)(0.023)(0.010) age23*non-Han0.1120.003age38*non-Han-0.0260.004 (0.036)***(0.013)(0.024)(0.012) age24*non-Han0.1350.005age39*non-Han-0.017-0.002 (0.040)***(0.013)(0.024)(0.013) age25*non-Han0.1080age40*non-Han-0.0080.019 (0.039)***(0.012)(0.022)(0.013) age26*non-Han0.0850.001age41*non-Han-0.0310.024 (0.043)**(0.012)(0.020)(0.015) age27*non-Han0.0770.001age42*non-Han-0.0290.005 (0.040)*(0.014)(0.025)(0.015) age28*non-Han0.0390age43*non-Han-0.0260.002 (0.039)(0.011)(0.027)(0.017) age29*non-Han0.0250.007age44*non-Han-0.043-0.011 (0.043)(0.013)(0.028)(0.019) age30*non-Han-0.001-0.003 (0.035)(0.011) Observations 45,85545,855 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.*at10% level,**at5%level,***at1%level.Bothregressionsareonobser- vationsfrom2-ChildrenProvinces.Allregressionscontrolsagecohortdummies,ethnicity dummies,genderofbirth,mother'sageatbirth,educationlevelsforbothparents, andcountyedeffects. 89 TableA.14:CoefofInteractionTermsforthe1stStageRegressionsRegardingGenderof theFirst-Birth(2-ChildrenProvinces) (1)(2)(1)(2) First Stage Reduced Form First Stage Reduced Form First-BornGirl0.067-0.002age31*First- BornGirl -0.012-0.007 (0.027)**(0.018)(0.029)(0.019) age17*First- BornGirl 0.3890.01age32*First- BornGirl -0.0630.003 (0.304)(0.021)(0.026)**(0.022) age18*First- BornGirl -0.2070.001age33*First- BornGirl -0.0360 (0.128)(0.019)(0.026)(0.019) age19*First- BornGirl 0.2370.002age34*First- BornGirl -0.039-0.001 (0.137)*(0.018)(0.026)(0.019) age20*First- BornGirl -0.072-0.026age35*First- BornGirl -0.0640.003 (0.091)(0.038)(0.026)**(0.016) age21*First- BornGirl -0.0210.012age36*First- BornGirl -0.0660.002 (0.057)(0.027)(0.028)**(0.018) age22*First- BornGirl -0.0960.003age37*First- BornGirl -0.0430.005 (0.051)*(0.019)(0.028)(0.020) age23*First- BornGirl -0.0720.005age38*First- BornGirl -0.0630.017 (0.038)*(0.020)(0.031)**(0.021) age24*First- BornGirl -0.0640.015age39*First- BornGirl -0.0450.005 (0.038)*(0.019)(0.028)(0.022) age25*First- BornGirl -0.041-0.008age40*First- BornGirl -0.0270.003 (0.034)(0.018)(0.030)(0.022) age26*First- BornGirl -0.05-0.001age41*First- BornGirl -0.0530.005 (0.036)(0.019)(0.030)*(0.031) age27*First- BornGirl -0.0360.01age42*First- BornGirl -0.0510.03 (0.031)(0.021)(0.027)*(0.022) age28*First- BornGirl -0.054-0.002age43*First- BornGirl -0.05-0.012 90 TableA.14(cont'd) (0.033)(0.019)(0.030)*(0.024) age29*First- BornGirl -0.0380.019age44*First- BornGirl -0.0610.007 (0.031)(0.020)(0.036)*(0.026) age30*First- BornGirl -0.053-0.001 (0.034)(0.019) Observations 26,77426,774 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.*at10% level,**at5%level,***at1%level.Bothregressionsareonobser- vationsfrom2-ChildrenProvinces.Allregressionscontrolsagecohortdummies,ethnicity dummies,genderofbirth,mother'sageatbirth,educationlevelsforbothparents, andcountyedeffects. 91 TableA.15:RobustnessCheckoftheEffectofAdditionalChildrenon FemaleLFP (1)(2)(3) OLS2SLS(Triple differenceas IV) 2SLS(Twin- ningasIV) kids2-0.002-0.07-0.069 (0.002)(0.081)(0.040)** Observations 612,785612,749612,749 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.*sig- at10%level,at5%level,at1%level.All regressionscontrolsforagecohortdummies,ethnicitydummies,genderof birth,mother'sageatbirth,educationlevelsforbothparents,andcounty edeffects.Allregressionsarebasedonobservationsfromthe1-boy-2-girl provinces.Col(1)isresultsfromOLS;Col(2)isresultsfrom2SLSusingtriple differenceasinstruments;Col(3)resultsfrom2SLSusingtwinningasinstru- ments. 92 APPENDIXB FIGURESFORCHAPTER1 FigureB.1:FemaleLaborForceParticipationandTotalFertilityRateintheU.S. DataSources:U.S.BureauofLaborStatisticsandWorldBank 93 FigureB.2:FemaleLaborForceParticipationandTotalFertilityRateinChina DataSources:Maurer-Fazioetal.(2005)andtheWorldBank. Measurementoffemalelaborforceparticipationisforthefemalepopulationages15andolder. ThefemaleLFPratesareforbothruralandurbanfemaletogether.AccordingtoMaurer-Fazioetal.(2005),therural femaleLFPare64.0,59.6,and56.3in1982,1990,2000inChina. 94 FigureB.3:TotalFertilityRateinChina DataSource:Yang(2004) Totalfertilityrate(TFR)ofapopulationinaperiodistheaveragenumberofchildrenthatwouldbeborntoawoman overherlifetimeifsheweretoexperiencetheexactcurrentfertilityratesthroughherlifetime,andshe weretosurvivefrombirththroughtheendofherreproductivelife(Yang,2004). 95 a:Ethnicityasinstrumentsb:Genderofbirthasinstruments FigureB.4:CoefofInteractionsofAgeCohortsandEthnicity/First-BornGirl(Heteroge- nousEffectsforWomenwithDifferentEducationLevels) 96 FigureB.5:GirlPercentageforFirst-BornsinDifferentYears(Ages) DataSource:1%samplefromthe1990ChinaPopulationCensus 97 a: nonHan ict d t b: First - BornGirl ict d t FigureB.6:CoefofInteractionsofAgeCohortsandEthnicity/First-BornGirl Fig6a:bluelineŒcoefof nonHan ict d t fromColumn(1),Table7 greenlineŒcoefof nonHan ict d t fromColumn(1),Table12 orangelineŒcoefof nonHan ict d t fromColumn(1),Table9 redlineŒcoefof nonHan ict d t fromColumn(2),Table12 Fig6b:blueline-coefof First - BornGirl ict d t fromColumn(1),Table8 greenlineŒcoefof First - BornGirl ict d t fromColumn(1),Table13 orangelineŒcoefof First - BornGirl ict d t fromColumn(1),Table10 redlineŒcoefof First - BornGirl ict d t fromColumn(2),Table13 98 APPENDIXC TABLESFORCHAPTER2 TableC.1:People'sCharacteristicsbyTheirDisplacementStatus MenWomen NeverAdjustedNeverAdjusted Displaced 1 Displaced 2 Differencesc 3 Displaced 1 Displaced 2 Differences 3 Age20.0529.809.75***20.4629.869.40*** Hispanic0.210.17-0.04***0.210.17-0.04*** Black0.330.27-0.06***0.330.29-0.04** Morethanhighschooleduc 0.260.460.20***0.320.440.11*** Single0.840.39-0.03***0.680.31-0.01 Married0.130.510.03***0.250.510.02 Divorce0.030.11-0.010.060.17-0.01 Familyincomelastyear 18449.5252541.774515.26***18261.9443810.142991.00*** Ageat1stmarriage 25.7725.16-0.61***23.1923.430.23 FertilityinaGivenYear Hadanadditionalchild 0.070.06-0.01**0.100.060.00 Numberofchildren 0.251.18-0.05**0.551.44-0.07** Observations1590208110952893 Note:DataisfromNLSY79.Forselectionrestriction,refertoSection3. 1 Fortime-variantvariables,meansarecalculatedbyusingvaluesthreeormoreyearspriortothedisplacement. 2 Fortime-variantvariables,meansarecalculatedbyusingvaluesofallperson-years. 3 Togettheadjusteddifferences,ageandyearedeffectsarecontrolled. 99 TableC.2:ImpactofDisplacementontheProbabilityofHaving anAdditionalChild(FixedEffect) AllSampleMenWomen Displacementyear-2-0.001-0.0130.015 (0.008)(0.011)(0.013) Displacementyear-100.007-0.011 (0.008)(0.010)(0.011) Displacementyear0.0030.011-0.01 (0.007)(0.010)(0.010) Displacementyear+10.0070.0060.004 (0.007)(0.010)(0.011) Displacementyear+2-30.002-0.0030.003 (0.006)(0.009)(0.009) Displacementyear+4-50.004-0.0050.009 (0.006)-0.009(0.009) Displacementyear+6-70-0.0150.011 (0.006)(0.008)*(0.009) Displacementyear+8+-0.001-0.0160.007 (0.005)(0.008)**(0.008) TotalEffectontheTreated-0.005-0.1100.033 (0.082)(0.126)(0.104) Obersations 156,43975,20581,234 Notes:Standarderrorsclusteredatindividuallevelarereportedinbrackets. at10%level;at5%level;at1% level. Allregressionscontrolsage,agesquare,yearandindividualedeffects. Standarderrorsfortotaleffectonthetreatedareobtainedthroughboot- strap. 100 TableC.3:HeterogeneousImpactofDisplacementontheProbabilityofHavinganAdditionalChild(FixedEffect) MenWithoutMenwithWomenwithoutWomenwith CollegeEducationCollegeEducationCollegeEducationCollegeEducation Displacementyear-2-0.019-0.0030.0070.044 (0.013)(0.017)(0.016)(0.023)* Displacementyear-10.011-0.008-0.0150.012 (0.013)(0.016)(0.014)(0.018) Displacementyear0.0050.022-0.008-0.002 (0.012)(0.017)(0.013)(0.018) Displacementyear+10.0040.0080.0120.005 (0.012)(0.017)(0.013)(0.018) Displacementyear+2-3-0.0090.0170.015-0.002 (0.011)(0.014)(0.011)(0.014) Displacementyear+4-5-0.0050.010.0220.008 (0.011)(0.015)(0.011)**(0.015) Displacementyear+6-7-0.005-0.0140.0230.017 (0.011)(0.013)(0.010)**(0.015) Displacementyear+8+-0.005-0.0030.0220.011 (0.010)(0.012)(0.009)**(0.012) TotalEffectontheTreated-0.015-0.0050.0830.004 (0.106)(0.141)(0.109)(0.149) Obersations 47,28427,92148,43832,796 Notes:Standarderrorsclusteredatindividuallevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,yearandindividualedeffects. Standarderrorsfortotaleffectonthetreatedareobtainedthroughbootstrap. 101 TableC.4:InteractionTermsofCollegeEduca- tionandDisplacementStatus College*Displacementyear+1-0.019 (0.020) College*Displacementyear+2-3-0.029 (0.015)** College*Displacementyear+4-5-0.026 (0.016)* College*Displacementyear+6-7-0.017 (0.016) College*Displacementyear+8+-0.023 (0.012)* F-statistic 4.58 N 81,234 Notes:Standarderrorsclusteredatindividuallevelare reportedinbrackets. at10%level;at5%level; at1%level. Regressioncontrolsage,agesquare,displacementin- dicators,yearandindividualedeffects. 102 TableC.5:ImpactofDisplacementontheProbabilityofHavinganAdditionalChild(Time TrendModel) WomenwithoutWomenwith MenWomenCollegeEducationCollegeEducation Displacementyear-2-0.0170.010.0070.018 (0.012)(0.014)(0.017)(0.025) Displacementyear-1-0.005-0.01-0.009-0.011 (0.013)(0.013)(0.015)(0.023) Displacementyear-0.006-0.0020.008-0.018 (0.013)(0.013)(0.016)(0.023) Displacementyear+1-0.010.0140.03-0.016 (0.014)(0.014)(0.017)*(0.024) Displacementyear+2-3-0.0210.0170.035-0.021 (0.013)(0.013)(0.017)**(0.022) Displacementyear+4-5-0.0240.0260.045-0.011 (0.014)*(0.015)*(0.018)**(0.025) Displacementyear+6-7-0.0350.0320.048-0.002 (0.015)**(0.015)**(0.019)**(0.026) Displacementyear+8+-0.0380.030.048-0.011 (0.016)**(0.017)*(0.021)**(0.027) Obersations 71,53477,24646,06331,183 Notes:Standarderrorsclusteredatindividuallevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,yearandindividualedeffects,interactiontermsofindividual edeffectandyear,andinteractiontermsofindividualedeffectandyearsquare. 103 TableC.6:ImpactofDisplacementontheProbabilityofHavinganAdditionalChild(Corre- latedRandomEffectProbitModel) WomenwithoutWomenwith MenWomenCollegeEducationCollegeEducation Displacementyear-2-0.010.0060.0010.036 (0.007)(0.009)(0.010)(0.022) Displacementyear-10.004-0.012-0.0130.007 (0.008)(0.007)*(0.008)(0.018) Displacementyear0.006-0.013-0.009-0.005 (0.007)(0.007)*(0.008)(0.015) Displacementyear+10.003-0.0040.004-0.001 (0.008)(0.008)(0.010)(0.016) Displacementyear+2-3-0.004-0.0060.006-0.009 (0.006)(0.007)(0.008)(0.012) Displacementyear+4-5-0.006-0.0030.01-0.002 -0.007(0.007)(0.009)(0.014) Displacementyear+6-7-0.014-0.0020.0120.004 (0.006)**(0.008)(0.010)(0.015) Displacementyear+8+-0.016-0.010.008-0.007 (0.006)**(0.007)(0.009)(0.012) TotalEffectontheTreated-0.010-0.0410.029-0.018 (0.010)(0.056)(0.053)(0.041) Obersations 75,20581,23448,25332,796 Notes:Standarderrorsclusteredatindividuallevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,meanofage,meanofagesquare,meanofdisplacementindicators, andyearedeffects. 104 TableC.7:ImpactofDisplacement_ExcludingPeopleSufferingFirstDisplacementAfter1994 WomenwithoutWomenwith MenWomenCollegeEducationCollegeEducation Displacementyear-2-0.0150.0150.0070.044 (0.011)(0.013)(0.016)(0.023)* Displacementyear-10.006-0.011-0.0150.013 (0.010)(0.011)(0.014)(0.018) Displacementyear0.009-0.01-0.007-0.001 (0.010)(0.011)(0.013)(0.018) Displacementyear+10.0030.0040.0140.006 (0.010)(0.011)(0.013)(0.018) Displacementyear+2-3-0.0070.0030.0170.001 (0.009)(0.009)(0.011)(0.014) Displacementyear+4-5-0.010.0150.0330.01 -0.009(0.009)(0.011)***(0.015) Displacementyear+6-7-0.020.0170.0320.021 (0.009)**(0.009)*(0.011)***(0.015) Displacementyear+8+-0.0210.0090.0270.011 (0.008)***(0.008)(0.010)***(0.012) TotalEffectontheTreated-0.1620.0590.1020.024 (0.116)(0.109)(0.141)(0.143) Obersations 64,27468,21939,86228,357 Notes:Standarderrorsclusteredatindividuallevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,yearandindividualedeffects. Standarderrorsfortotaleffectonthetreatedareobtainedthroughbootstrap. 105 TableC.8:ImpactofDisplacementforObseravationsduring1984-1992 WomenwithoutWomenwith MenWomenCollegeEducationCollegeEducation Displacementyear-2-0.0190.0090.0060.033 (0.011)*(0.013)(0.016)(0.024) Displacementyear-1-0.003-0.022-0.0210.003 (0.012)(0.012)*(0.015)(0.021) Displacementyear-0.002-0.029-0.015-0.03 (0.012)(0.013)**(0.016)(0.020) Displacementyear+1-0.008-0.0070.015-0.017 (0.013)(0.014)(0.018)(0.023) Displacementyear+2-3-0.02-0.0120.013-0.022 (0.013)(0.013)(0.017)(0.020) Displacementyear+4-5-0.026-0.0070.027-0.022 (0.014)*(0.016)(0.020)(0.026) Displacementyear+6-7-0.04-0.0020.0290.009 (0.016)**(0.019)(0.023)(0.033) Displacementyear+8+-0.05-0.0260.017-0.033 (0.023)**(0.029)(0.034)(0.053) TotalEffectontheTreated-0.071-0.0130.017-0.008 (0.057)(0.048)(0.595)(0.087) Obersations 28,25130,14117,58612,555 Notes:Standarderrorsclusteredatindividuallevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,yearandindividualedeffects. Standarderrorsfortotaleffectonthetreatedareobtainedthroughbootstrap. 106 TableC.9:ComparisonsbetweenNon-displacedPeopleandPeopleLostaJobDuetoFirmClosure MenWomen NeverAdjustedNeverAdjusted Displaced 1 Displaced 2 Differencesc 3 Displaced 1 Displaced 2 Differences 3 Age20.7429.809.06***20.6129.869.25*** Hispanic0.180.17-0.020.190.17-0.01 Black0.340.27-0.07***0.280.290.01 Morethanhighschooleduc 0.300.460.16***0.330.440.11*** Single0.810.39-0.04**0.660.310.01 Married0.150.510.05***0.260.510.01 Divorce0.040.11-0.01*0.070.17-0.02 Familyincomelastyear 19851.4252541.774699.83***18534.5243810.142719.02*** Ageat1stmarriage 25.6725.16-0.4922.7823.430.67** FertilityinaGivenYear Hadanadditionalchild 0.080.06-0.010.090.060.01 Numberofchildren 0.301.18-0.030.521.44-0.03 Observations 44520813772893 Note:DataisfromNLSY79.Forselectionrestriction,refertoSection4. 1 Fortime-variantvariables,meansarecalculatedbyusingvaluesthreeormoreyearspriortothedisplacement. 2 Fortime-variantvariables,meansarecalculatedbyusingvaluesofallperson-years. 3 Togettheadjusteddifferences,ageandyearedeffectsarecontrolled. 107 TableC.10:ImpactofJobLossDuetoFirmClosureontheProbabilityofHavinganAddi- tionalChild(FixedEffect) WomenwithoutWomenwith MenWomenCollegeEducationCollegeEducation Displacementyear-2-0.0010.0110.0130.014 (0.017)(0.020)(0.025)(0.034) Displacementyear-10.014-0.014-0.002-0.039 (0.018)(0.016)(0.021)(0.022)* Displacementyear0.028-0.018-0.008-0.031 (0.018)(0.015)(0.019)(0.024) Displacementyear+10.015-0.019-0.018-0.008 (0.018)(0.014)(0.017)(0.027) Displacementyear+2-30.01-0.0040.008-0.017 (0.013)(0.011)(0.014)(0.021) Displacementyear+4-500.0180.0340 (0.012)(0.012)(0.016)**(0.020) Displacementyear+6-7-0.0140.0070.0160.001 (0.011)(0.011)(0.012)(0.021) Displacementyear+8+-0.0120.010.0160.013 (0.008)(0.008)(0.009)*(0.014) TotalEffectontheTreated-0.0200.0380.0540.016 (0.095)(0.085)(0.101)(0.098) Obersations 53,97667,57839,35028,228 Notes:Standarderrorsclusteredatindividuallevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,yearandindividualedeffects. 108 TableC.11:HeterogeneousImpactofDisplacementontheProbabilityofHavinganAdditionalChild(Fixed EffectPropensityScoreMatching) MenWithoutMenwithWomenwithoutWomenwith CollegeEducationCollegeEducationCollegeEducationCollegeEducation Displacementyear-20.006-0.013-0.0370.012 (0.012)(0.011)(0.014)***(0.019) Displacementyear-1-0.007-0.021-0.0290.010 (0.01)(0.01)**(0.013)**(0.022) Displacementyear0.036-0.012-0.011-0.026 (0.011)***(0.009)(0.013)(0.016) Displacementyear+1-0.005-0.005-0.015-0.014 (0.01)(0.01)(0.012)(0.019) Displacementyear+2-3-0.002-0.007-0.007-0.009 (0.007)(0.007)(0.008)(0.013) Displacementyear+4-5-0.0050.0050.015-0.010 (0.008)(0.006)(0.008)*(0.012) Displacementyear+6-7-0.0170.0090.0180.009 (0.007)**(0.007)(0.008)**(0.011) Displacementyear+8+-0.0010.0030.010-0.003 (0.003)(0.002)(0.003)***(0.005) TotalEffectontheTreated-0.035-0.0030.007-0.040 (0.030)(0.007)(0.006)(0.045) Note:DataisfromNLSY79. Standarderrorsareobtainedthroughbootstrapandreportedintheparenthesis. at10%level;at5%level;at1%level. 109 APPENDIXD FIGURESFORCHAPTER2 FigureD.1:TheU.S.FertilityRateHasFallenDuringRecessions DataSources:OECDDemographyData. 110 FigureD.2:UnemploymentandBirthRateintheU.S. DataSources:NationalVitalStatisticsSystemandBureauofLaborStatistics. 111 FigureD.3:EffectsofWageDecreaseforWomenwithVeryHighWage Notes:Theoriginfortwodifferencecurvesisatthenortheasterncorner,withn-axisdenotingthehorizontalaxis. 112 FigureD.4:DisplacementRatefromNLSY79Vs.OfUnemploymentintheU.S.(1984-1994) DataSources:NLSY79andBureauofLaborStatistics 113 FigureD.5:DisplacementRateforDifferentAgeCohorts.(1984-1994) DataSources:NLSY79 114 FigureD.6:DisplacementandBirthRateforMen Notes:Fordisplacedmen,thex-axisdenotestimebeforeandafterjobdisplacement.Fornon-displacedmen,the x-axisdenotestimebeforeandafterafakejobdisplacement.Theyearofthefakedisplacementisgeneratedby randomizationwiththeprobabilityforeachyearbasedonthedistributionofoccurrenceratesofdisplacementforthe displacedgroup. 115 FigureD.7:DisplacementandBirthRateforWomen Notes:Fordisplacedwomen,thex-axisdenotestimebeforeandafterjobdisplacement.Fornon-displacedwomen, thex-axisdenotestimebeforeandafterafakejobdisplacement.Theyearofthefakedisplacementisgeneratedby randomizationwiththeprobabilityforeachyearbasedonthedistributionofoccurrenceratesofdisplacementforthe displacedgroup. 116 FigureD.8:DisplacementandBirthRateforWomenwithCollegeEducation Notes:Fordisplacedwomen,thex-axisdenotestimebeforeandafterjobdisplacement.Fornon-displacedwomen, thex-axisdenotestimebeforeandafterafakejobdisplacement.Theyearofthefakedisplacementisgeneratedby randomizationwiththeprobabilityforeachyearbasedonthedistributionofoccurrenceratesofdisplacementforthe displacedgroup. 117 FigureD.9:DisplacementandBirthRateforWomenwithoutCollegeEducation Notes:Fordisplacedwomen,thex-axisdenotestimebeforeandafterjobdisplacement.Fornon-displacedwomen, thex-axisdenotestimebeforeandafterafakejobdisplacement.Theyearofthefakedisplacementisgeneratedby randomizationwiththeprobabilityforeachyearbasedonthedistributionofoccurrenceratesofdisplacementforthe displacedgroup. 118 FigureD.10:PropensityHistogrambyDisplacementStatusin1984_Men DataSources:NLSY79 119 APPENDIXE TABLESFORCHAPTER3 TableE.1:SummaryStatisticsforKeyVariables (1)(2)(3) Men0.4681.0000.000 (0.005)(0)(0) CES-D8.8727.7649.845 (0.065)(0.088)(0.093) CES-D 100.3980.3240.463 (0.005)(0.007)(0.007) #ofChildren2.9562.8673.034 (0.015)(0.022)(0.022) #ofChildren > 10.8860.8760.894 (0.003)(0.005)(0.004) First-BornGirl0.4740.4770.472 (0.005)(0.007)(0.007) Age58.59159.06158.178 (0.094)(0.135)(0.13) Ageat1stBirth24.02425.16223.024 (0.041)(0.062)(0.051) YearofSchooling4.6776.1253.404 (0.04)(0.054)(0.053) AtLeastPrimarySchool0.5040.6660.361 (0.005)(0.007)(0.007) #ofSiblings3.9213.8383.994 (0.02)(0.029)(0.027) Married0.8490.8940.810 (0.004)(0.005)(0.005) GoodHealthduringChildhood0.7450.7510.740 (0.004)(0.006)(0.006) Observations 965745175140 Notes:Dataisfrom2011CHARLS. Col(1)isonthewholesample;Col(2)isformenonly;Col(3)isfor womenonly. 120 TableE.2:DIDEstimatesRegardingGenderofFirstBirth ProbabilityofHaving2orMoreChildren OldCohortsYoungCohortsDifference First-BornBoy0.9470.802-0.145 (s.d./s.e.)0.22330.39860.0094 First-BornGirl0.9570.872-0.084 (s.d./s.e.)0.20370.29700.0075 Difference0.0090.0700.061 (s.e.)0.00670.00820.0121 NumberofChildren First-BornBoy3.9812.230-1.751 (s.d./s.e.)1.68331.00640.0351 First-BornGirl4.1342.587-1.546 (s.d./s.e.)1.67771.14020.0380 Difference0.1520.3570.205 (s.e.)0.05270.02500.0516 CES-D-10Score First-BornBoy9.9758.315-1.660 (s.d./s.e.)6.56076.26280.1858 First-BornGirl9.5098.571-0.938 (s.d./s.e.)6.48736.38160.1924 Difference-0.4660.2560.722 (s.e.)0.21970.15680.2674 Notes:OldCohortsincludeindividuals62yearsoldandabove;Young Cohortsincludeindividualsbelow62. 121 TableE.3:FirstStageResultsforTwoorMoreChildren (1)(2)(3) Age < 62*First-BornGirl0.1020.1010.106 (0.017)***(0.019)***(0.020)*** First-BornGirl0.0080.0080.007 (0.009)(0.011)(0.012) Age < 62-0.231-0.170.578 (0.033)***(0.037)***(0.057)*** Men0.009 (0.006) YearofSchooling-0.001-0.0030.001 (0.001)(0.001)**(0.002) #ofSiblings0.0020.0020.002 (0.002)(0.002)(0.003) TemporarySeperated0.0110.0080.009 (0.016)(0.019)(0.025) Seperated-0.163-0.016-0.251 (0.062)***(0.062)(0.085)*** Divorced-0.191-0.207-0.19 (0.078)**(0.140)(0.090)** Widowed-0.03-0.007-0.045 (0.010)***(0.010)(0.019)** Nevermarried-0.609-0.104-0.696 (0.104)***(0.273)(0.087)*** AgeCohortFEYYY Cragg-DonaldWaldF 84.6239.2443.23 R-squared 0.120.110.14 Observations 9,6575,1404,517 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Col(1)isforthewholesample;Col(2)isformenonly;Col(3)isforwomen only. 122 TableE.4:FirstStageResultsforNumberofChildren (1)(2)(3) Age < 62*First-BornGirl0.2630.2620.256 (0.059)***(0.079)***(0.070)*** First-BornGirl0.1230.1210.141 (0.055)**(0.074)(0.063)** Age < 62-4.011-3.7790.209 (0.095)***(0.134)***(0.169) Men-0.046 (0.021)** YearofSchooling0.0030.0040.001 (0.004)(0.004)(0.006) #ofSiblings0.00100.002 (0.006)(0.009)(0.009) TemporarySeperated-0.0120.004-0.014 (0.055)(0.061)(0.089) Seperated-0.1050.307-0.433 (0.201)(0.297)(0.277) Divorced-0.642-0.977-0.477 (0.201)***(0.378)**(0.202)** Widowed-0.080.001-0.296 (0.054)(0.063)(0.077)*** Nevermarried-1.748-0.87-1.917 (0.306)***(0.155)***(0.412)*** AgeCohortFEYYY Cragg-DonaldWaldF 33.415.3318.53 R-squared 0.450.470.44 Observations 9,6575,1404,517 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Col(1)isforthewholesample;Col(2)isformenonly;Col(3)isforwomen only. 123 TableE.5:OLSResultsforEffectsofFertilityonParent'sMentalHealth TwoorMoreChildrenNumberofChildren (1)(2)(3)(4)(5)(6) Fertility0.2140.0830.320.0990.0590.126 (0.236)(0.322)(0.341)(0.068)(0.086)(0.089) AgeatFirstBirth0.0350.065-0.0180.0430.07-0.008 (0.018)*(0.024)***(0.026)(0.018)**(0.024)***(0.026) Men-1.74-1.747 (0.135)***(0.133)*** YearofSchooling-0.15-0.187-0.105-0.15-0.193-0.103 (0.021)***(0.030)***(0.032)***(0.021)***(0.029)***(0.032)*** #ofSiblings0.0310.068-0.0050.0250.073-0.02 (0.036)(0.046)(0.055)(0.035)(0.045)(0.055) TemporarySeperated0.5931.2190.3090.5181.0910.271 (0.310)*(0.529)**(0.364)(0.305)*(0.518)**(0.363) Seperated5.7356.73.7634.8465.413.489 (1.294)***(1.494)***(2.261)*(1.146)***(1.460)***(2.111)* Divorced3.8865.3651.3433.9065.3991.442 (1.112)***(1.379)***(1.772)(1.100)***(1.371)***(1.802) Widowed1.3251.6431.3321.3371.7081.306 (0.272)***(0.480)***(0.313)***(0.267)***(0.472)***(0.313)*** Nevermarried1.8482.5671.6631.541.6641.812 (1.903)(3.137)(2.296)(1.532)(1.771)(2.134) R-squared 0.070.050.030.070.050.03 Observations 9,4624,4065,0569,6574,5175,140 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)andCol(4)areonthewholesample;Col(2)andCol(5)areformenonly;Col(3)andCol(6)areforwomenonly. 124 TableE.6:2SLSResultsforEffectsofFertilityonParent'sMentalHealth TwoorMoreChildrenNumberofChildren (1)(2)(3)(4)(5)(6) Fertility8.3284.9259.8343.0211.9013.33 (2.980)***(3.486)(4.197)**(1.255)**(1.506)(1.602)** AgeatFirstBirth0.1440.1340.1060.3240.240.307 (0.043)***(0.054)**(0.061)*(0.122)***(0.141)*(0.161)* Men-1.85-1.609 (0.144)***(0.153)*** YearofSchooling-0.14-0.189-0.077-0.162-0.198-0.117 (0.023)***(0.031)***(0.038)**(0.025)***(0.032)***(0.036)*** #ofSiblings0.0190.064-0.0240.0310.068-0.008 (0.037)(0.048)(0.058)(0.041)(0.048)(0.064) TemporarySeperated0.5451.1610.3030.5281.110.231 (0.324)*(0.555)**-0.383(0.348)(0.530)**(0.427) Seperated6.6597.63.5775.0076.0952.303 (1.451)***(1.674)***(2.324)(1.312)***(1.558)***(2.418) Divorced5.5626.3693.4255.8256.4264.684 (1.453)***(1.734)***(2.549)(1.516)***(1.709)***(3.221) Widowed1.4831.8091.3651.5692.2891.268 (0.280)***(0.505)***(0.322)***(0.328)***(0.684)***(0.372)*** Nevermarried4.5124.7962.796.6315.2234.551 (2.629)*(3.714)(4.573)(2.579)**(3.323)(3.015) R-squared -0.050-0.11-0.19-0.06-0.27 Observations 9,4414,3735,0339,6374,4845,118 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)andCol(4)areonthewholesample;Col(2)andCol(5)areformenonly;Col(3)andCol(6)areforwomenonly. 125 TableE.7:2SLSEstimatesofEffectsofTwoorMoreChildrenonParent'sMentalHealthbyParents' Education LessThanPrimarySchoolAtLeastPrimarySchool (1)(2)(3)(4)(5)(6) #ofChildren > 16.59310.5976.4017.4275.8669.601 (6.675)(9.082)(8.598)(3.460)**(4.689)(5.069)* AgeatFirstBirth0.1180.1690.0750.140.1480.091 (0.077)(0.098)*(0.103)(0.063)**(0.085)*(0.103) Men-1.931-1.816 (0.233)***(0.182)*** YearofSchooling0.033-0.1050.082-0.187-0.241-0.062 (0.069)(0.129)(0.092)(0.047)***(0.061)***(0.084) #ofSiblings-0.050.001-0.0570.0760.0950.075 (0.062)(0.095)(0.087)(0.049)(0.062)(0.095) TemporarySeperated0.8311.770.5380.2640.5570.064 (0.406)**(0.875)**(0.495)(0.523)(0.774)(0.714) Seperated5.9719.4992.548.1677.45913.518 (1.695)***(2.558)***(2.596)(2.301)***(2.536)***(0.880)*** Divorced5.1555.0834.6826.1846.8420.903 (2.500)**(4.643)(3.196)(1.838)***(2.046)***(1.754) Widowed1.3082.0831.1891.9681.8442.121 (0.367)***(0.796)***(0.400)***(0.461)***(0.727)**(0.676)*** Nevermarried3.7554.3722.537.8297.026 (2.835)(5.276)(3.810)(3.741)**(4.961) R-squared -0.02-0.13-0.03-0.06-0.04-0.15 Observations 4,6701,4243,2074,7412,9081,769 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)andCol(4)areonthewholesample;Col(2)andCol(5)areformenonly;Col(3)andCol(6)areforwomenonly. 126 TableE.8:2SLSEstimatesofEffectsofNumberofChildrenonParent'sMentalHealthbyParents' Education LessThanPrimarySchoolAtLeastPrimarySchool (1)(2)(3)(4)(5)(6) #ofChildren1.4392.5671.5034.6593.5175.329 (1.415)(2.192)(1.871)(2.584)*(5.999)(3.270) AgeatFirstBirth0.1830.310.1450.4660.5770.456 (0.136)(0.201)(0.186)(0.252)*(0.580)(0.314) Men-1.737-1.679 (0.241)***(0.231)*** YearofSchooling0.044-0.0840.095-0.181-0.196-0.109 (0.067)(0.131)(0.096)(0.053)***(0.083)**(0.100) #ofSiblings-0.034-0.074-0.0140.0690.14-0.138 (0.057)(0.101)(0.074)(0.056)(0.081)*(0.130) TemporarySeperated0.7041.8180.4150.1540.014-0.302 (0.415)*(0.968)*(0.523)(0.525)(1.099)(0.804) Seperated4.798.1481.8096.756.60813.524 (1.531)***(2.282)***(2.839)(2.304)***(3.356)**(7.316)* Divorced4.6794.7434.7767.4088.5962.57 (2.204)**(5.018)(3.326)(2.026)***(3.546)**(5.005) Widowed1.4442.9491.2151.9383.2650.705 (0.408)***(1.104)***(0.425)***(0.479)***(2.048)(1.165) Nevermarried3.8637.2442.7988.4048.909 (3.110)(5.844)(2.883)(4.532)*(7.373) R-squared -0.03-0.26-0.05-0.46-0.83-0.5 Observations 4,7651,4703,2574,8422,9741,805 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)andCol(4)areonthewholesample;Col(2)andCol(5)areformenonly;Col(3)andCol(6)areforwomen only. 127 TableE.9:2SLSResultsforEffectsofFertilityonParent'sSelf-ReportedHealth TwoorMoreChildrenNumberofChildren (1)(2)(3)(4)(5)(6) Fertility-0.664-0.148-0.981-0.275-0.094-0.358 (0.374)*(0.503)(0.518)*(0.154)*(0.205)(0.191)* AgeatFirstBirth-0.011-0.004-0.015-0.028-0.009-0.037 (0.006)*(0.008)(0.008)*(0.015)*(0.019)(0.019)* Men0.1540.129 (0.020)***(0.021)*** YearofSchooling0.010.0150.0010.0120.0160.005 (0.003)***(0.004)***-0.005(0.003)***(0.004)***(0.004) #ofSiblings0.006-0.0040.0140.005-0.0020.01 (0.005)(0.007)(0.007)**(0.005)(0.007)(0.007) TemporarySeperated0.002-0.040.0060.001-0.0270.003 (0.041)(0.065)(0.053)(0.041)(0.064)(0.054) Seperated-0.326-0.319-0.067-0.228-0.255-0.013 (0.134)**(0.197)(0.163)(0.112)**(0.179)(0.190) Divorced-0.0060.131-0.19-0.0490.114-0.33 (0.185)(0.235)(0.242)(0.187)(0.237)(0.256) Widowed0.0550.1340.0120.0350.0980.01 (0.032)*(0.060)**(0.040)(0.035)(0.089)(0.043) Nevermarried-0.477-0.346-0.193-0.874-0.587-0.374 (0.286)*(0.242)(0.136)(0.326)***(0.414)(0.324) R-squared 0.010.04-0.06-0.050.03-0.15 Observations 9,4394,3725,0329,6354,4835,117 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)andCol(4)areonthewholesample;Col(2)andCol(5)areformenonly;Col(3)andCol(6)areforwomen only. 128 TableE.10:2SLSEstimatesofEffectsofFertilityonDiagnosisofChronicDiseases ArthritisorRheumatismHypertensionDigestiveDiseases #ofChildren > 1#ofChildren#ofChildren > 1#ofChildren#ofChildren > 1#ofChildren AllSample 0.3460.127-0.115-0.031-0.016-0.013 (0.222)(0.084)(0.186)(0.071)(0.175)(0.067) MenOnly 0.1280.0360.0520.038-0.032-0.024 (0.273)(0.109)(0.259)(0.105)(0.240)(0.098) WomenOnly 0.2790.109-0.093-0.022-0.089-0.037 (0.297)(0.104)(0.267)(0.095)(0.267)(0.096) Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. Allregressionscontrolsage,agesquare,genderofbirth,ageatbirth,sex,education,numberofsiblings,marital status,self-reportedhealthduringchildhood,andcountyedeffects. 129 TableE.11:2SLSResultsforEffectsofFertilityonParent'sMentalHealth(ControllingSelf-Reported Health) TwoorMoreChildrenNumberofChildren (1)(2)(3)(4)(5)(6) Fertility6.7734.6577.2372.4041.7422.426 (2.685)**(3.293)(3.743)*(1.111)**(1.433)(1.431)* AgeatFirstBirth0.1180.1270.0670.2610.2240.214 (0.038)***(0.051)**(0.054)(0.107)**(0.134)*(0.142) Men-1.497-1.325 (0.135)***(0.145)*** YearofSchooling-0.115-0.158-0.073-0.134-0.166-0.103 (0.021)***(0.029)***(0.035)**(0.023)***(0.030)***(0.033)*** #ofSiblings0.0320.0560.0130.0420.0650.018 (0.035)(0.045)(0.054)(0.038)(0.045)(0.059) TemporarySeperated0.551.0790.3170.531.0550.237 (0.300)*(0.513)**(0.352)(0.322)(0.501)**(0.388) Seperated5.9076.9543.4064.5025.5892.281 (1.426)***(1.778)***(2.000)*(1.306)***(1.638)***(2.109) Divorced5.5466.6442.9245.7166.6783.851 (1.325)***(1.565)***(2.488)(1.399)***(1.565)***(3.085) Widowed1.6212.0851.4131.6642.4981.318 (0.264)***(0.470)***(0.301)***(0.300)***(0.641)***(0.333)*** Nevermarried3.4054.1042.2854.6824.0913.616 (2.540)(3.669)(4.859)(2.343)**(3.194)(3.580) Self-ReportedHealth-2.309-2.048-2.616-2.232-2.039-2.487 (0.089)***(0.122)***(0.115)***(0.097)***(0.127)***(0.137)*** R-squared 0.090.10.0700.05-0.02 Observations 9,4394,3725,0329,6354,4835,117 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)andCol(4)areonthewholesample;Col(2)andCol(5)areformenonly;Col(3)andCol(6)areforwomenonly. 130 TableE.12:2SLSResultsforEffectsofFertilityonCo-residencewithChildren TwoorMoreChildrenNumberofChildren (1)(2)(3)(4)(5)(6) Fertility-0.281-0.261-0.276-0.106-0.101-0.098 (0.246)(0.290)(0.294)(0.094)(0.115)(0.107) AgeatFirstBirth0.0070.0070.0080.0010.0020.002 (0.004)*(0.004)*(0.005)*(0.009)(0.011)(0.011) Men-0.002-0.009 (0.008)(0.009) YearofSchooling-0.003-0.005-0.001-0.002-0.0050 (0.002)(0.002)*(0.003)(0.002)(0.002)**(0.002) #ofSiblings00.003-0.002-0.0010.003-0.003 (0.003)(0.004)(0.004)(0.003)(0.004)(0.004) TemporarySeperated0.006-0.0390.030.002-0.0430.028 (0.028)(0.045)(0.032)(0.028)(0.047)(0.032) Seperated-0.18-0.252-0.022-0.14-0.2250.018 (0.087)**(0.116)**(0.109)(0.083)*(0.108)**(0.112) Divorced-0.152-0.159-0.159-0.168-0.166-0.201 (0.086)*(0.105)(0.121)(0.096)*(0.110)(0.153) Widowed0.0950.0510.1170.0950.0290.122 (0.021)***(0.032)(0.023)***(0.021)***(0.047)(0.023)*** Nevermarried-0.301-0.351-0.134-0.316-0.365-0.194 (0.169)*(0.229)(0.093)(0.181)*(0.248)(0.100)* R-squared 0.010.010.020.030.020.02 Observations 9,5904,4605,0959,5904,4605,095 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)andCol(4)areonthewholesample;Col(2)andCol(5)areformenonly;Col(3)andCol(6)areforwomenonly. 131 TableE.13:2SLSResultsforEffectsofFertilityonParent'sMentalHealth(ControllingLivingArrange- ments) TwoorMoreChildrenNumberofChildren (1)(2)(3)(4)(5)(6) Fertility8.1625.0659.4212.9141.8823.188 (2.992)***(3.492)(4.244)**(1.239)**(1.458)(1.608)** AgeatFirstBirth0.140.1360.10.3190.2430.302 (0.041)***(0.052)***(0.058)*(0.120)***(0.138)*(0.162)* Men-1.833-1.617 (0.142)***(0.150)*** YearofSchooling-0.141-0.188-0.084-0.162-0.195-0.124 (0.023)***(0.031)***(0.038)**(0.024)***(0.032)***(0.036)*** #ofSiblings0.0160.064-0.0290.0270.068-0.012 (0.038)(0.048)(0.059)(0.041)(0.048)(0.064) TemporarySeperated0.5291.110.2860.521.0730.231 (0.327)(0.562)**(0.383)(0.347)(0.534)**(0.427) Seperated6.5897.5783.5244.91962.319 (1.449)***(1.671)***(2.329)(1.298)***(1.540)***(2.406) Divorced5.4946.3683.2545.7186.3944.447 (1.448)***(1.730)***(2.531)(1.496)***(1.688)***(3.180) Widowed1.5061.8611.3611.5962.351.254 (0.282)***(0.509)***(0.323)***(0.326)***(0.681)***(0.366)*** Nevermarried4.1944.7972.2266.3615.1343.893 (2.633)(3.694)(4.256)(2.554)**(3.230)(2.769) Self-ReportedHealth-0.335-0.348-0.272-0.495-0.459-0.394 (0.177)*(0.206)*(0.216)(0.204)**(0.243)*(0.251) R-squared 0.0400.10.170.060.24 Observations 9,3984,3515,0129,5904,4605,095 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)andCol(4)areonthewholesample;Col(2)andCol(5)areformenonly;Col(3)andCol(6)areforwomenonly. 132 APPENDIXF FIGURESFORCHAPTER3 FigureF.1:NumberofChildrenandParent'sCES-DScore DataSources:ChinaHealthandRetirementLongitudinalSurvey(2011) 133 FigureF.2:NumberofChildrenandParent'sVignetteQuestionScore DataSources:ChinaHealthandRetirementLongitudinalSurvey(2011) 134 FigureF.3:CES-DScoreandVignetteQuestionScore DataSources:ChinaHealthandRetirementLongitudinalSurvey(2011) 135 APPENDIXG APPENDICESFORCHAPTER1 TableG.1:CoefofInteractionTermsforFirstStageandReducedFormwhenFocusingon Mothers 30 (1)(2)(1)(2) First Stage Reduced Form First Stage Reduced Form age30*non- Han 0.079-0.012age40*First- BornGirl 0.043-0.012 (0.014)***(0.006)**(0.006)***(0.004)*** age31*non- Han 0.089-0.014age31*First- BornGirl 0.058-0.01 (0.011)***(0.006)**(0.006)***(0.004)*** age32*non- Han 0.088-0.005age32*First- BornGirl 0.06-0.007 (0.010)***(0.005)(0.006)***(0.003)** age33*non- Han 0.082-0.012age33*First- BornGirl 0.062-0.01 (0.010)***(0.005)**(0.006)***(0.003)*** age34*non- Han 0.072-0.013age34*First- BornGirl 0.055-0.007 (0.009)***(0.005)***(0.006)***(0.003)** age35*non- Han 0.058-0.012age35*First- BornGirl 0.047-0.007 (0.008)***(0.005)***(0.005)***(0.003)** age36*non- Han 0.052-0.01age36*First- BornGirl 0.036-0.009 (0.007)***(0.005)*(0.004)***(0.003)*** age37*non- Han 0.033-0.002age37*First- BornGirl 0.022-0.005 (0.006)***(0.005)(0.004)***(0.004) age38*non- Han 0.029-0.016age38*First- BornGirl 0.015-0.006 (0.006)***(0.004)***(0.003)***(0.003)* age39*non- Han 0.014-0.003age39*First- BornGirl 0.011-0.005 (0.006)**(0.005)(0.003)***(0.004) Observations 475,344475,344 Observations 339,118339,118 136 TableG.1(cont'd) Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.at 10%level;at5%level;at1%level.Col(1)and(2)areon observationsfromrestrictedprovinces(1-ChildProvincesand1-Boy-2-GirlProvinces); Col(3)and(4)areonobservationsfrom1-Boy-2-GirlProvinces.Allregressionscontrols agecohortdummies,ethnicitydummies,genderofbirth,mother'sageatbirth, educationlevelsforbothparents,countyedeffects. 137 TableG.2:OLSand2SLSEstimatesoftheEffectofAdditionalChildrenonFemale LFPwhenFocusingonMothers 30 (1)(2)(3)(4) OLS2SLSOLS2SLS kids20.01-0.1210.009-0.129 (0.002)***(0.046)***(0.003)***(0.037)*** non-Han0.0090.011 (0.008)(0.008) First-BornGirl-0.0010.008-0.0010.006 (0.001)(0.003)***(0.001)(0.002)*** Cragg-DonaldWaldFstatistic46.88680.629 Observations 475,344475,304339,118339,085 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.at10%level; at5%level;at1%level.Allregressionscontrolsagecohortdummies, mother'sageatbirth,educationlevelsforbothparents,countyedeffects.Col(1)and(2)are onobservationsin1-ChildProvincesand1-Boy-2-GirlProvinces,usinginteractionsofagecohort andethnicityasinstruments;Col(3)and(4)areonHanpeopleonlyin1-Boy-2-GirlProvinces,using interactionsofagecohortandgenderofbirthasinstruments. 138 TableG.3:FertilityonFemaleLFP: T i non-HanasIV (1)(2)(3)(4) FirstStageReducedFormOLS2SLS kids20-0.255 (0.001)(0.090)*** non-Han-0.1370.040.0030.006 (0.039)***(0.012)***(0.008)(0.008) First-BornGirl0.059-0.001-0.0010.015 (0.003)***(0.000)(0.000)(0.005)*** T i non-Han0.166-0.042 (0.044)***(0.010)*** Observations 778,754778,754778,754778,710 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.*sig- at10%level,**at5%level,***at1%level.All regressionscontrolsagecohortdummies,mother'sageatbirth,educationlev- elsforbothparents,countyedeffects.Allregressionsonobservationsfrom 1-ChildProvincesand1-Boy-2-GirlProvinces. FollowWuandLi(2012),Iuseinteractionofgirldummyandameasure oftimeexposuretoOCPasaninstrument.Thevalueof T i canbeexpressedas below, T i = 8 > < > : 1 forwomenyoungerthan 18 at 1979 55 ageat 1979 55 18 forwomenbetIen 18 and 55 at 1979 0 forwomenolderthan 55 at 1979 139 TableG.4:FertilityonFemaleLFP: T i First-BornGirlasIV (1)(2)(3)(4) FirstStageReducedFormOLS2SLS kids2-0.002-0.86 (0.002)(0.429)** non-Han0.0110.0020.0020.011 (0.004)**(0.008)(0.008)(0.010) First-BornGirl0.0240.016-0.0010.037 (0.008)***(0.005)***(0.001)*(0.019)** T i First-BornGirl0.023-0.019 (0.010)**(0.005)*** Observations 612,785612,785612,785612,749 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.* at10%level,**at5%level,***sat1%level.Allregres- sionscontrolsagecohortdummies,mother'sageatbirth,educationlevelsfor bothparents,countyedeffects.Allregressionsonobservationsfrom1-Boy-2-Girl Provinces. FollowWuandLi(2012),Iuseinteractionofgirldummyandameasureof timeexposuretoOCPasaninstrument.Thevalueof T i canbeexpressedasbelow, T i = 8 > < > : 1 forwomenyoungerthan 18 at 1979 55 ageat 1979 55 18 forwomenbetIen 18 and 55 at 1979 0 forwomenolderthan 55 at 1979 140 TableG.5:FertilityonFemaleLFP:DIDBasedonEthnicityasIV OLS2SLSFirstStageReducedForm kids20.005-0.139 (0.002)***(0.044)*** age29*non-Han0.083-0.03 (0.016)***(0.011)*** age30*non-Han0.08-0.03 (0.015)***(0.010)*** age31*non-Han0.088-0.032 (0.014)***(0.011)*** age32*non-Han0.085-0.023 (0.013)***(0.010)** age33*non-Han0.08-0.03 (0.013)***(0.010)*** age34*non-Han0.07-0.031 (0.012)***(0.010)*** age35*non-Han0.057-0.03 (0.012)***(0.010)*** age36*non-Han0.051-0.028 (0.011)***(0.010)*** age37*non-Han0.031-0.019 (0.011)***(0.010)* age38*non-Han0.028-0.033 (0.011)***(0.010)*** age39*non-Han0.013-0.021 (0.010)(0.010)** age40*non-Han0.005-0.025 (0.010)(0.010)** age41*non-Han-0.003-0.028 (0.010)(0.010)*** age42*non-Han-0.012-0.01 (0.010)(0.011) age43*non-Han-0.003-0.014 (0.010)(0.010) age44*non-Han-0.002-0.011 (0.011)(0.010) Observations 593,411593,358593,411593,411 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets.*at 10%level,**at5%level,***at1%level.Allregressionsareon observationsfromrestrcitedprovinces(1-childprovincesand1-boy-2-girlprovinces), excludingmothersyoungerthan27.Allregressionscontrolsagecohortdummies,eth- nicitydummies,genderofbirth,mother'sageatbirth,educationlevelsforboth parents,countyedeffects. 141 TableG.6:CoefofTrippleInteractionTerms (1)(2)(1)(2) First Stage Reduced Form First Stage Reduced Form age16*non-Han*First- BornGirl 0.058age31*non-Han*First- BornGirl -0.0350.005 (0.012)***(0.014)**(0.010) age17*non-Han*First- BornGirl -0.047age32*non-Han*First- BornGirl -0.022-0.011 (0.094)(0.012)*(0.007) age18*non-Han*First- BornGirl 0.036age33*non-Han*First- BornGirl -0.0210.006 (0.157)(0.012)*(0.009) age19*non-Han*First- BornGirl -0.04age34*non-Han*First- BornGirl -0.028-0.002 (0.087)(0.013)**(0.010) age20*non-Han*First- BornGirl -0.079age35*non-Han*First- BornGirl -0.045-0.012 (0.048)*(0.011)***(0.008) age21*non-Han*First- BornGirl 0.002age36*non-Han*First- BornGirl -0.0280.007 (0.035)(0.010)***(0.008) age22*non-Han*First- BornGirl -0.009age37*non-Han*First- BornGirl -0.015-0.003 (0.026)(0.010)(0.008) age23*non-Han*First- BornGirl 0.039-0.004age38*non-Han*First- BornGirl -0.0310.015 (0.025)(0.011)(0.009)***(0.008)* age24*non-Han*First- BornGirl -0.0150.011age39*non-Han*First- BornGirl -0.011-0.01 (0.019)(0.009)(0.011)(0.011) age25*non-Han*First- BornGirl -0.009-0.01age40*non-Han*First- BornGirl -0.01-0.011 (0.017)(0.008)(0.011)(0.011) age26*non-Han*First- BornGirl -0.0370.013age41*non-Han*First- BornGirl 0.0010.003 (0.015)**(0.008)*(0.011)(0.011) age27*non-Han*First- BornGirl -0.0220.005age42*non-Han*First- BornGirl 0.0120.011 (0.014)(0.006)(0.013)(0.013) age28*non-Han*First- BornGirl 0.0020.01age43*non-Han*First- BornGirl 0.009-0.029 142 TableG.6(cont'd) (0.016)(0.009)(0.013)(0.014)** age29*non-Han*First- BornGirl 0.020.004age44*non-Han*First- BornGirl -0.0090.002 (0.020)(0.012)(0.015)(0.014) age30*non-Han*First- BornGirl 0.0080.007 (0.016)(0.011) Observations 612,785612,785 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. *at10%level,**at5%level,***at1%level. Allregressionscontrolsagecohortdummies,ethnicity,genderofbirth,interactionsofageandethnicity, interactionsofageandgenderofbirth,mother'sageatbirth,educationlevelsforbothparents,county edeffects. 143 FigureG.1:Correlationbetweennumberofchildrenandlaborforceparticipationrate DataSource:1%samplefrom1990ChinaPopulationCensus. 144 FigureG.2:Averagenumberofchildrenforwomenatdifferentage. DataSource:1%samplefromthe1990ChinaPopulationCensus. 145 a:Restrictedprovincesb:2-Childrenprovinces FigureG.3:CumulativeDistributionofAgeWhenGivingSecondBirth DataSource:1%samplefromthe1990ChinaPopulationCensus 146 FigureG.4:SexRatioatBirthandAbortionRateinChina(1978-1991) PictureSource:Chenetal.(2013) 147 APPENDIXH APPENDICESFORCHAPTER2 TableH.1:ImpactofDisplacementontheProbabilityofHavinganAdditionalChild(First Difference) WomenwithoutWomenwith MenWomenCollegeEducationCollegeEducation Displacementyear-2-0.0230.0060.0110.010 (0.016)(0.017)(0.022)(0.032) Displacementyear-1-0.005-0.024-0.006-0.027 (0.016)(0.016)(0.022)(0.030) Displacementyear-0.004-0.0210.013-0.039 (0.016)(0.017)(0.025)(0.032) Displacementyear+1-0.009-0.0060.041-0.036 (0.017)(0.017)(0.027)(0.035) Displacementyear+2-3-0.0160.0060.072-0.050 (0.018)(0.018)(0.033)**(0.038) Displacementyear+4-5-0.0230.0350.119-0.046 (0.023)(0.023)(0.041)***(0.052) Displacementyear+6-7-0.0360.0540.147-0.034 (0.026)(0.027)**(0.049)***(0.063) Displacementyear+8+-0.0360.0410.144-0.053 (0.029)(0.030)(0.054)***(0.072) Obersations 71,53477,24646,06331,183 Notes:Standarderrorsclusteredatindividuallevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,individualandyearedeffects. 148 TableH.2:ImpactofDisplacementontheProbabilityofHavinganAdditionalChild(Including marriagestatus) WomenwithoutWomenwith MenWomenCollegeEducationCollegeEducation Displacementyear-2-0.0130.0160.0080.043 (0.011)(0.013)(0.016)(0.023)* Displacementyear-10.007-0.01-0.0150.01 (0.010)(0.011)(0.013)(0.018) Displacementyear0.011-0.008-0.0070 (0.009)(0.010)(0.013)(0.017) Displacementyear+10.0080.0060.0120.007 (0.010)(0.011)(0.013)(0.017) Displacementyear+2-300.0030.014-0.001 (0.008)(0.009)(0.011)(0.014) Displacementyear+4-50.0010.0090.0210.008 (0.008)(0.009)(0.011)**(0.014) Displacementyear+6-7-0.0080.0110.020.017 (0.008)(0.008)(0.010)**(0.014) Displacementyear+8+-0.010.0080.0190.014 (0.007)(0.007)(0.009)**(0.011) Married0.110.0990.0640.122 (0.004)***(0.004)***(0.005)***(0.005)*** Divorce0.0420.0420.0230.045 (0.005)***(0.004)***(0.006)***(0.007)*** Obersations 75,20581,23448,43832,796 Notes:Standarderrorsclusteredatindividuallevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,individualandyearedeffects. 149 FigureH.1:DisplacementandBirthRateforMenwithCollegeEducation DataSources:NLSY79. 150 FigureH.2:DisplacementandBirthRateforMenwithCollegeEducation DataSources:NLSY79. 151 FigureH.3:DisplacementandBirthRateforMenwithoutCollegeEducation DataSources:NLSY79. 152 APPENDIXI APPENDICESFORCHAPTER3 TableI.1:ListofCES-D-10itemstomeasurementalhealth ItemDescription 1Iwasbotheredbythingsthatdon'tusuallybotherme. 2IhadtroublekeepingmymindonwhatIwasdoing. 3Ifeltdepressed. 4IfelteverythingIdidwasaneffort. 5Ifelthopefulaboutthefuture. 6Ifeltfearful. 7Mysleepwasrestless. 8Iwashappy. 9Ifeltlonely. 10Icouldnotget"going." Notes:The10itemsaboverefertohowindividualhavefeltandbehavedduringthelastweek.Individualaare requestedtochoosetheappropriateresponseas:(1)Rarelyornoneofthetime(<1day)(2)Someoralittleof thetime(1-2days)(3)Occasionallyoramoderateamountofthetime(3-4days)(4)Mostorallofthetime(5-7 days) Foreachindividual,weusethisformulatocalculateCES-D-10: CES-D-10=item1-1+item2-1+item3-1+item4-1+item6-1+item7-1+item9-1+item10-1+4-item5+4-item8 153 TableI.2:CoefofInteractionTermsforthe1stStageRegressiononTwoorMoreChildren (1)(2)(3)(1)(2)(3) age45*First-Born Girl 0.1740.1420.17age59*First-Born Girl 0.0510.0260.046 (0.062)***(0.063)**(0.045)***(0.030)*(0.035)(0.027)* age46*First-Born Girl 0.160.140.135age60*First-Born Girl 0.0310.020.052 (0.047)***(0.049)***(0.040)***(0.025)(0.030)(0.020)** age47*First-Born Girl 0.1540.1530.209age61*First-Born Girl -0.0050.0280.04 (0.037)***(0.043)***(0.032)***(0.026)(0.030)(0.024)* age48*First-Born Girl 0.2060.2080.157age62*First-Born Girl -0.006-0.0430.027 (0.037)***(0.036)***(0.031)***(0.025)(0.027)(0.022) age49*First-Born Girl 0.0950.1090.132age63*First-Born Girl -0.0330.014-0.013 (0.037)**(0.041)***(0.030)***(0.026)(0.028)(0.027) age50*First-Born Girl 0.1360.1750.17age64*First-Born Girl -0.062-0.059-0.024 (0.038)***(0.043)***(0.031)***(0.031)**(0.047)(0.022) age51*First-Born Girl 0.1320.0880.138age65*First-Born Girl 0.0120.0290.022 (0.043)***(0.044)**(0.032)***(0.027)(0.040)(0.027) age52*First-Born Girl 0.0840.1510.093age66*First-Born Girl 0.0260.01-0.006 (0.042)**(0.036)***(0.041)**(0.031)(0.038)(0.023) age53*First-Born Girl 0.0380.0930.071age67*First-Born Girl -0.037-0.038-0.058 -0.036(0.038)**(0.038)*(0.032)(0.030)(0.023)** age54*First-Born Girl 0.1070.1140.054age68*First-Born Girl -0.0140.003-0.031 154 TableI.2(cont'd) (0.036)***(0.036)***(0.031)*(0.030)(0.024)(0.025) age55*First-Born Girl 0.0590.1340.069age69*First-Born Girl -0.008-0.039-0.031 (0.033)*(0.036)***(0.034)**(0.040)(0.042)(0.033) age56*First-Born Girl 0.0970.0840.057age70*First-Born Girl -0.007-0.0310.013 (0.034)***(0.036)**(0.031)*(0.029)(0.038)(0.024) age57*First-Born Girl 0.040.0730.052 (0.03)(0.029)**(0.030)* Cragg-Donald WaldF 6.793.134.1 age58*First-Born Girl 0.0240.0390.045 R-squared 0.130.130.12 (0.030)(0.039)(0.031) Observations 9,4624,4065,056 Notes: Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsagecohortdummies,genderofbirth,ageatbirth,sex,education,numberofsiblings, maritalstatus,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)isonthewholesample;Col(2)isformenonly;Col(3)isforwomenonly. 155 TableI.3:CoefofInteractionTermsforthe1stStageRegressiononNumberofChildren (1)(2)(3)(1)(2)(3) age45*First- BornGirl 0.4160.510.508age59*First- BornGirl 0.266-0.078-0.131 (0.176)**(0.167)***(0.225)**(0.144)*(0.169)(0.174) age46*First- BornGirl 0.2180.3190.263age60*First- BornGirl 0.2330.071-0.073 (0.129)*(0.153)**(0.154)*(0.139)*(0.149)(0.171) age47*First- BornGirl 0.2120.2990.377age61*First- BornGirl 0.2170.0290.023 (0.144)(0.165)*(0.169)**(0.151)(0.169)(0.174) age48*First- BornGirl 0.3180.4070.197age62*First- BornGirl 0.3850.0090.022 (0.126)**(0.146)***(0.161)(0.144)***(0.164)(0.181) age49*First- BornGirl 0.2190.2740.133age63*First- BornGirl 0.1620.0980.093 (0.129)*(0.137)**(0.170)(0.166)(0.166)(0.193) age50*First- BornGirl 0.4490.4190.336age64*First- BornGirl -0.148-0.0210.022 (0.132)***(0.168)**(0.177)*(0.176)(0.184)(0.174) age51*First- BornGirl 0.3510.1260.209age65*First- BornGirl 0.0830.160.215 (0.146)**(0.166)(0.189)(0.189)(0.185)(0.214) age52*First- BornGirl 0.4230.6560.054age66*First- BornGirl 0.130.025-0.033 (0.153)***(0.170)***(0.174)(0.167)(0.187)(0.241) age53*First- BornGirl 0.2370.2820.068age67*First- BornGirl -0.2990.027-0.253 (0.141)*(0.161)*(0.169)(0.203)(0.184)(0.220) age54*First- BornGirl 0.2730.1630.002age68*First- BornGirl 0.0290.294-0.035 156 TableI.3(cont'd) (0.142)*(0.189)(0.177)(0.223)(0.209)(0.235) age55*First- BornGirl 0.2690.24-0.063age69*First- BornGirl -0.1140.0880.046 (0.139)*(0.148)(0.193)(0.231)(0.245)(0.262) age56*First- BornGirl 0.3530.172-0.101age70*First- BornGirl -0.0810.1880.13 (0.133)***(0.160)(0.161)(0.217)(0.191)(0.219) age57*First- BornGirl 0.2910.075-0.037 (0.137)**(0.163)(0.175) Cragg-Donald WaldF 2.591.441.72 age58*First- BornGirl 0.330.178-0.08 R-squared 0.450.420.45 (0.145)**(0.174)(0.175) Observations 9,6574,5175,140 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsagecohortdummies,genderofbirth,ageatbirth,sex,education,numberofsiblings, maritalstatus,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)isonthewholesample;Col(2)isformenonly;Col(3)isforwomenonly. 157 TableI.4:CoefofInteractionTermsforthe1stStageRegressiononTwoorMoreChildren (1)(2)(3)(1)(2)(3) age45*First- BornGirl 0.1850.1550.175age54*First- BornGirl 0.1180.1290.061 (0.061)***(0.061)**(0.044)***(0.035)***(0.029)***(0.029)** age46*First- BornGirl 0.1710.1530.141age55*First- BornGirl 0.070.1490.076 (0.045)***(0.046)***(0.039)***(0.031)**(0.032)***(0.031)** age47*First- BornGirl 0.1650.1660.215age56*First- BornGirl 0.1080.0990.063 (0.036)***(0.041)***(0.030)***(0.033)***(0.032)***(0.028)** age48*First- BornGirl 0.2170.2220.163age57*First- BornGirl 0.0520.0880.058 (0.035)***(0.032)***(0.029)***(0.029)*(0.025)***(0.027)** age49*First- BornGirl 0.1060.1240.138age58*First- BornGirl 0.0350.0540.052 (0.035)***(0.037)***(0.027)***-0.028(0.034)(0.028)* age50*First- BornGirl 0.1470.1890.177age59*First- BornGirl 0.0620.0420.053 (0.038)***(0.043)***(0.029)***(0.028)**(0.031)(0.025)** age51*First- BornGirl 0.1430.1020.144age60*First- BornGirl 0.0420.0340.059 (0.042)***(0.041)**(0.029)***(0.023)*(0.024)(0.017)*** age52*First- BornGirl 0.0950.1660.1age61*First- BornGirl 0.0060.0430.047 (0.041)**(0.033)***(0.038)***(0.023)(0.025)*(0.020)** age53*First- BornGirl 0.0490.1080.078 Cragg-Donald WaldF 2.6711.6661.49 -0.035(0.035)***(0.036)** Observations 9,4624,4065,056 158 TableI.4(cont'd) Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsagecohortdummies,genderofbirth,ageatbirth,sex,education,numberofsiblings, maritalstatus,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)isonthewholesample;Col(2)isformenonly;Col(3)isforwomenonly. Cohortsageabove62areusedascontrol. 159 TableI.5:CoefofInteractionTermsforthe1stStageRegressiononNumberofChildren (1)(2)(3)(1)(2)(3) age45*First- BornGirl 0.390.460.489age54*First- BornGirl 0.2480.099-0.019 (0.148)***(0.149)***(0.197)**(0.109)**(0.156)(0.113) age46*First- BornGirl 0.1910.2670.243age55*First- BornGirl 0.2420.177-0.085 (0.096)**(0.130)**(0.104)**(0.114)**(0.099)*(0.138) age47*First- BornGirl 0.1850.2420.357age56*First- BornGirl 0.3270.106-0.123 (0.113)(0.136)*(0.118)***(0.107)***(0.118)(0.103) age48*First- BornGirl 0.2910.3490.177age57*First- BornGirl 0.2630.013-0.06 (0.088)***(0.110)***(0.103)*(0.102)**(0.125)(0.118) age49*First- BornGirl 0.1940.2150.113age58*First- BornGirl 0.3020.114-0.102 (0.094)**(0.100)**(0.112)(0.114)***(0.142)(0.115) age50*First- BornGirl 0.4230.3610.314age59*First- BornGirl 0.236-0.142-0.152 (0.101)***(0.128)***(0.116)***(0.109)**(0.129)(0.122) age51*First- BornGirl 0.3230.0630.186age60*First- BornGirl 0.2060.005-0.093 (0.114)***(0.128)(0.133)(0.105)*(0.112)(0.114) age52*First- BornGirl 0.3950.5930.033age61*First- BornGirl 0.189-0.0360.003 (0.123)***(0.146)***(0.116)(0.122)(0.127)(0.126) age53*First- BornGirl 0.2110.2190.049 Cragg-Donald WaldF 9.6354.8545.442 (0.108)*(0.127)*(0.108) Observations 9,6574,5175,140 160 TableI.5(cont'd) Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsagecohortdummies,genderofbirth,ageatbirth,sex,education,numberofsiblings, maritalstatus,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)isonthewholesample;Col(2)isformenonly;Col(3)isforwomenonly. Cohortsageabove62areusedascontrol. 161 TableI.6:2SLSResultsforEffectsofHavingTwoChildrenonParent'sMentalHealth (1)(2)(3) Having2Children8.0044.1957.191 (5.131)(5.662)(7.019) AgeatFirstBirth0.2170.1670.155 (0.129)*(0.149)(0.188) Men-1.816 (0.226)*** YearofSchooling-0.173-0.276-0.112 (0.032)***(0.045)***(0.055)** #ofSiblings0.010.086-0.051 (0.067)(0.091)(0.094) TemporarySeperated0.7611.880.745 (0.496)(0.841)**(0.570) Seperated9.5698.5666.978 (2.349)***(2.901)***(3.565)* Divorced7.2848.8463.514 (1.963)***(2.197)***(3.074) Widowed2.2942.2721.204 (0.655)***(1.085)**(0.713)* Nevermarried4.4327.332-3.546 (3.632)(4.841)(4.726) R-squared -0.120.03-0.1 Observations 4,2091,9952,158 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Observationsareparentswitheither1or2children. Allregressionscontrolsage,agesquare,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)isonthewholesample;Col(2)isformenonly;Col(3)isforwomenonly. 162 TableI.7:OLSEstimatesofEffectsofTwoorMoreChildrenonParent'sMentalHealthbyParents'Education LessThanPrimarySchoolAtLeastPrimarySchool (1)(2)(3)(4)(5)(6) #ofChildren > 1-0.039-0.029-0.1120.2160.0550.325 (0.442)(0.837)(0.523)(0.275)(0.362)(0.478) AgeatFirstBirth0.0450.061-0.0010.0160.048-0.063 (0.023)*(0.038)(0.031)(0.029)(0.034)(0.060) Men-1.848-1.739 (0.217)***(0.174)*** YearofSchooling0.035-0.0350.065-0.194-0.23-0.136 (0.067)(0.108)(0.087)(0.043)***(0.055)***(0.072)* #ofSiblings-0.029-0.022-0.0260.0730.1040.018 (0.055)(0.084)(0.069)(0.048)(0.058)*(0.085) TemporarySeperated0.7971.5080.540.410.81-0.075 (0.416)*(0.836)*(0.505)(0.473)(0.695)(0.652) Seperated5.3498.292.7317.1386.46914.155 (1.507)***(1.890)***(2.544)(1.915)***(2.080)***(0.766)*** Divorced3.652.1183.4094.9255.785-1.37 (1.589)**(3.217)(2.146)(1.564)***(1.632)***(2.258) Widowed1.2612.0561.171.7591.5172.126 (0.358)***(0.726)***(0.388)***(0.439)***(0.655)**(0.683)*** Nevermarried2.4392.947-0.0150.832 (1.779)(3.267)(0.425)(0.518) R-squared 0.040.050.020.060.050.04 Observations 469714643233476529421823 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)andCol(4)areonthewholesample;Col(2)andCol(5)areformenonly;Col(3)andCol(6)areforwomenonly. 163 TableI.8:OLSEstimatesofEffectsofNumberofChildrenonParent'sMentalHealthbyParents'Education LessThanPrimarySchoolAtLeastPrimarySchool (1)(2)(3)(4)(5)(6) #ofChildren-0.004-0.2330.0440.2520.2340.307 (0.082)(0.147)(0.106)(0.100)**(0.109)**(0.189) AgeatFirstBirth0.0450.05200.0390.067-0.015 (0.024)*(0.040)(0.032)(0.028)(0.034)**(0.059) Men-1.836-1.778 (0.214)***(0.172)*** YearofSchooling0.036-0.0140.059-0.193-0.232-0.121 (0.067)(0.107)(0.086)(0.042)***(0.054)***(0.070)* #ofSiblings-0.042-0.02-0.0410.0780.119-0.022 (0.054)(0.083)(0.068)(0.047)*(0.057)**(0.086) TemporarySeperated0.7081.2340.4910.290.632-0.214 (0.413)*(0.846)(0.499)(0.464)(0.679)(0.652) Seperated5.0467.3992.7094.6854.4895.978 (1.481)***(1.950)***(2.552)(1.876)**(2.097)**(5.853) Divorced3.631.8813.5035.0125.886-1.237 (1.572)**(3.177)(2.147)(1.542)***(1.614)***(2.234) Widowed1.2641.9861.1411.7791.5792.056 (0.353)***(0.700)***(0.388)***(0.434)***(0.645)**(0.675)*** Nevermarried0.9380.0941.8562.8543.188 (1.158)(1.520)(2.106)(3.730)(3.414) R-squared 0.040.050.020.060.050.04 Observations 4,7921,5093,2834,8653,0081,857 Notes:Standarderrorsclusteredatcountylevelarereportedinbrackets. at10%level;at5%level;at1%level. Allregressionscontrolsage,agesquare,self-reportedhealthduringchildhood,andcountyedeffects. Col(1)andCol(4)areonthewholesample;Col(2)andCol(5)areformenonly;Col(3)andCol(6)areforwomenonly. 164 BIBLIOGRAPHY 165 BIBLIOGRAPHY Adsera,A.(2005).Vanishingchildren:Fromhighunemploymenttolowfertilityindeveloped countries. AmericanEconomicReview2 (95),189Œ193. Adsera,A.(2011).Wherearethebabies?labormarketconditionsandfertilityineurope. European JournalofPopulation/RevueeuropéennedeDémographie27 (1),1Œ32. Adsera,A.andA.Menendez(2009).Fertilitychangesinlatinamericainthecontextofeconomic uncertainty.Technicalreport,IZAdiscussionpapers. Agüero,J.M.andM.S.Marks(2011).Motherhoodandfemalelaborsupplyinthedeveloping worldevidencefrominfertilityshocks. JournalofHumanResources46 (4),800Œ826. Alesina,A.,R.DiTella,andR.MacCulloch(2004).Inequalityandhappiness:areeuropeansand americansdifferent? JournalofPublicEconomics88 (9),2009Œ2042. Amialchuk,A.(2013).Theeffectofhusband'sjobdisplacementonthetimingandspacingof birthsintheunitedstates. ContemporaryEconomicPolicy31 (1),73Œ93. Ananat,E.O.,C.Gibson-Davis,andA.Gassman-Pines(2011).Thegreatrecessionandlocaljob loss:Effectsonfertility.WorkingPaper. Angeles,L.(2010).Childrenandlifesatisfaction. JournalofhappinessStudies11 (4),523Œ538. Angrist,J.D.andW.N.Evans(1998).Childrenandtheirparents'laborsupply:Evidencefrom exogenousvariationinfamilysize. AmericanEconomicReview88 (3),450Œ477. Angrist,J.D.andW.N.Evans(2000).Schoolingandlabormarketconsequencesofthe1970state abortionreforms. ResearchinLaborEconomics18 ,75Œ113. Angrist,J.D.andG.W.Imbens(1995).Two-stageleastsquaresestimationofaveragecausal effectsinmodelswithvariabletreatmentintensity. JournaloftheAmericanstatisticalAssocia- tion90 (430),431Œ442. Bailey,M.J.(2006).Morepowertothepill:theimpactofcontraceptivefreedomonwomen'slife cyclelaborsupply. TheQuarterlyJournalofEconomics121 (1),289Œ320. Banerjee,A.,X.Meng,andN.Qian(2010).Thelifecyclemodelandhouseholdsavings:Micro evidencefromurbanchina.WorkingPaper. Becker,G.S.(1960).Aneconomicanalysisoffertility.In Demographicandeconomicchangein developedcountries ,pp.209Œ240.ColumbiaUniversityPress. Ben-Porath,V.(1973).Short-terminfertilityandeconomicactivityinisrael. Demog- raphy10 (2),185Œ204. 166 Berkman,L.F.,C.S.Berkman,S.Kasl,D.H.Freeman,L.Leo,A.M.Ostfeld,J.Cornoni-Huntley, andJ.A.Brody(1986).Depressivesymptomsinrelationtophysicalhealthandfunctioningin theelderly. AmericanJournalofEpidemiology124 (3),372Œ388. Black,D.A.,N.Kolesnikova,S.G.Sanders,andL.J.Taylor(2013).Arechildrenfinormalfl? The ReviewofEconomicsandStatistics95 (1),21Œ33. Black,S.E.,P.G.Devereux,andK.G.Salvanes(2004).Themorethemerrier?theeffectof familycompositiononchildren'seducation.WorkingPaper. wer,D.G.(2009).Internationalevidenceonwell-being.In Measuringthesubjective well-beingofnations:Nationalaccountsoftimeuseandwell-being ,pp.155Œ226.Universityof ChicagoPress. Bloom,D.E.,D.Canning,G.Fink,andJ.E.Finlay(2009).Fertility,femalelaborforceparticipa- tion,andthedemographicdividend. JournalofEconomicGrowth14 (2),79Œ101. Boey,K.W.(1999).Cross-validationofashortformoftheces-dinchineseelderly. International journalofgeriatricpsychiatry14 (8),608Œ617. Buber,I.andH.Engelhardt(2008).Children'simpactonthementalhealthoftheiroldermothers andfathers:fromthesurveyofhealth,ageingandretirementineurope. European JournalofAgeing5 (1),31Œ45. Burton,R.P.(1998).Globalintegrativemeaningasamediatingfactorintherelationshipbetween socialrolesandpsychologicaldistress. JournalofHealthandSocialBehavior39 (3),201Œ215. Butz,W.P.andM.P.Ward(1979).Theemergenceofcountercyclicalusfertility. TheAmerican EconomicReview69 (3),318Œ328. Cáceres-Delpiano,J.andM.Simonsen(2012).Thetolloffertilityonmothers'wellbeing. Journal ofHealthEconomics31 (5),752Œ766. Charles,K.K.andM.Stephens(2004).Jobdisplacement,disability,anddivorce. Journalof LaborEconomics22 (2),489Œ522. Chen,Y.,H.Li,andL.Meng(2013).Prenatalsexselectionandmissinggirlsinchina:Evidence fromthediffusionofdiagnosticultrasound. JournalofHumanResources48 (1),36Œ70. Cong,Z.andM.Silverstein(2008).Intergenerationaltime-for-moneyexchangesinruralchina: Doesreciprocityreducedepressivesymptomsofoldergrandparents? ResearchinHumanDe- velopment5 (1),6Œ25. Couch,K.A.andD.W.Placzek(2010).Earningslossesofdisplacedworkersrevisited. The AmericanEconomicReview100 (1),572Œ589. Cruces,G.andS.Galiani(2007).Fertilityandfemalelaborsupplyinlatinamerica:Newcausal evidence. LabourEconomics14 (3),565Œ573. 167 Currie,J.andH.Schwandt(2014).Short-andlong-termeffectsofunemploymentonfertility. ProceedingsoftheNationalAcademyofSciences111 (41),14734Œ14739. Dean,A.,B.Kolody,andP.Wood(1990).Effectsofsocialsupportfromvarioussourceson depressioninelderlypersons. JournalofHealthandSocialBehavior31 (2),148Œ161. Dehejia,R.andA.Lleras-Muney(2004).Booms,busts,andbabies'health*. TheQuarterly journalofeconomics119 (3),1091Œ1130. DelBono,E.,A.Weber,andR.Winter-Ebmer(2012).Clashofcareerandfamily:Fertility decisionsafterjobdisplacement. JournaloftheEuropeanEconomicAssociation10 (4),659Œ 683. Demyttenaere,K.,R.Bruffaerts,J.Posada-Villa,I.Gasquet,V.Kovess,J.Lepine,M.Angermeyer, S.Bernert,G.DeGirolamo,P.Morosini,etal.(2004).Prevalence,severity,andunmetneedfor treatmentofmentaldisordersintheworldhealthorganizationworldmentalhealthsurveys. JAMA:thejournaloftheAmericanMedicalAssociation291 (21),2581Œ2590. DiTella,R.,R.J.MacCulloch,andA.J.Oswald(2003).Themacroeconomicsofhappiness. ReviewofEconomicsandStatistics85 (4),809Œ827. Earle,J.R.,M.H.Smith,C.T.Harris,andC.F.LonginoJr(1997).Women,maritalstatus,and symptomsofdepressioninamidlifenationalsample. Journalofwomen&aging10 (1),41Œ57. Ebenstein,A.(2009).Whenisthelocalaveragetreatmentclosetotheaverage?evidencefrom fertilityandlaborsupply. JournalofHumanResources44 (4),955Œ975. Eliason,M.(2012).Lostjobs,brokenmarriages. JournalofPopulationEconomics25 (4),1365Œ 1397. Ermisch,J.(1988).Econometricanalysisofbirthratedynamicsinbritain. JournalofHuman Resources23 (4),563Œ576. Ermisch,J.F.(1980).Timecosts,aspirationsandtheeffectofeconomicgrowthongermanfertil- ity*. OxfordBulletinofEconomicsandStatistics42 (2),125Œ143. Evenson,R.J.andR.W.Simon(2005).Clarifyingtherelationshipbetweenparenthoodand depression. JournalofhealthandSocialBehavior46 (4),341Œ358. Fang,H.,K.N.Eggleston,J.A.Rizzo,andR.J.Zeckhauser(2010).Femaleemploymentand fertilityinruralchina.Technicalreport,HarvardUniversity,JohnF.KennedySchoolofGov- ernment. Feng,W.(2010).China'spopulationdestiny:Theloomingcrisis. CurrentHistory109 (728), 244Œ251. Gilbert,D.(2009). Stumblingonhappiness .VintageCanada. Gove,W.R.andM.R.Geerken(1977).Theeffectofchildrenandemploymentonthemental healthofmarriedmenandwomen. SocialForces56 (1),66Œ76. 168 Greenhalgh,S.(1986).Shiftsinchina'spopulationpolicy,1984-86:Viewsfromthecentral, provincial,andlocallevels. PopulationandDevelopmentReview12 (3),491Œ515. Greenhalgh,S.(2003).Science,modernity,andthemakingofchina'sone-childpolicy. Population andDevelopmentReview29 (2),163Œ196. Gronau,R.(1977).Leisure,homeproduction,andwork-thetheoryoftheallocationoftimerevis- ited. TheJournalofPoliticalEconomy85 (6),1099Œ1123. Hank,K.(2010).Childbearinghistory,later-lifehealth,andmortalityingermany. Population studies64 (3),275Œ291. Heckman,J.J.andJ.R.Walker(1990).Therelationshipbetweenwagesandincomeandthe timingandspacingofbirths:evidencefromswedishlongitudinaldata. Econometrica:journal oftheEconometricSociety58 (6),1411Œ1441. HildaL.Solis,J.M.G.(2012). ChartingInternationalLaborComparisons (2012Editioned.). U.S.DepartmentofLabor,BureauofLaborStatistics. Hotz,V.J.,J.A.Klerman,andR.J.Willis(1997).Theeconomicsoffertilityindevelopedcoun- tries. Handbookofpopulationandfamilyeconomics1 (Part1),275Œ347. Hotz,V.J.andR.A.Miller(1993).Conditionalchoiceprobabilitiesandtheestimationofdynamic models. TheReviewofEconomicStudies60 (3),497Œ529. Hoynes,H.(2002).Theemployment,earnings,andincomeoflessskilledworkersoverthebusi- nesscycle.InR.BlankandD.Card(Eds.), FindingJobs:WorkandWelfareReform .Russell SageFoundation. Hu,T.-w.,Y.He,M.Zhang,andN.Chen(2007).Economiccostsofdepressioninchina. Social psychiatryandpsychiatricepidemiology42 (2),110Œ116. Hurt,L.S.,C.Ronsmans,andS.L.Thomas(2006).Theeffectofnumberofbirthsonwomen's mortality:systematicreviewoftheevidenceforwomenwhohavecompletedtheirchildbearing. Populationstudies60 (1),55Œ71. Huttunen,K.andJ.Kellokumpu(2012).Theeffectofjobdisplacementoncouples'fertility decisions.Workingpaper,HelsinkiCenterforEconomicResearch. Irwin,M.,K.H.Artin,andM.N.Oxman(1999).Screeningfordepressionintheolderadult:crite- rionvalidityofthe10-itemcenterforepidemiologicalstudiesdepressionscale(ces-d). Archives ofInternalMedicine159 (15),1701Œ1704. Islam,A.andR.Smyth(2010).Childrenandparentalhealth:Evidencefromchina.Working Paper. Jacobsen,J.P.,J.W.PearceIII,andJ.L.Rosenbloom(1999).Theeffectsofchildbearingon marriedwomen'slaborsupplyandearnings:usingtwinbirthsasanaturalexperiment. Journal ofHumanResources34 (3),449Œ474. 169 Jacobson,L.S.,R.J.LaLonde,andD.G.Sullivan(1993).Earningslossesofdisplacedworkers. TheAmericanEconomicReview83 (4),685Œ709. Jones,L.E.,A.Schoonbroodt,andM.Tertilt(2010).Fertilitytheories:Cantheyexplainthenega- tivefertility-incomerelationship?In DemographyandtheEconomy ,pp.43Œ100.Universityof ChicagoPress. Kendig,H.,P.A.Dykstra,R.I.vanGaalen,andT.Melkas(2007).Healthofagingparentsand childlessindividuals. JournalofFamilyIssues28 (11),1457Œ1486. Klerman,J.A.(1999).Usabortionpolicyandfertility. TheAmericanEconomicReview89 (2), 261Œ264. Kletzer,L.G.andR.W.Fairlie(2003).Thelong-termcostsofjobdisplacementforyoungadult workers. IndustrialandLaborRelationsReview56 (4),682Œ698. Kohler,H.-P.,J.R.Behrman,andA.Skytthe(2005).Partner+children=happiness?theeffectsof partnershipsandfertilityonwell-being. Populationanddevelopmentreview31 (3),407Œ445. Kohout,F.J.,L.F.Berkman,D.A.Evans,andJ.Cornoni-Huntley(1993).Twoshorterformsof theces-ddepressionsymptomsindex. JournalofAgingandHealth5 (2),179Œ193. Koropeckyj-Cox,T.(1998).Lonelinessanddepressioninmiddleandoldage:Arethechildless morevulnerable? TheJournalsofGerontologySeriesB:PsychologicalSciencesandSocial Sciences53 (6),S303ŒS312. Kruk,K.E.andS.Reinhold(2014).Theeffectofchildrenondepressioninoldage. SocialScience &Medicine100 ,1Œ11. Lechner,M.(1999).Earningsandemploymenteffectsofcontinuousgff-the-jobtrainingineast germanyafter JournalofBusiness&EconomicStatistics17 (1),74Œ90. Lee,D.(2002).Fertilityandfemalelaborsupplyinruralchina.WorkingPaper. Levine,P.,D.Staiger,T.Kane,D.Zimmerman,etal.(1999).Roevwadeandamericanfertility. AmericanJournalofPublicHealth89 (2),199. Li,H.,J.Yi,andJ.Zhang(2011).Estimatingtheeffectoftheone-childpolicyonthesexratio imbalanceinchina:basedonthedifference-in-differences. Demography48 (4), 1535Œ1557. Li,H.,J.Zhang,andY.Zhu(2005).Theeffectoftheone-childpolicyonfertilityinchina: basedondifferences-in-differences.WorkingPaper. Lindo,J.M.(2010).Arechildrenreallyinferiorgoods?evidencefromdisplacement-drivenin- comeshocks. JournalofHumanResources45 (2),301Œ327. Lu,C.,R.G.Frank,Y.Liu,andJ.Shen(2009).Theimpactofmentalhealthonlabourmarket outcomesinchina. Thejournalofmentalhealthpolicyandeconomics12 (3),157. 170 Mason,K.O.andA.M.Taj(1987).Differencesbetweenwomen'sandmen'sreproductivegoals indevelopingcountries. PopulationandDevelopmentReview13 (4),611Œ638. Maurer-Fazio,M.,R.Connelly,L.Chen,andL.Tang(2011).Childcare,eldercare,andlaborforce participationofmarriedwomeninurbanchina,1982Œ2000. JournalofHumanResources46 (2), 261Œ294. Maurer-Fazio,M.,J.Hughes,andD.Zhang(2005).Economicreformandchangingpatternsof laborforceparticipationinurbanandruralchina.Technicalreport,WilliamDavidsonInstitute attheUniversityofMichigan. McElroy,M.andD.T.Yang(2000).Carrotsandsticks:fertilityeffectsofchina'spopulation policies. TheAmericanEconomicReview90 (2),389Œ392. Merrigan,P.andY.S.Pierre(1998).Aneconometricandneoclassicalanalysisofthetimingand spacingofbirthsincanadafrom1950to1990. JournalofPopulationEconomics11 (1),29Œ51. Mirowsky,J.andC.E.Ross(2002).Depression,parenthood,andageatbirth. SocialScience &Medicine54 (8),1281Œ1298. Mu,R.(2013).Regionaldisparitiesinself-reportedhealth:Evidencefromchineseolderadults. Healtheconomics23 (5),529Œ49. Mundlak,Y.(1978).Onthepoolingoftimeseriesandcrosssectiondata. Econometrica46 (1), 69Œ85. Oreopoulos,P.,M.Page,andA.H.Stevens(2008).Theintergenerationaleffectsofworkerdis- placement. JournalofLaborEconomics26 (3),455Œ000. Pei,X.andV.K.Pillai(1999).Oldagesupportinchina:Theroleofthestateandthefamily. InternationalJournalofAgingandHumanDevelopment49 (3),197Œ212. Peng,L.,C.D.Meyerhoefer,andS.H.Zuvekas(2013).Theeffectofdepressiononlabormarket outcomes.Technicalreport,NationalBureauofEconomicResearch. Peng,P.(1996). EncyclopediaofBirthControlPoliciesinChina .Beijing:ThePeople'sPress. Perry,C.(2003). HowDoFemaleEarningsAffectFertilityDecisions? Ph.D.thesis,Massachusetts InstituteofTechnology. Porter,M.andE.M.King(2010).Fertilityandwomen'slaborforceparticipationindeveloping countries.WorkingPaper. Qian,N.(2009).Quantity-qualityandtheonechildpolicy:Theonly-childdisadvantageinschool enrollmentinruralchina.Technicalreport,NationalBureauofEconomicResearch. Rasul,I.(2008).Householdbargainingoverfertility:Theoryandevidencefrommalaysia. Journal ofDevelopmentEconomics86 (2),215Œ241. 171 Rosenbaum,P.R.andD.B.Rubin(1983).Thecentralroleofthepropensityscoreinobservational studiesforcausaleffects. Biometrika70 (1),41Œ55. Rosenzweig,M.R.andK.I.Wolpin(1980).Life-cyclelaborsupplyandfertility:Causalinfer- encesfromhouseholdmodels. TheJournalofPoliticalEconomy88 (2),328Œ348. Rosenzweig,M.R.andK.I.Wolpin(2000).Natural"naturalexperiments"ineconomics. Journal ofEconomicLiterature38 (4),827Œ874. Ruhm,C.J.(1991).Areworkerspermanentlyscarredbyjobdisplacements? TheAmerican EconomicReview81 (1),319Œ324. Scharping,T.(2003). BirthControlinChina1949-2000 .LondonandNewYork:Routledge Curzon. Schultz,T.P.(1985).Changingworldprices,women'swages,andthefertilitytransition:Sweden, 1860-1910. TheJournalofPoliticalEconomy93 (6),1126Œ1154. Schultz,T.P.(2009).Howdoesfamilyplanningpromotedevelopment?evidencefromaso- cialexperimentinmatlab,bangladesh,1977-1996.Technicalreport,YaleUniversity,Economic GrowthCenter,NewHaven,Conneticut. Schultz,T.P.andY.Zeng(1995).Fertilityofruralchina.effectsoflocalfamilyplanningand healthprograms. JournalofPopulationEconomics8 (4),329Œ350. Short,S.E.andF.Zhai(1998).Lookinglocallyatchina'sone-childpolicy. StudiesinFamily Planning29 (4),373Œ387. Silverstein,M.,Z.Cong,andS.Li(2006).Intergenerationaltransfersandlivingarrangements ofolderpeopleinruralchina:Consequencesforpsychologicalwell-being. TheJournalsof GerontologySeriesB:PsychologicalSciencesandSocialSciences61 (5),S256ŒS266. Singh,I.,L.Squire,J.Strauss,etal.(1986). Agriculturalhouseholdmodels:extensions,applica- tions,andpolicy. JohnsHopkinsUniversityPress. Staiger,D.O.andJ.H.Stock(1994).Instrumentalvariablesregressionwithweakinstruments. Stevens,A.H.(1997).Persistenteffectsofjobdisplacement:Theimportanceofmultiplejob losses. JournalofLaborEconomics15 (1),165Œ188. Stevens,A.H.andJ.Schaller(2011).Short-runeffectsofparentaljoblossonchildren'sacademic achievement. EconomicsofEducationReview30 (2),289Œ299. Umberson,D.andW.R.Gove(1989).Parenthoodandpsychologicalwell-beingtheory,measure- ment,andstageinthefamilylifecourse. JournalofFamilyIssues10 (4),440Œ462. Walker,J.R.(2002,12).Acommentonalitasiran's"wageandincomeeffectsonthetimingand spacingofbirthsinswedenandintheunitedstates". JournalofPopulationEconomics15 (4), 773Œ782. 172 Wang,F.(2000).Chinacensusesof1982and1990.InP.K.Hall,R.McCaa,andG.Thorvaldsen (Eds.), HandbookofInternationalHistoricalMicrodataforPopulationResearch ,Chapter4,pp. 45Œ60.MinnesotaPopulationCenter. Ward,M.P.andW.P.Butz(1980).Completedfertilityanditstiming. TheJournalofPolitical Economy88 (5),917Œ940. Wei,S.-J.andX.Zhang(2011).Thecompetitivesavingmotive:Evidencefromrisingsexratios andsavingsratesinchina. JournalofPoliticalEconomy119 (3),511Œ564. White,T.(1991).Birthplanningbetweenplanandmarket:Theimpactofreformonchina's one-childpolicy. China'sEconomicDilemmasinthe1990s.TheProblemsofReforms,Mod- ernization,andInterdependence1 ,252Œ69. Willis,R.(1974).Economictheoryoffertilitybehavior.In EconomicsoftheFamily:Marriage, Children,andHumanCapital ,pp.25Œ80.UMI. Willis,R.J.(1973).Anewapproachtotheeconomictheoryoffertilitybehavior. TheJournalof PoliticalEconomy81 (2),S14ŒS64. Wooldridge,J.M.(2010). Econometricanalysisofcrosssectionandpaneldata .MITpress. Wu,X.andL.Li(2012).Familysizeandmaternalhealth:evidencefromtheone-childpolicyin china. JournalofPopulationEconomics25 (4),1341Œ1364. Yang,F.(2004). HistorialResearchonFamilyPlanningofContermporaryChina. Ph.D.thesis, ZhejiangUniversity. Zhang,X.,X.-O.Shu,Y.-T.Gao,G.Yang,H.Li,andW.Zheng(2009).Pregnancy,childrearing, andriskofstrokeinchinesewomen. Stroke40 (8),2680Œ2684. Zhao,Y.,J.Strauss,G.Yang,J.Giles,Y.Hu,andA.Park(2012).Thecharlsuserguide.Technical report,PekingUniversityChinaCenterforEconomicResearch. Zimmer,Z.andJ.Kwong(2003).Familysizeandsupportofolderadultsinurbanandruralchina: Currenteffectsandfutureimplications. Demography40 (1),23Œ44. 173