TheGermanOccupationoftheSovietUnion:

TheLong‐TermHealthOutcomes

January,2016

Preliminaryandincomplete

OlgaBelskayaUNC‐ChapelHill

belskaya@email.unc.edu

KlaraSabirianovaPeterIZAandCEPRUNC‐ChapelHillkpeter@unc.edu

ChristianPossoUNC‐ChapelHill

cpossosu@email.unc.edu

AbstractThisstudyexaminesthelong‐termhealthconsequencesoftheearly‐lifeshockscausedbytheNazioccupationoftheSovietUnionduringWWIIin1941‐1945.Wefocusonindividualswho,atthetimeof thewar,were in utero or in their early childhood until age 5, lived under occupation ofNaziGermanyandsurviveduntilage50whenwestartobservingtheirhealthoutcomes.Thestudydesignreliesontheprecisetimingofoccupation,specificgeographic locationofoccupiedmunicipalities,andtheunexpectedandrapidadvancementoftheNazyarmyintotheSovietUnion.Wetestforthepresence of the critical periods in child development affecting late‐life health outcomes such asmortality, chronic heart conditions, diseases of respiratory system, diseases of digestive andgenitourinarysystems,spinaldisorders,hypertension,depression,under‐orover‐weight,aswellassubjectivehealthassessmentsandlifesatisfaction.Theaveragetreatmenteffectisestimatedusingadifference‐in‐differenceestimatorwithafullsetofregionofbirthandbirthyearfixedeffectsandwith accounting for attrition and selective mortality bias. The spatial regression discontinuityestimator is also implemented. In addition to the average treatment effect, we recover theheterogeneity of treatment by gender, by the length of individual exposure to shocks (from onemonthtoafewyears)andbytheregionalsub‐divisiondefinedbasedonthelevelofdestructionandhumanlosses.OurpreliminaryestimatesshowthattheGermanoccupationoftheSovietUnionhadalargedetrimentaleffectonhealthofsurvivedchildrenofthewar,andthattheeffectsarethelargestiftheexposuretoshocksoccurredinuteroorininfancy.

Keywords:health,earlychildhoodshocks,difference‐in‐difference,survival,WWII,Russia,Ukraine

JELCode:

Acknowledgement:WearethankfultoMicheleMendezforusefulcomments.WealsoacknowledgehelpfulcommentsreceivedfromparticipantsoftheUNCAppliedMicroeconomicsWorkshopandCarolinaPopulationCenterSeminarSeries.Webappendix: 

1. Introduction Thereisagrowinginterestinunderstandingofhowtheformationofhumancapabilitiesin

early stages of life influences adult outcomes (Conti and Heckman, 2010). Recent economicsliteraturehasbuiltatheoreticallife‐cycleframeworkthatdrawsattentiontomechanismsthroughwhichthechangesinchildcapabilitiesandinvestmentsinchilddevelopmentdeterminefuturelifeoutcomes(seeCunhaandHeckman,2007;Heckman,2007,AlmondandCurrie,2011aand2011b).Under this framework, the literature started seeking evidence on the role of early childhood inshapinglate‐lifeconsequencesbystudyingunexpected,disastrousevents,suchaswarsandfamines,experiencedearlyinlife.1

Ourstudyexaminesthelong‐termimpactofearlylifeshocksfromtheWorldWarII(WWII)usingthecaseoftheleaststudiedandmostharmedsideoftheconflict–theformerUnionofSovietSocialistRepublics(USSRorsimplySovietUnion).Specifically,wefocusonindividualswho,atthetimeofthewar,wereinuteroorintheirearlychildhood,livedunderoccupationofNaziGermanyandsurviveduntilage50whenwestartobservingtheirhealthoutcomes.

Despitethenon‐aggressionMolotov‐RibbentroppactsignedbetweentheSovietUnionandGermanyin1939,NaziGermanyattackedonJune22,1941andwithinthenextfewmonthsoccupiedBelarus,Moldova,Ukraine,Balticrepublics,andasignificantpartofthecentralandsouthernRussia.Almost85millionpeopleor44.5percentofpre‐warpopulationofUSSRlivedintheterritoriesthatwereoccupiedduringWWII(Goskomstat,2015).TheSovietUnionsufferedimmensely.Injustfouryears,1941‐1945,around8.7millionRedArmysoldiersdiedinthecourseofthewar,11.3millioncivilian individualsdiedbecauseof themilitary activity and crimesagainsthumanity,6.5millionSovietcitizensdiedbecauseof thewar‐related famineandinfectiousdiseases. Intotal, theSovietUnionlost26.5millionpeopleinWorldWarII,or15.5percentofits1939population(EllmanandMaksudov,1994).Basedonhistoricevidence,weconjecturethatNazioccupationmayhavehadalong‐lastingeffectonhealthoutcomesofsurvivedchildrenofthewarthroughavarietyofchannels,includingmalnutrition,diseases,stress,thereducedparentalinvestmentinchilddevelopment,thelackofmaternalandhealthcare,andthedestructionofhousingandotherformsofmaterialwealth.

TheoccupationoftheSovietUnionduringWWIIpresentsauniquequasi‐experimentalsetupwithseveralimportantproperties.First,theoccupationwassharplydefinedintimeandspace.TheoccupationborderlineshowninFigure1splitsthecountryroughlyinhalfintermsofthesizeofpre‐war population. Second, there is a significant spatial heterogeneity in terms of the timing andduration of occupation, which also can be seen in Figure 1. The duration of occupation variedanywhere from 3 days to 1413 days, with the mean of 690 days across municipalities. Third,occupationwasnotanticipatedbythegeneralpopulation.BecauseWWIIstartedunexpectedlyandtheGermanoccupationoccurredrapidly,mostpeoplecouldneitherescapefromitsharshconditions,norprepare for it. Finally, livingconditionsonoccupiedterritoriesweresoseverethatsurvivedchildrenarelikelytocarrythepermanentimpactofthisadverseperiodtotheiradulthood.

Bycombiningthisquasi‐experimentalsetupwithuniquedatasourcesandrigorousempiricalanalysis,ourstudymakesseveralimportantcontributionstotheexistingliteratureonthistopic.Tostartwith,thisisthefirsteconometricstudytoevaluatethelong‐termhealthconsequencesofthenegativeshocksassociatedwiththeeasternfrontoftheWWII.Nootherstudyhasdoneitdespite

                                                            1Forexample,previousstudiesevaluated the long‐term impactof theDutch famineof1944onnumeroushealthoutcomes(Lumeyetal.,2007), famine inEuropeduringWWIIonheight(vanderBergetal.,2015),WWII inEuropeonhealth(Kesternichetal.,2014),bombinganddestructionof infrastructure inGermanyduringWWIIonheight,mortality,obesityandchronicdiseases(Akbulut‐Yuksel,2014and2015),theKoreanWarof1950‐1953ondisabilityandmortality(Lee,2014),theChinesefamineof1959‐1961onheight(ChenandZhou,2007;Gorgensetal.,2012;MengandQian,2009),obesity(Fung,2009),disability(Almondetal.,2010),andmentaldisorders(St.Clairetal.,2005),theNigerianCivilWarof1967‐1970onheight(Akresh,2011),andtheEthiopianfamineof1984onheight(DerconandPorter,2010). 

thefactthat“thewarassumeda fargranderscale intheEastthaninanyofthe frontswheretheWesternAllieswereinvolved”(Davies,2006).

WeutilizethedetailednationallyrepresentativesurveysofthetwolargestcountriesoftheSovietUnion,Russia andUkraine,2 and link these surveyswith thenovel dataset on thedatesofoccupationandliberationofUSSRmunicipalitiesandalsowiththeregionalmeasuresofdestructionandhumanlossesduringWWII. Thetwosurveysourceshaveneverbeenpreviouslyused intheWWIIcontext,whilethemunicipality‐leveldatahavebeencollectedbytheauthorsinthecourseofthisresearchproject.

Inouridentificationstrategy,werelyontheindividualdateofbirth,municipalityofbirth(cityorcounty),andexogenousvariationinthelocationandtimingoftheGermanoccupationofeachmunicipality duringWWII. Thus,wemeasure an individual‐specificexposure to treatment at thehighlydisaggregatelevel,whichisanimprovementoverpreviousstudiesthatidentifytreatmentatthecountrylevel(HavariandPeracchi,2015;Kesternichetal.,2014;VanbenBergetal.,2015,usingthe survey of European countries SHARELIFE) or region/province level (Akbulut, 2014, usingGermanSocio‐EconomicPanel). Anotherimportantfeatureofourstudydesignisacontrolgroupcomprisingofindividualsborninnon‐occupiedmunicipalitiesduringthewaraswellasindividualsbornafterthewarinformerlyoccupiedandnon‐occupiedmunicipalities.Thisallowsustoexplorethedisaggregatedspatialandtemporalvariationintreatmentsimultaneously.

Thisstudyalsocontributestothegrowingliteraturethatcomparesthelong‐termeffectsofshocksacrossdifferentstagesofchildhood.Mostofthepreviousstudiesfocusonspecificstages,suchas in‐utero (Lee,2014), infancy (Ampaabeng andTan,2013;Portraitetal., 2005), or childrenofschoolage(Akbulut,2014).Wedistinguishfourdifferentstages:in‐utero(fromconceptiontobirth),infants(frombirthto12monthsofage),toddlers(1‐3yearsold),andpreschoolers(3‐5yearsold).Thisallowsustotestforthepresenceofthecriticalperiodsinchilddevelopmentaffectinglong‐termhealth. We find astonishing heterogeneity in the treatment effect by age, with kids exposed toadverseeventsduringtheprenatalstageoflifeenduringthelargestnegativeconsequences.

Unlikemanyexistingstudies,ourmeasureofthe intensityofhealthshockvarieswiththedateofconceptionandbirth,which,togetherwiththedatesofoccupation,determinethelengthofindividualexposure tohardships in theearly childhoodunderNazioccupation. Additionally,werecovertheheterogeneityoftreatmentbygenderandbytheregionalsub‐divisiondefinedbasedonthelevelofdestructionandhumanlosses.

2. Background OnJune22,1941,NaziGermanylaunchesOperationBarbarossa–aninvasionoftheSoviet

Union by three million German soldiers and half a million troops from German allies (Finland,Romania,Hungary,Italy,Slovakia,andCroatia)alonga1,800milesfrontextendingfromtheArcticOceantotheBlackSea.Barbarossawasthelargestmilitaryoperationintheworldhistoryinbothmanpowerandcasualties,andregionsaffectedbyBarbarossawerethesiteofsomeofthelargestbattles,highest casualties, andmosthorrific conditions. German forceswereorganized into fourarmies: army Norway operated in far northern Scandinavia; army North aimed at taking ordestroyingthecityofLeningrad(nowSt.Petersburg)bymarchingthroughtheBalticsintonorthernRussia;armyCenterwasplannedtomarchthroughpresentdayBelarusandtakeMoscow;andarmySouthaimedatstrikingUkraineandtakingthecontrolovertheoil‐richCaucasus.ThemainobjectivewastoreachthesocalledA‐Aline(Arkhangelsk‐Astrakhan):

“InquickpursuitalineisthentobereachedfromwhichtheRussianAirForcewillnolongerbeabletoattacktheterritoryoftheGermanReich.Theultimateobjectiveoftheoperationistoestablish

                                                            2 The RussianLongitudinalMonitoringSurvey–HigherSchoolofEconomics(RLMS‐HSE)andtheUkrainianLongitudinalMonitoringSurvey(ULMS),respectively.

a cover against Asiatic Russia from the general line Volga‐Archangel.” (German Archive, Hitler’sDirectiveNo.21OperationBarbarossa,December18,1940)

ByearlySeptember1941,GermanforcesencircledLeningradinthenorthandtookSmolenskinthecenterandDnepropetrovskinthesouth.InearlyDecember1941,theycameclosetothecityofMoscowbutwerequicklydrivenawayfromthecountrycapitalbytheRedArmy.Inthesummerof1942,Germanforcesattackedinthedirectionof thecityofStalingrad(nowVolgograd)ontheVolgaRiverandtheoilfieldsoftheCaucasus.Thus,inSeptember1942,theGermanoccupationofthe Soviet Union reached its furthest geographical extension, whichwas still far away from thedesiredA‐A line,aswecansee fromFigure1. Theendof1942became the turningpointof theGermaninvasionoftheSovietUnion.SixweeksofferociouscombatnearthecityofStalingradledtoheavycasualtiesonbothsides,themajorvictoryoftheRedArmy,andthebeginningofliberationoftheSovietterritories.During1943,theRedArmyclearedGermanforcesfrommostoftheterritoryof Russia, eastern Belarus andUkraine. In 1944, thewestern parts of Belarus and Ukraine, thenorthwesternpartofRussia,Moldova,andmostoftheBalticcountrieswereliberated.

TheGermanoccupationoftheSovietUnionwasthemostbrutalanddestructiveoccupationinEurope.AccordingtoNaziplans,JewishpopulationandmembersoftheCommunistPartyweretobe annihilated, Slavic population significantly reduced as part of the Hunger Plan, and Sovieteconomicresourcesdestroyedorexpropriated. Wehavealreadymentionedthe immense lossofmanymillionsof lives. Inaddition,during theoccupationof theSovietUnion, theNazi invaderstotally or partially ruinedandburned1710 cities and townships andmore than 70,000 villages,burnedanddestroyedmore than6millionbuildingsand31,850 industrialenterprises, renderedaround 25million persons homeless, and caused great damage to the infrastructure and publicservices(Goskomstat,2015;mostnumbersaretakenfromtheNurembergTrial).

Here,weonlyhighlightsomeofthemainchannelsthroughwhichtheoccupationmayhavehadaneffectonthelong‐termhealthoutcomesofsurvivedchildrenofthewar.

Malnutrition – According to the “Hunger Plan” developed by Germans in early 1941 andsubsequentlyimplemented,theSovietUnionwasdividedintotwoagriculturalzones–adeficitzone(BelarusandNorthernandCentralRussia)andasurpluszone(Ukraine,SouthernRussiaandtheCaucuses).Germanyextractedfoodproductsfromthesurpluszoneandsealedoffthedeficitzonefromgettingfoodsupplies(Gerhard,2009).Largeindustrialandurbancentersinthesurpluszonewerealsocutofffromfoodsupplies.Thus,famineonamassscalewasunavoidableandengineeredwith expectations that tensofmillionsof Soviet peoplewouldbecome superfluous anddie fromstarvation. As a result of inadequate food rationingandblockading food supplies, urban civilianpopulation plummeted.3 Among survivors, the short‐term health effects from undernourishmentwereimmediatelynoticeable. Forexample,inLeningrad,theaveragebirthweightwas100gramslessin1941comparedtothebirthweightin1938‐1940.In1942,thebirthweightdeclinedfurtherby600grams,averageheight–by2cm,chestcircumference–by1.5cm,andheadcircumference–by1.3cm(GeorgievskyandGavrilov,1975).Theseandotherchangesinhumanbodyduringprenatalandearlychildhoodstagesoflifemayhaveproducedlong‐lastingeffectsonhumancapabilitiesinadulthoodandinfluencedsubsequentlifequalityandduration.Unfortunately,scientificevaluationofthelong‐termhealtheffectsoftheGermanhungerpolicyintheUSSRhasneverbeenattemptedonalargecountryscale.4

                                                            3Siegedfor872daysbetweenSeptember1941andJanuary1944,Leningradexperiencedthelongestandthemostdestructivesiegesintheworldhistoryleadingto1,500,000deathsduetothefamineandartilleryshelling(Salisbury1969).PopulationoftheUkrainiancityofKharkovdroppedfromamillionbeforethewarto250,000two years later, which cannot be attributed to migration as entering and leaving the town was strictlyprohibited(Mazower2008).4Afewepidemiologicalstudiesexaminedthelong‐termconsequencesoftheSiegeofLeningradonmortalityandhealthoutcomesusingthesampleofLeningradresidents(Koupiletal.,2009;Sparenetal.,2003;Stanner

Diseases – It is known that undernourishment can make a person more susceptible toinfectionandotherdiseases.Inadditiontopoornutrition,otherwar‐relatedfactorssuchasuncleanwater, poor sanitation, inadequatehealth care, and the lackof shelterprovide conditions for thespread of infectious diseases during the wartime. The available scarce data on epidemiologicalsituation in occupied territories of the Soviet Union indicate a dramatic rise in the incidence ofdiseasesduringtheoccupationperiod.Forexample,in1944comparedtothepre‐waryearof1940,the registered cases of illness in Belarus increased by 45 times for typhus, 5 times for sexuallytransmitteddiseases,4timesformalaria,2.6timesfortyphoidfever,and2timesfortuberculosis(Logvinenko,2010). Basedonrobustresults instudiesof the1918 influenzapandemic(Almond2006,LinandLiu,2014),itisplausibletoexpectthattheincreasedexposuretoinfectiousdiseasesinearlychildhoodduringWWIImayalsohavelong‐lastinghealthconsequences.

Stress–Thewarexperienceisundoubtedlytraumaticonmanylevels;itmaytriggerstressreactionsinchildrenandberesponsibleforavarietyofhealthproblemslaterinlife.Themostlikelystress triggersduring thewar include thedeathofparents, separation fromparentsdue to theirinvolvementinmilitaryservice,witnessingpublichangingsandotherformsofviolence,experiencingphysical abuse by German soldiers, starvation, watching parental stress reactions, etc. In ourUkrainiansurvey,almost26percentoftherespondentsbornin1936‐1945reportnothavingafatheraliveatage14;4percentdidnothaveamotherwhentheywere14yearsold.Inlatercohorts(bornin1949‐1958), thesenumberschangedramatically–only5percenthadadeceased fatherand1percentadeceasedmotheratage14.Alargenumberofexperimentalandepidemiologicalstudiesdemonstratethathealthinthelaterstagesoflifemaybestronglyinfluencedbystressfulexperiencesinprenatalandearlychildhoodstagesoflife(seereviewofstudiesinVaiserman,2015).

Reducedparentalinvestmentinchilddevelopment–Thelossofaparentorabsenceofaparentonwarservicenotonlytriggersstressinchildren,butalsopreventsparentalinvolvementduringcriticalperiodsofearlychilddevelopmentandthusdiminishchild’s futurecapabilities. Even forchildrenwhoseparentswerepresentduringthewar,parentalinvestmentintochilddevelopmentislikelytobereducedduetodeterioratedparents’health,insufficientparentalincome,starvation,andotherreasons.Previousliteraturefindsthatchildren’shealthisnegativelyassociatedwithparentaldeath(Beegleetal.,2010;Corak,2001;Krause,1998),parentaljobloss(Lindo2011),andadverseincomeshocks(Caseetal.,2002)

Lackofmaternalandhealthcare–Thedestructionofhospitalsandclinicsalongwithbarredmedicalsuppliesisanotherchannelthatmaycontributetopoorhealthoutcomeslaterinlife.DuringtheGermanoccupationoftheSovietUnion,40,000healthcarefacilitiesweredestroyedorburned(Goskomstat,2015).AsofJanuary1,1945,formeroccupiedterritoriesoftheSovietUnionwereleftwith48percentofpre‐warbirthcarecenterbedsand52percentofchildrenhospitalbeds. Themedicalstaffwasseverelydownsized.Bytheendof1944,thetworepublicsthatwerefullyoccupied,BelarusandUkrainecombined,hadonly46percentofdoctorsfromthepre‐warlevel,42percentofgynecologists,39percentofpediatricians,54percentofnurses,and46percentofmidwifes(HealthCareintheUnionofSSR,1946).Eventhoughamethodologicallysolidpaperonthelong‐termhealthconsequencesofthedestructionofthehealthcaresystemisyettobewritten,thecausallinkhereisapparent.

Destructionofthematerialwealth–Manymaterialassetsthatareessentialforchildrenhealth–suchaswatersupplies,sewer,heatingsystemsneededforsurvivalinthewinter,andhousing–wereseverelydamagedduringtheGermanoccupation.Over50percentofhousesandapartmentbuildingsinoccupiedcitiesweredestroyed(Voznesensky,1948).InsomecitieslikePskov,Vitebsk,

                                                            etal.,1997;Vageroetal.,2013).Theresearchdesignwasnotidealbytoday’sstandard;forexample,thecontrolgroupinadvertentlyincludedindividualsthatcouldhavesufferedfrommalnutritioninotheroccupiedareasofthecountryandalsoindividualsexposedtothepost‐war1946‐47famine.Thefindingsfromthesestudiesareofteninconclusiveandsometimesconflicting.

Voronezhandothers,theshareoflivablehousingslumpedtolessthan10percentofthepre‐warhousingstock.Peoplehadtoliveinsheds,bunkersanddangerousunheatedhousingconditions.Thedestructionofotherformsofmaterialwealth‐electricity,communication,transport,manufacturing,agriculture, etc. – was also devastating (see Voznesensky (1948) for aggregate statistics ondestruction). Production ofmaterial goodsmay have an indirect effect on children health, as itprovides parentswith jobs and income, produces and supplies food andmedicine, and providespeoplewithotherbasicessentials.

TheabovelistofchannelsthroughwhichtheNazioccupationmayhaveimpactedchildrenofthewarisnotcomplete,butitshowsthatthetimeofoccupationwasdevastatingforpeopleandthatthechannelsarenumerousandintertwinedwitheachother.Forthedisasterofthismagnitudesuchas WWII, disentangling the impact of a specific channel is methodologically unfeasible due tochannels’ complementarity and reciprocity. What is feasible is to recover the treatmentheterogeneitywithrespecttodifferentmargins,includingchild’sageatwhichtheimpactoccurred,thelengthofindividualexposuretowarhardships,andtheextentofregionaldestructionandhumanlosses.Therefore,inwhatfollows,ourtreatmentvariablerepresentsacombinationofallchannels,butitisinteractedwithvarioustreatmentmargins.

3. Theoretical Framework

3.1.GeneralSet‐UpInthissection,wedescribethetheoreticalframeworkthatguidesourresearchprocess.We

modelhealthoutcomesusingthehealthproductionfunction,inwhichhumancapabilitiesdeterminehealthoutcomesateachstageof life. FollowingContiandHeckman(2014),CunhaandHeckman(2007)andHeckman(2007),weassumethatthehealthoutcomeisafunctionofasetofhealthandotherhumancapabilitiescreatedinamulti‐stagetechnology.5Thekeycharacteristicofthemodelisthat each stage of life may have different technology and/or the set of human capabilities.Furthermore,theearlystagesofchilddevelopment(e.g.,inutero,infancy)affectlaterstagesoflife.Inthismodel, thekeymechanismsthroughwhichtheNazioccupationoftheSovietUnionaffectsadult health outcomes are dynamic complementarity and self‐productivity. Dynamiccomplementarity means that health and other capabilities produced at one stage increase theproductivityof investment inhealth andother capabilities at subsequent stages.Self‐productivityimpliesthathealthandothercapabilitiesproducedatonestageaugmentthecapabilitiesgeneratedatlaterstages,whichinturnimproveshealthandsurvival.

Ourmodelisasimplifiedversionoftheoverlappinggenerationsmodel(CunhaandHeckman2007).Wemodelsixstagesoflife:in‐utero( 0),infancy( 1),toddler( 2),preschool(3),youngadulthood( 4),andadulthood( 5).Thefirstfourstagesfromconceptiontoage5arethemainfocusofourresearch.AsAlmondandCurrie(2011a,2011b)argue,theadverseshocksexperienced in utero and during early childhood may be more influential than shocks in laterchildhood.Thesefourstagesareoftendubbedasthecriticalperiodofchilddevelopment.AccordingtoHeckman(2007),acriticalperiodisastageinthelifespanwhenlate investmentisnotabletocompensateforthelackofearlyinvestment.

The thhealthoutcomeattime , ,isfunctionofavectorofhealthandotherindividualcapabilities, :

(1)Forexample,the thhealthoutcomeinadulthoodisgivenby .Theproductiontechnologyofcapabilitiesatstage 1is

                                                            5Analternativemodelfortwoperiods(childhoodandadulthood)ispresentedinAlmondandCurrie(2011b).Amulti‐stagetechnologyfitsbetterwithourempiricalfindinginthatthedifferentstagesofchildhoodatwhichtheshockoccurredhavedifferentialeffectsonfuturecapabilities.

, , , (2)where ∙ is increasing and twice continuously differentiable in all arguments; denotesinvestmentinhealthandothercapabilitiesmadebyparentsinearlystages,anindividualhimselfinlaterstages,community,andgovernmentinperiod , denotesexogenousparentalcapabilities,suchasgenesorparenteducation,and denotesconstantcommunitycharacteristicsaffectingcapabilitiessuch as climate, regional food practices, water sources, urban density, etc. define the initialcapabilitieseachindividualisbornwithandwhichareformedinutero.

Tomakethetechnologyfunctionforcapabilitiesmoreflexible,wecanaddagroup‐specificproductivityshifter forgroupj:

, , , (2b)Forexample,aproductivityshiftercanaccountforgenderdifferencesinthegenerationofhumancapabilities. In fact, our empirical exercise finds significant differences in the effect of early‐lifeshocksonfutureadultcapabilitiesbetweenmalesandfemales.

If currenthealth andother capabilities createhigher capabilities in thenextperiod, then

technology ∙ isself‐productive( 0).Similarly,ifhealthandothercapabilitiesaccumulated

until the previous stage complement investment at the current stage, then technology ∙ is

complementary ( 0). This complementarity is dynamic if early investment makes later

investmentmoreproductive.Substituting , … , repeatedlyfor inequation(2),humancapabilitiesinadulthoodcan

bewrittenasafunctionofparentalcapabilities,healthandothercapabilitiesformedin‐utero, ,andallsubsequentinvestments:

, , , , , , , (3)Usingequation(3),the thhealthoutcome inadulthoodisgivenby

, , , , , , , (4)ThroughthechannelsdiscussedinSection2,theformationofhumancapabilitiesduringthe

initialstagesoflife( , 0, 1, 2, 3)andinvestmentatthesestagesarelikelytobeadverselyaffectedbywar‐relatedexogenousshocks, (i.e., ⁄ 0and ⁄ 0).Thesubscript indicatesthattheshockcouldvaryatsomegeographiclevel,whichmayprovideanexogenousvariationininitial capabilities and early childhood investment. In particular, the intensity of shock, ,maydependuponthedurationofoccupationorthelevelofdestructionthateachcity/regionsuffered.Thetechnologyfunction doesnotneedtobelinearandmayallowfornon‐lineareffectsoftheintensityofshock, ,withineachstageoflife.

BylinkingthediscussioninSection2withtheabovetheoreticalframework,wecanhighlightatleastthreelong‐termcausalmechanismsthroughwhichtheNazioccupationoftheSovietUnionmayhaveaffectedhealthinthelaststageofhumanlifeoradulthoodinthemodel( = ).

⟹ 1 ↓ ⟹↓ ⟹ ⋯⟹↓ ⟹↓ 2 ↓ ⟹↓ ⟹↓ … ⟹↓ ⟹↓ 3 ↓ ⟹↓ ⟹ ⋯⟹↓ ⟹↓

First, the occupation may lower health capabilities in‐utero and in early childhood, forexample,viamalnutrition,wartimediseases,physicalabuse,injuries,andhighstresslevels.Sincetechnology is self‐productive, the deteriorated capabilities in early stages affect the formation ofcapabilities in later stages, including the last stage, , and through this channel reduces healthoutcomesinadulthood.Theliteratureinepidemiologyandmedicinehasshownthatin‐uteroandinfancyarecriticalperiodsforthephysical,cognitive,sensoryandmotordevelopment,andadverse

environmentsduringthistimehavelastingeffectsonhealthoutcomes(Barker1990,1998,Karpetal1995,GluckmanandHanson2005,Schultz2010).6

Second,giventhecomplementaritybetweenthepreviousaccumulationofhumancapabilitiesand current investment in health, a significantwar‐caused decline in health capabilities in earlystages of child development reduces the productivity of investments in later stages (i.e.,, , , , ),andthroughthischannelnegativelyaffectshealthoutcomesduringadulthood.

Third,theoccupationdirectlyreducesparental investmentduringchildhoodasaresultofthelossofaparent,theabsenceofaparentonwarservice,deterioratedparents’health,thelackofparental jobs and income, and destroyed family housing. The community and governmentinvestments in health are also harmfully affected, leading to the lack of basic utilities andinfrastructure,poorsanitation,andthelackofmaternalandhealthcare.Thesereducedinvestmentsduringearlychildhoodarelikelytolowerhealthoutcomesinadulthoodthroughthechannelsofself‐productivityanddynamiccomplementarity.

3.2.EmpiricalModel

ThemainobjectiveoftheempiricalmodelistoestimatetheeffectofshockstriggeredbytheNazioccupationof theSovietUniononhealthoutcomes inadulthood. Theempiricalmodel isanapproximationofequation(4).Itislinearinparameters,butnotnecessarilylinearinvariables.

Our unique quasi‐experimental setup exploits the individual date of birth (d), themunicipalityofbirth(m),andexogenousvariationofthelocationandtimingofGermanoccupationduringWWII.Theidentificationofthetreatmenteffectreliesonthisquasi‐experimentalvariationandonafullsetofregionofbirth(r)andbirthyear(t)fixedeffects,whererisanext‐levelaggregationofm,andtisanannualaggregationofd.7

Thestartingspecificationisthefollowing: (5)

where isalate‐lifehealthoutcomeobservedinsurveyyearsforanindividualiwhoisborninyeartandbirthregionr, isthebirthregionfixedeffect, isthebirthyearfixedeffect, istheinteractionofthebirthregionfixedeffectwithsomefunctionalformfort, istheyear‐of‐survey fixedeffect, isavectorofobservable individualcharacteristics,and isarandomerrorterm.

Ourtreatmentvariable isadummyvariablethattakesthevalueof1forindividualswhowere born in an occupiedmunicipality of the Soviet Union andwhowere in a specific stage ofchildhoodunderoccupation. Thecontrolgroupincludestwotypesof individuals:(1) individualswhowereborninnon‐occupiedmunicipalitiesandwhowereinaspecificstageoflifeduringthewarperiod; and (2) individuals born after the war in both formerly occupied and non‐occupiedmunicipalities. Further specifics on definitionswill be discussed in Section4. Bothparts of thecontrolgrouparenecessary,astheyprovidespatialandintertemporaldimensionsforthedifference‐in‐differenceestimation.Withoutindividualsbornafterthewar,itwouldnotbefeasibletoisolatetheeffectsofNazioccupationfromotherspecificcharacteristicsofregionssuchasfactorzincludedin the production technology of capabilities at each stage given by equations (2)‐(4). Includingindividuals born before thewar into the control groupwill be inconsistentwith our theoreticalframework.Sinceinvestmentandcapabilitiesofpre‐warcohortsarelikelytobeharmedbythewaratlaterstagesofchilddevelopment( , , , ),thesewarshocks,viaself‐productivityanddynamiccomplementarity,maygenerateanadverseeffectonlate‐lifehealthoutcomesofpre‐warcohorts.

                                                            6Forinstance,environmentalconditionsinuteromodifytheepigenomeandthroughthischannelincreasetheriskofchronichealthconditionsinadulthood(Thornburgetal.2010,Petronis2010).7Thecombinationofmanddinourdataidentifiesauniqueindividualifor99.6percentofsurveyedindividuals.Thecoincidenceofseveralrespondentsbeingbornonthesamedayinthesamecityisnegligible.Thus,usingthesubscriptiisequivalenttousingthesubscript(md)inourcase.

Here we are interested in estimating . Region‐of‐birth fixed effects account for thesystematicdifferencesinhealthoutcomesacrossregions,whileyear‐of‐birthdummiescontrolforthepermanentdifferencesinhealthoutcomesacrossbirthcohorts.Theinteractionterm is included to account for time‐varying region‐specific policies (e.g., different speeds of post‐warreconstruction and development across regions). Year fixed effects capture country‐widecontemporary shocks in health outcomes. To avoid including endogenous individual‐levelcovariates,welimitthevector togender,ethnicity,typeofthebirthsettlement,andparents’education.Thelattervariableservesasaproxyforparentalcapabilitiespinequations(2)‐(4).

Ourmainidentifyingassumptionthat isneededtoconsistentlyestimate isthatoncewecontrolforthevectorofobservableindividualcharacteristics ,asetofregion‐of‐birthandyear‐of‐birthfixedeffects , ,theinteractionterm ,andannualmacroshocks ,theerrorterm isuncorrelatedwiththeNazioccupation.Intheend,ouridentificationstrategyreliesonwithin‐regioncross‐cohortvariationtoidentifytheeffectsofNazioccupation.

Specification(5)isestimatedforeachofthefourstagesoflifeconsideredinourtheoreticalframework:in‐utero,infants,toddlers,andpreschoolers.Wefurthertestforgenderdifferencesintreatmenteffectsbyincludingatwo‐wayinteractionbetweenthetreatmentvariableandtwogendercategories, :

. (6)Following our theoretical framework, we also estimate the model using three different

proxiesfortheintensityofshocks, :thedurationofoccupation,thelengthofexposuretoshocks,andthelevelofdestruction.Wecalltheseproxiesastreatmentmargins.

, (7)where istheintensityofshockexperiencedbyachildbornondaydinmunicipalitym.

The intensityofshockmaydependupon thedurationofoccupation ineachmunicipality.Thismarginvariesatthemunicipalitylevelmbutnotatthedlevel.Forinstance,wecantestifthelate‐life health outcomes for infants born during occupation in cities like Stavropol which wasoccupiedfor169daysarethesamethanforinfantsbornunderoccupationincitieslikePskovwhereoccupationlastedformorethanthreeyears(1110days).

Theintensityofshockmayalsovarywiththedateofconceptionandbirth,which,togetherwiththedatesofoccupation,determinethelengthofindividualexposuretoadverseshocks.8Thismarginvariesatbothlevelsmandd.IntheexampleofthecityofPskov, ’smightbedifferentforindividualsconceivedatthebeginningofoccupationwithalongerexposuretoshockscomparedtoindividualsconceivedattheendoftheoccupationwithashorterperiodofexposureinthesamecityofbirth.

Finally,athirdmargincapturesnon‐lineareffectsassociatedwiththespatialdifferencesinthelevelofdestructionandpopulationlossesthateachcity/regionsuffered.Thismarginvariesatthelevelofmonly.

Thefactthatweobserveindividualsaffectedbythewarinlaterstagesoftheirlife(ages50‐77) has two important methodological implications. First, captures the treatment effect onsurvivors.Itislikelytobebiasediftheprobabilityofsurvival(andthustheselectionofanindividualintoourestimationsample) ispositivelycorrelatedwith inhealthoutcomes. Wecanpartlyaddress this concern for those survived till the first year of the survey bymodeling subsequentsurvivalbetweenages50and77andre‐estimatingequation(5)withtheinversesurvivalprobabilityweights.Second,healtheffectsofthepastcompensatoryinvestmentmadebyanindividual,parentsorcommunitiestooffsetnegativeearly‐lifeshocksaregoingtobeabsorbedinthetreatmenteffect.Even though the remediationof adverse shocks is generally found tobe lesseffective after early                                                            8Thedurationoftheconflictisthemostcommonmarginusedintheliteratureassociatedwiththelongtermeffectsofwarsorviolentconflicts. This margin does not necessarily reflect the real exposure of an individual to the conflict. 

childhood,someevidenceofcompensatoryinvestmentexistsespeciallyinnon‐cognitiveskillssuchasperseverance,motivation, self‐esteem, self‐control,etc. (Conti andHeckman,2014). Assumingsomepersistenceinnon‐cognitiveskillsoverthelifespanandusingthesamefunctionalform(5),wecan test if individuals in the treatment group have developed a different set of non‐cognitivecapabilities.

4.Data In this section, we describe survey data, define “treated” municipalities, and discuss thecomposition of treatment and control groups. We also introduce various margins of treatmentintensity, including municipality‐specific duration of occupation, individual‐specific length ofexposuretooccupation,andregion‐specificmeasuresofdestructionandpopulationloss.Next,weprovidethet‐testresultsforthepre‐treatmentconditionsattheregionallevel.Weconcludethedatasectionbypresentingsummarystatisticsonlong‐termhealthoutcomesandmodelcovariates.

4.1.SurveydataOurstudydrawsontwolongitudinalsurveysofRussiaandUkraine–RLMS‐HSEandULMS,

respectively.TheRLMS‐HSEisoneofthelongestnationallyrepresentativelongitudinalsurveysofhouseholds.9Themainsampleofhouseholdsissurveyedannuallyfrom1994andonwards,withtheexceptionofyears1997and1999.Thenumberofrespondentsperyearfluctuatedbetween10,500and14,700individualsuntil2009androsetomorethan20,000respondentsin2010‐2013.Weuseallsurveywavesbeginningyear2000whentheRLMS‐HSEaddedmanyhealth‐relatedquestions.

The ULMS is a household and labor force survey based on a statistically representativesampleoftheUkrainianworking‐agepopulationbetweentheagesof15and72.10 TheUkrainiansurveystartedin2003andrepeatedin2004,2007,and2012.Thedatafromthe2012surveyisnotyetpubliclyavailableandwillnotbeusedinourstudy.DuringthefirstthreewavesoftheULMS,9,902individualsfrom4,232householdsparticipatedinthesurvey.

Themostrecentdetailsonsurveyinstruments,sampledesign,andattritioninbothRussianandUkrainiansurveyscanbefoundinGerryandPapadopoulos(2015),KozyrevaandSabirianovaPeter(2015),andLehmannetal.(2012).Theattritionduetoaging(i.e.,reachingtheupperagelimitof72)inULMSdoesnotrepresentaproblemforus,asitconcernspeoplebornbefore1935.Theattritionduetomortalityisanimportantissueinbothsurveysandwillbeaddressedusingastandardtoolofinversepropensityweighting.

Theresearchdesigndescribedintheprevioussectionrequireshavingsomenon‐occupiedterritories in the controlgroupand thus canbeappliedeither toRussiaaloneas theonlySovietrepublicthatwaspartiallyoccupiedortothecombinedRussia‐Ukrainesample.TheULMSemulatedmanyquestionsfromtheRLMS‐HSE,hencemakingiteasiertoappendthetwosurveys.Weaccountfordifferentsamplesizesanddifferentnumberofsurveywavesintwocountriesbyre‐weightingthecountry composition of the sample to match the country composition of the combined generalpopulation,ofwhich75percentresidedinRussiaand25percentinUkrainein2000‐2013.

                                                            9 The RLMS‐HSE is organized by the National Research University Higher School of Economics, MoscowtogetherwiththeCarolinaPopulationCenterattheUniversityofNorthCarolinaatChapelHillandtheInstituteofSociologyattheRussianAcademyofSciences.TheRLMS‐HSEsurveyedindividualsin32outof83regionsinallsevenfederaldistrictsoftheRussianFederation(accordingtotheofficialclassificationofregionsasofJanuary1,2010).10 The ULMS is organized by the Institute for the Study of Labor (IZA) in collaboration withmany otherinstitutions,includingtheEuropeanUnion,theWorldBank,theWilliamDavidsonInstituteoftheUniversityofMichigan,andothers.ThedatacollectionwasentrustedtotheKievInternationalInstituteofSociology.Thesurveywasdoneinall27regionsofUkraine.

Westartwiththecohortofrespondentswhoarebornbetween1935and1958andwhosemunicipalityofbirth(cityorruralcounty)isknownwithcertainty.InULMS,thequestionsaboutthecountry,county,andsettlementofbirthwereaskedinanopen‐endedformin2003and2007.Thisinformationisusedtoinferthebirthplaceforthesamerespondentsinthe2004survey.Thus,wehaveanearlyfullcoverageinUkraine.Theresponserateisveryhigh–94percentofobservationsfrom the 1935‐1958 birth cohorts have complete information on the municipality of birth.11Unfortunately,inRLMS‐HSE,anopen‐endedquestiononbirthplacewasonlyaskedonce,in2013.Inallotheryears,themunicipalityofbirthcanbeinferredeitherfromthe2013surveyiftherespondentparticipatedinearliersurveysorfromthecurrentresidenceiftherespondentisborninthesameplaceasherpresentmunicipality.Outof68,605observationsfromthe1935‐1958birthcohortsandfromallRLMS‐HSEsurveyyears,70percenthavetheidentifiablenameofbirthcityorbirthcounty;5 additional percent canbeunambiguously classified intoeitheroccupiedornon‐occupiedareasbasedontheinformationprovidedontheregionorrepublicofbirth;theremaining25percentwithunknownbirthplacehadtobedroppedfromtheanalysis.

4.2.TreatmentattheMunicipalityLevelandtheDurationofOccupationInourstudy,theword“treatment”hasanegativeconnotationofbeingsubjectedtoGerman

occupationinchildhood.Themunicipalitywhereapersonisbornisclassifiedasbeing“treated”ifitwas(i) fullyoccupiedbyGermanforcesandtheirallies, (ii)partiallyoccupiedsuchas thecityofStalingrad (now Volgograd), or (iii) sieged such as the city of Leningrad (now St. Petersburg).AlthoughLeningradhadnotbeenoccupied,its900‐daysiegeclaimed650,000Leningraderlivesin1942 alone, mostly from starvation, disease, and shelling from German artillery (EncyclopediaBritannica).Basedontheseandmanyotherhorrificfactsonthedeadliestsiegeinhistory,Leningradclearly belongs to the “treated category”. Some ambiguitymay arisewith respect to other non‐occupiedcities,whichwereseverelybombardedbutnotencircled,suchasthecityofMoscow.Wedecidednottoincludethebombardednon‐occupiedcitiesintothetreatmentgroupsincetheirfoodsuppliesandevacuationroutesremainedintact.

Weassigntheoccupationstatustogeographicunitsinthe2011GADMspatialdatabaseofGlobalAdministrativeAreasoftheformerSovietUnioncountries.Thegeographicunitsarechosenatthelevelofmunicipalitythatcouldbeeitheracityorruralcounty.12Altogetherwehave3,857municipalities. This number excludes uninhabited geographic areas, within‐city districts, andmunicipalitiesforwhichtheoccupationstatusisambiguoussuchasKaliningradregion,KlaipedainLithuania,SouthernSakhalin,andTranscarpathia,whichwereacquiredbytheSovietUnionin1945asaresultofWWII.

Next, for eachoccupiedmunicipality,we collect informationon thebeginning and endofoccupation.WeaddruralmunicipalitiesandsmallercitiestothepreviouslypublisheddatabaseonthedatesofoccupationandliberationoftheUSSRcitiesduringtheWWII(Dudarenkoetal.,1985).Figure1plotsthelengthofoccupationindaysonthecontemporarymapoftheformerSovietUnioncountries. Thismap is summarized inTable 1. Almost 37percent of 3,858municipalitieswereoccupiedbytheGermanarmy.Theoccupationlastedforlessthanonemonthin43municipalities,includingsomecountiesclosetotheneversurrenderedcitiesofMoscowandTula;1to6months–in200municipalities,includingStalingradandnearbyareasandalsolocationsintheNorthCaucasus;morethan6monthsbutlessthanayear–in113municipalities,suchasthecityofRostov‐on‐Don

                                                            11Also,1percentarebornoutsidetheUSSRandexcludedfromtheestimationsample;2.5percenthavemissinginformationonthemunicipalityofbirth,buttheanswersprovidedonthecountryandregionofbirtharestillsufficientenoughtodetermineiftheplaceofbirthwasoccupiedornotduringtheWWII,andonly2.5percentofobservationsarediscardedbecauseofrefusalsorunclearanswers.12 Further details on the GADM database and geographic coding used in this paper are provided in webAppendix1.

andsomeareasontheEasterncoastoftheBlackSeaandtheSeaofAzov.Abouttwothirdsofalloccupiedmunicipalitiessufferedundertheoccupationformorethanayear–theoccupationlastedfor 12 to 24 months in 257 municipalities of the Central Russian Upland and Eastern Ukraine,including cities of Belgorod, Donetsk, Kharkov, Kursk, and Voronezh; 24 to 36months – in 565municipalities,includingwell‐knowncitiesofKiev,Leningrad,andSevastopol;and3yearsormore–in241municipalitiesofWesternBelarus,WesternUkraine,theBaltics,Moldova,andNorthwestRussia.

We use the same GADM classification of Global Administrative Areas to geocode therespondents’answerstoopen‐endedquestionsonbirthplace;thegeocodingprocessisdescribedinwebAppendix1.ThegeocodedindividualobservationsinRLMS‐HSEandULMSarethenlinkedtoexternaldatabases,includingtheoneonthedatesofoccupationandliberation.Table1providesthedistributionofmunicipalities inthecombinedRLMS‐ULMSsamplefrom2000‐2013surveys. Thenumberofmunicipalitieswhererespondentsarebornisfairlylarge(morethan2,000),andmostofmunicipalitiesarelocatedinRussiaandUkraine.Theshareofoccupiedmunicipalitiesinoursampleis45percent,whichislargerthanthatfortheentirecountryoftheSovietUnion,butthedistributionofmunicipalitiesbythedurationofoccupationissimilar.

4.3.TreatmentattheIndividualLevelandtheLengthofIndividualExposureThetreatmentgroupisdefinedattheindividuallevel.Itisconstructedbasedonthebirth

date(day,month,andyear)andthedatesofoccupationofthemunicipalityofbirth.Anindividualbelongs to the “treatment group” if she is conceived, born or lived until age 5 in occupiedmunicipalitiesduringtheperiodofNazioccupation.Theconceptiondateisapproximate–266daysor38weekssubtractedfromthebirthdate.13Theoverlapbetweentheageintervalfromconceptiontoage5andtheoccupationintervaldeterminesthelengthof individualexposuretohardshipsinchildhoodunderNazioccupation,anditisoneofthemainmeasuresoftreatmentintensity.

Thedetailsonchildrenmigrationarenotavailableinourdata,sowehavetoassumethatchildren stay in theirmunicipality of birth until age 5. This assumption is not unreasonable. Anumber of WWII history books provide evidence that the German occupation of the USSR wasunanticipated by general population, the retreat of the Red Army was rapid, the evacuation ofcivilianswaspoorlyorganized,andtheescaperouteswerecutoffonceGermanforcesmovedintocitiesandvillages(EhrenburgandGrossman,2003;Erickson,1999;Manley,2009).

Toexaminetheroleofdifferentstagesofchilddevelopmentinfuturehealthoutcomes,wedistinguishthefollowingfourageintervalswithinthetreatmentgroup:in‐utero(fromconceptiontobirth),infants(frombirthto12monthsofage),toddlers(1‐3yearsold),andpreschoolers(3‐5yearsold).Thetreatmentvariableforeachstageoflifeisequaltooneifthereisanyoverlaybetweenthecorrespondingageintervalandtheoccupationintervalforthemunicipalityofbirth.Wenotethatinthemunicipalitiesthatwereoccupiedforseveralyears,thesameindividualmayappearinmultipletreatmentgroupsbygoingthroughseveralstagesoflifeunderoccupation.

Following the theoretical framework presented in Section 3, the controlled observationsconsistoftwosetsofindividuals:(i)individualswhoareconceived,born,orliveduntilage5duringthewarinnon‐occupiedmunicipalities(orthewarcontrolgroup)and(ii)individualswhoareborninallmunicipalitiesafterthewar(orthepost‐warcontrolgroup).ThewarperiodforthefirstgroupisfixedbetweenthebeginningandendofGermanoccupationinoursurveysample,June22,1941(occupationofthecityofBrest)andOctober15,1944(liberationofthecityofRiga).Thus,childreninthewarcontrolgroupareborninnon‐occupiedmunicipalitiesbetweenJune22,1936(forkids

                                                            13Forexample,thecityofLeningradwasundersiegefromSeptember8,1941untilJanuary27,1944.Thus,the “treatmentgroup” includes individualswhoareborn in this citybetweenSeptember8, 1936 (forkidsreachingage5onthefirstdayoftheblockade)andOctober18,1944(forchildrenconceivedonthelastdayofthecityblockade).

reachingage5onthefirstdayofoccupation)andJuly8,1945(forkidsconceivedonthelastdayofoccupationinoursample).Similartothetreatmentgroup,wealsodistinguishthefourstagesofchilddevelopmentwithinthewarcontrolgroup:inutero,infants,toddlers,andpreschoolers,asillustratedinFigure2.

Indefining thepost‐war control group,we follow the related literatureby skippingbirthcohortsbornimmediatelyafterthewar(Akbulut‐Yuksel,2014;ChenandZhou,2007).Thisisdonetoavoidthecontaminationofthecontrolgroupwithindividualsaffectedintheaftermathofthewar.For example, the major famine hit the USSR in 1946‐1947 due to the drought and post‐wardevastations.Ourpost‐warcontrolgroupincludesindividualsbornintheUSSRin1950‐1953,i.e.,conceivedonasearlyasApril1949. The four‐yearperiod is chosen toapproximatelymatch thelengthof thewar. Suchapproach todefining thepost‐warcontrolgroupbychoosing fixedbirthcohortsisprevalentintheliteratureandwillbeadoptedinourstudy.However,withthisapproach,thetimeintervalsbetweenthebeginningofthetreatmentandthefirstyearofthepost‐warcontrolgrouparedifferentacrossstagesoflife:upto14yearsforpreschoolersandupto9yearsforbabiesinutero,ascanbeseeninFigure2.Tomakethetimespellscomparable,weimplementanalternativeapproachbyshiftingtheentiretimeframefor1936‐1945threeyearslaterfor1949‐1958(Figure2).

4.4.TreatmentMarginsattheRegionalLevel:TheDestructionoftheIndustrialBaseandPopulationChange

Inattempttocaptureadditionalheterogeneityinthetreatmenteffectintermsoftheseverityofshocks,weusetwoproxiesfortheextentofdestructionatamoreaggregatelevel.First,recentlydeclassifieddocumentscontainregionaldataonelectricityproductionin1940and1944(CentralStatisticalDirectorateof theSovietUnion,1946). Theabsenceof anationalgrid systemmade itimpossibletoreplacepowerlossessufferedinoneregionoftheSovietUnionbypowerimportsfromotherareas(Hoeffding,1970).Thus,thelogdifferenceinelectricityproductionbetween1941and1944canserveasareasonableproxyforthedestructionoftheindustrialbaseandinfrastructureattheregional level. Weplot this indicator inFigure3 for theEuropeanpartofSovietUnion. Theelectricityproduction fellby18.8percentoverallandby72.6percent inoccupiedregions. Someoccupiedregionslostpracticallyallpowergenerationcapacity– inBelarus,Moldova,partsoftheUkrainianSSR(Nikolayev,Odessa,andZaporozhyeregions),andinpartsoftheRussianSFSRin1945borders(Crimea,Pskov,andSmolenskregions),electricityproductionfellbymorethan95percentfromitspre‐warlevel.Tosupporttheexpandedmilitaryproduction,newpowerplantswerebuiltfrom scratch in non‐occupied regions. For instance, regions ofWestern Siberia increased theirelectricityproductionby107.6percentbetween1941and1944,andregionsoftheUrals–by94.2percent.Thedeclineinelectricitygenerationinteractedwithadummyfortreatedmunicipalitiesisstronglycorrelatedwiththedurationofoccupation(–0.86)

Thesecondproxyisthelogdifferenceinregionalpopulation(adjustedforpre‐warmortalityrates) between the two censuses 1939 and 1959. Specifically, we calculate a counterfactualpopulationbetweentheages20to70in1959,whichistheregion’spopulationaged20‐70thatwouldhavebeenachievedin1959ifthemortalityratesstayedatthepre‐warlevelandthenetmigrationwaszero.14Then,foreachregion,wetakethelogdifferencebetweenactualpopulationaged20‐70in1959andcounterfactualpopulationforthesameagegroup.Thisdifferenceshowstheregionalpopulationchangeduetoexcesswarmortalityandalsoduetointerregionalmigrationduringthewarandpost‐warperiods.15Bynomeans,thismeasurerepresentsthecountofhumanlosses,butit

                                                            14 We take regionalage‐specificpopulationaged0‐50inthe1939Census,adjustitfor20‐yearmortalitywithage‐specificpre‐warmortalityrates,andrecalculateitin1959bordersofregions. 15Partofmigrationflowscouldberelatedtothewar,includingrushedanddisorganizedevacuationsinthechaosoftheSovietretreatandtheflightofpeoplefromdestroyedunlivableareasrightaftertheliberation.The

could representanotherdimensionof the level ofhardship inoccupied regions. The correlationbetweenthe20‐yearpopulationchangeinteractedwithadummyfortreatedmunicipalitiesandthedurationofoccupationis–0.67.

Thesummarystatisticsonpopulationchangeinexcessofnaturalmortalityoverthe20‐yearperiodisreportedinTable2,PanelA.Itindicateslargerpopulationlossesinoccupiedregions(–0.233logpoints)comparedtonon‐occupiedregions(–0.127).Byfurtherdisaggregatingtheadjustedmeasurebyagein1939andgender,weuncovershatteringlossesofmalepopulationinthe15‐50agegroups inbothoccupied(‐0.529)andnon‐occupiedareas (‐0.449). Youngmales inoccupiedregions seem also affected in a big way (–0.218). Relatively smaller changes in population areobservedinnon‐occupiedregionsamongyoungmales(–0.048),adultfemales(–0.045),andyoungfemales(0.002).InPanelBofTable2,wealsoreporttherawdifferencesinpopulationbetweenthetwocensuseswithoutaccountingforpre‐warmortalityrates.Itisnoteworthythatthelargestdeclineinpopulationoccurredinasocalled“deficitzone”outlinedintheHitlerHungerPlaninearly1941;seeSection2.16

4.5.Pre‐TreatmentConditionsBefore proceeding to the estimation, we compare the pre‐war levels of economic

developmentbetweentheoccupiedandnon‐occupiedterritories.Table3showsthatoccupiedandnon‐occupied territoriesprior to theWWIIhad similar levels of gross outputper capita andperworker,energypowercapacity,electricityproduction,employmentparticipation,andwagesofstaffandworkers.Thetwopartsofthecountryexhibitedsimilarinvestmentlevelsinhealthcarein1935‐1936,asthereadercanseefromthepopulation‐adjustedstatisticsonphysicians,hospitalbeds,birthcenters,inpatientclinicsformaternalandinfanthealth,andpediatrictreatmentcenters.Twoyearsbeforethewar,thecompositionofpopulationwasalsonotthatdifferentbetweentheoccupiedandnon‐occupiedregionsintermsoftheshareoffemales,thedegreeofurbanization,theattainmentofsecondaryandhighereducation,femaleschoolingachievements,andethnicdiversity.17

Onlyafewpre‐treatmentindicatorsexhibitstatisticallysignificantdifferencesbetweenthetwo groups, all favoringWestern, soon‐to‐be‐occupied territories. In 1939, population in theseterritorieshadalowerinfantmortalityrate(145vs.192infantsdeathsunderoneyearoldper1,000livebirths),alowercrudebirthrate(32vs.36infantsunderoneyearoldper1,000people),asmallernumberofchildrenperawomanofreproductiveage(439vs.506childrenunderage5per1,000womenaged15‐49),andabetteradultilliteracyrate(19vs.24percentofilliteratepopulationage15andabove).18Ifthepopulationinoccupiedregionsindeedhadbetterhealthconditionspriorto

                                                            escapeduringtheoccupationwaspracticallyimpossible.Anotherpartofmigrationflowsisassociatedwithpost‐warpoliciesoflaborreallocationtotheEastandmassdeportationsofentireethnicgroups.16 Regionsthatlostthemostpopulationaged20‐70inexcessofthepre‐warlevelsofmortalityarealllocatedina“deficitzone”andincludeSmolenskregion(–50%dropinadjustedpopulation),Vitebskregion(–42%),Novgorodregion(–41%),Pskovregion(–41%),Leningradregion(–40%),andothernearbyregions. 17TheSovietUnionwasanethnicallydiversecountry.In1939,14ethnicgroupshadpopulationofalmostonemillionormore,withethnicRussiansbeingthelargestgroup(58.4percentoftotalpopulation).Eventhoughbothoccupiedandnon‐occupiedterritorieshadasimilaraveragevalueofethnicfractionalization,theyhaddifferentethniccomposition.Beforethewarstarted,thefourlargestethnicgroupsintheoccupiedterritorieswereRussians(49percent),Ukrainians(31percent),Belarusians(12percent),andJewish(3.5percent).80percent of Jewish population resided in the occupied territories,which also included the so called Pale ofSettlement or the area of the Imperial Russia beyond which Jewish permanent residency was disallowed(Grosfeldetal.2013).Theresidencybanwasliftedin1917,butthemajorityofJewishpopulationcontinuedlivingclosertotheWesternUSSRborders.18 A closer investigation reveals that statistically significant differences between the occupied and non‐occupied territories in birth and fertility rates are largely driven by pre‐dominantlyMuslim regions withtraditionallyhigherbirthandfertilityrates.

theWWII(astheinfantmortalitystatisticssuggests),thenthiswouldbiasourresultstowardsfindinga smaller effect of the WWII on the long‐term health outcomes, thus providing a lower boundestimate. To control for potential regional differences in pre‐treatment conditions, we includeregional fixedeffectsandthetypeofbirthplace(city, townshiporvillage) inallspecifications. Insomespecifications,wealso focusonnearbymunicipalities inpartially‐occupiedregionsonbothsidesoftheoccupationboundary.

4.6.OutcomesBothRLMSandULMSoffera largearrayofpotentialhealthoutcomesduring the late‐life

years, including:1)anindicatorforhavingbadorverybadhealth;2)havingachronicdiseaseofheart,respiratoryorgans,liver,kidney,stomach,intestines,and/orspine);3)hadaheartattackorhavingachronicheartdisease;4)havingachronicrespiratorydisease;5)havingachronicdiseaseofdigestiveandgenitourinarysystems;6)havingaspinaldisorder;7)sufferedfromhypertension;8)hadasurgicaloperationinthelast12months(RLMSonly);9)hadadepressioninthelast12months(RLMSonly,2003,2004,2011‐2013);10)individualheight;11)underweight,BMI<18.5;and12)obese,BMI>30.Unlessnotedotherwise,alloutcomesareavailablefortheentiresampleperiod,2000‐2013.Individualheightisusedtoconstructtwoproxiesforchildmalnutrition,includingthedeviation of height from themean for each gender aged 20 and above to account for biologicaldifferences in height (residual height) and having a height below the 25th percentile of genderdistribution(lowheight).InlinewithBertoni(2015)whoshowstheimpactoffamineinearlylifeonlife satisfaction inadulthood,weuse two indicatorsas theoutcomemeasuresof subjectivewell‐being: being satisfied or fully satisfied with life (life satisfaction) and being dissatisfied or fullydissatisfiedwithlife(lifedissatisfaction).ThedetaileddefinitionsforallvariablescanbeviewedinappendixTableA1.

Table4reportsthedescriptivestatisticsoftheaboveoutcomesfortreatmentandcontrolsamplesaswellasthemeancomparisont‐test.Comparedtoindividualsinthecontrolgroup,thetreatment group has a significantly higher incidence of reporting bad health (39 percent vs. 21percent),chronicdiseases(67percentvs.56percent),heartdiseases(41percentvs.25percent),anddigestivesystemdisorders(38percentvs.33percent).Individualsinthetreatmentgrouparealsoshorterandmorelikelytobedissatisfiedwiththeirlives(51percentvs.43percent). Otherhealthoutcomesdonotexhibitstatisticallysignificantgroupdifferencesinmeans.

RLMSprovides additionalopportunities to look atnon‐cognitive self‐assessments ona4‐pointscalein2002‐2005and2011surveys,including1)respondentdeterminesownfuture;2)hasgoodcharacteristics;3)feelslikealoser;4)hasapositiveattitude;5)issatisfiedwithself;6)feelsuseless;and7)feelslikeabadperson,amongothers.Oneself‐assessmentquestionona9‐pontscaleforfeelingwellrespectedisavailableinallsurveywaves.Thesummarystatisticsforthesevariables(alsoreportedinTable4)doesnotpaintaconsistentpicture: thetreatmentgroupappearstobedissatisfiedwithlife,butsatisfiedwithself; lesslikelytodetermineownfuturebutmorelikelytohaveapositiveattitude;tobecriticalofhavinggoodcharacteristics,butlesslikelytofeellikeabadperson.Wewillleavetheinterpretationofnon‐cognitiveself‐assessmentstilltheresultsofmodelestimation.

5.ResultsRe‐introducethemodelDiscussXsTocontrol for thedifferences inhealthoutcomesbetweengendersandethnicgroups,we

includeafemaledummyanddummiesforindividualethnicity(Russian‐Ukrainianmixed,Russian,Ukrainian, Jewish, deported nationalities). Mother’s and father’s level of education (college,secondaryeducationormissing)helpuscontrolforparentalcapabilities.Typeofthebirthplace(city,

township, or village) is also likely to react differently to health shocks associated with Germanoccupation.

Womencomprise64.8percentoftreatmentand61.3percentofcontrolsample.Treatmentsample consists of 42.2 percent Russian, 45.2 percent Ukrainian, and 7.3 percent deportednationalities. In the control sample, 78.4 percent are Russian and 9.3 percent are Ukrainian.Individuals in the treatment group aremore likely to be born in a city (64.3 percent) than in atownship(5.8percent),oravillage(29.9percent)whileonly46.3percentofthecontrolgroupisborninacity(8.4percentareborninatownshipand45.3percent–inavillage).Finally,individualsintreatmentgrouparelesslikelytohaveparentswithcollegeorsecondaryeducation(0.8percentofmothersand1percentoffathershavecollegeeducation)perhapsduetothefactthatindividualsbornafterWWII(hencepartofthecontrolgroup)aremorelikelytohaveeducatedparentsascollegeeducationwasbecomingmoreavailableinlate1950sintheUSSR.

Table1reportstheestimatesofourmodelforadultswhowerebetweenin‐uteroand5yearsoldat the timeof theoccupationaswellas for4stagesofchildhoodseparately (in‐utero, infant,toddler, andpreschool). Living in anoccupiedmunicipalityduring theperiodofWorldWar II issignificantly associatedwith reporting one’s subjective health in adulthood as bad (18.3 percentmore likely),havingoneormorechronicdiseases(7.5percentmore likely),havingarespiratorydisease (7.9 percentmore likely), a digestive disease (21.8 percentmore likely), and any healthproblems(7.6percentmorelikely).Additionally,theseresultsmasktheheterogeneityoftheeffectofoccupationatdifferentstagesofindividual’schildhood.Wefindthenegativeeffectsofoccupationtobethemostnumerousandthelargestintheirmagnitudeforthoseadultswhowereexposedtooccupation in‐utero. Compared to adults who were not affected by occupation in‐utero (whosemothers lived in unoccupied territories during the war or who were conceived after the war),affectedadultsare37.5percentmorelikelytoreporthavingbadhealth,47.2percentmorelikelytohaveaheartdisease,28.4percentmorelikelytohaverespiratoryproblems,29.2percentmorelikelytohavedigestivedisorders,16.7percentmorelikelytosufferfromspinalproblems,8.7percentmorelikelytohavehadasurgeryinthelast30days,9.12centimetersshorter,37.4percentmorelikelytobe obese, and 24.5 percent more likely to be unsatisfiedwith their lives. We find the effect ofoccupation to be similar for adultswhowere exposed in infancy,with some effects (on chronicdiseases,hypertension,andhealthproblems)beingevenstronger.Individualsexposedininfancyarealso4.95centimetersshorterand4.9percentmorelikelytobeunderweight.Ourresultssuggestthatthenegativeimpactofoccupationislessdetrimentaliftheexposureoccurslaterinchildhood.Thus,individualswhowereexposedwhentheywereatoddlerorapreschoolerarefoundtobemorelikelyto report being in bad health, having chronic disease, respiratory problems, digestive disorders,spinalproblems,andrecentsurgery,butnomorelikely(comparedtounexposedadults)tosufferfromtheheartdiseaseorhypertension.However,theeffectsontheirhealtharefoundtobemuchsmaller.Thereisalsonostatisticallysignificanteffectofoccupationontheirheightinadulthood.

Toexplorethedifferencesintheeffectsofoccupationonhealthacrossmalesandfemales,weinclude an interaction between the occupation dummy and gender in ourmodel and report theresultsinTable2.Topreservethespace,wechoosetofocusonfiveoutcomes,includingbadhealth,chronic diseases, height, lowheight, and life dissatisfaction. First,we find the negative effects ofoccupation on bad health, chronic diseases, and life dissatisfaction to be significantly larger forfemales.However,duetotheshorterlifeexpectancyofmales(comparedtofemales)inRussiaandUkraine,malesthatweobserveinoursamplearelikelytobethehealthiestintheircohort.19Atthe

                                                            19  In 1995,male and female life expectancies inRussiawere58.1 and71.5 years, respectively (13.4 yearsdifference).In2013,65.1and76.3years,respectively(11.2yearsdifference)(GoskomstatXXXX).In1995,maleandfemalelifeexpectanciesinUkrainewere61.2and72.5years,respectively(11.3yearsdifference).In2013,66.3and76.2years,respectively(9.9yearsdifference).http://www.gks.ru/free_doc/new_site/population/demo/demo26.htm 

sametime,femalesampleislesslikelytobebiasedduetotheselectivemortalityasmanyfemalessurvivetotheirolderages,whichcouldexplainwhywearefindingstrongereffectsofoccupationontheirhealth.Ifonlytheheathiestmalessurvivetotheirolderages,theeffectofoccupationonmalesislikelytoprovidethelowestboundontheeffectontheirhealth.Second,wefindlargereffectofoccupationonmaleheight.Malesexposedtooccupationin‐uteroare9.395centimetersshorterwhilefemalesexposedin‐uteroare7.807centimetersshorter.Theeffectsare5.214and4.695centimetersformaleandfemaleindividualsexposedtooccupationintheirinfancy.Thisfindingisconsistentwithsomeoftheexistingstudies(EvelethandTanner1990;Kuhnetal.1991)andcouldbeexplainedbyadelayinmenarchealageasaresultofthehungerwhichsubsequentlylengthenstheperiodoffemalegrowth(Koupiletal.2009).

                                                            http://ukrstat.gov.ua/operativ/operativ2007/ds/nas_rik/nas_u/nas_rik_u.html

Table1:DistributionofMunicipalities

USSR

CombinedRLMS‐ULMSsample

Treatmentstatusisknown,N 3,857 2,021

Sharebygeographicarea,% Russia 62.6 63.1Ukraine 15.9 26.1Balticrepublics 2.2 1.0Belarus&Moldova 4.0 2.8Caucusesrepublics 4.0 1.7CentralAsia 11.3 5.2

OccupiedduringWWII,N 1,419 916OccupiedduringWWII,% 36.8 45.3

SharebythelengthofGermanoccupation,%

<1month 3.0 3.11‐6months 14.1 11.57‐12months 8.0 7.213‐24months 18.1 21.425‐36months 39.8 44.8Morethan36months 17.0 12.1

In1939borders,N 3,542 1,868Notes:ThelistofUSSRmunicipalitiesistakenfromtheGADMdatabaseofGlobalAdministrativeAreas.WebAppendix1describesadjustmentstothelist.Thelastcolumnshowsthedistributionofmunicipalitiesinwhichrespondentssurveyedin2000‐2013arebornbetween1935and1958.

Table2:ChangeinPopulationbetween1939and1959Censuses

Total Males Females

OccupiedNon‐

occupied OccupiedNon‐

occupied OccupiedNon‐

occupied

A.Adjusted Age0‐50in1939 ‐0.233 ‐0.127 ‐0.386 ‐0.251 ‐0.115 ‐0.025 (0.038) (0.027) (0.042) (0.030) (0.036) (0.026)Age0‐14in1939 ‐0.164 ‐0.022 ‐0.218 ‐0.048 ‐0.113 0.002 (0.061) (0.042) (0.062) (0.043) (0.060) (0.042)Age15‐50in1939 ‐0.281 ‐0.213 ‐0.529 ‐0.449 ‐0.114 ‐0.045 (0.025) (0.019) (0.031) (0.023) (0.023) (0.017)

B.Unadjusted Age0‐50in1939 ‐0.385 ‐0.277 ‐0.563 ‐0.425 ‐0.244 ‐0.152 (0.038) (0.027) (0.042) (0.030) (0.035) (0.025)Age0‐14in1939 ‐0.281 ‐0.139 ‐0.343 ‐0.172 ‐0.224 ‐0.109 (0.060) (0.042) (0.061) (0.043) (0.059) (0.042)Age15‐50in1939 ‐0.456 ‐0.389 ‐0.745 ‐0.664 ‐0.255 ‐0.184 (0.026) (0.019) (0.031) (0.023) (0.023) (0.017)Notes:Thetableshowstheaveragelogdifferenceinregionalpopulationbetween1939and1959weightedbytheshareofUSSRregionin1939population.Regionalbordersaretakenasin1959Census.PanelAreportsthelogdifferenceafteradjustingforthepre‐warage‐specificmortality.

Table3:PretreatmentStatistics–EconomicandSocialIndicatorsbeforetheSecondWorldWar

Occupiedregions

Non‐occupiedregions

t‐test Nofregions

EconomicDevelopment Grossoutputinmillionsofrublesper1000people,1935 0.433 0.456 ‐0.137 64Grossoutputinmillionsofrublesper1000workers,1935 2.579 2.396 0.466 64Powercapacityinthousandkilowattper1000people,1935 0.056 0.047 0.387 64Electricityproductioninmillionkilowatt‐hoursper1000people,

19350.232 0.175 0.528 64

Employmentparticipationrate,%,asofJan.1,1936 0.158 0.168 ‐0.270 64Averagewageofstaffandworkers,rubles,March1936 193.668 204.549 ‐0.853 64

HealthCare 64Numberofhospitalbedsper1000peopleasofJan.1,1936 2.897 3.304 ‐0.747 64Numberofdoctorsper1000peopleasofJan.1,1936 0.539 0.475 0.449 64Numberofbirthcentersandinpatientclinicsformaternaland

infanthealthper100,000peopleasofJan.1,19362.411 2.191 0.764 64

Numberofpediatrictreatmentclinicsper100,000peopleasofJan.1,1936

0.271 0.238 0.775 64

Coefficientofinfantmortality‐numberofinfantdeathstothenumberofborn,1939

145.360 191.613 ‐5.816 81

Census1939 Percentshareoffemalepopulation 52.494 51.976 0.936 111Percentshareofurbanpopulation 36.375 36.128 0.030 111Crudebirthrate‐numberofinfantsper1000people 32.074 35.692 ‐2.723 111Child‐womanratio‐numberofchildrenunder5per1000women

aged15‐49439.339 506.352 ‐2.431 111

Percentshareofilliteratepopulation(age15+) 18.983 23.858 ‐2.368 111Numberofpeoplewithsecondaryeducationper1000people 90.724 77.754 0.850 111Numberoffemaleswithsecondaryeducationper1000people 78.753 68.769 0.613 111Numberofpeoplewithhighereducationper1000people 7.137 7.645 ‐0.149 111Numberoffemaleswithhighereducationper1000people 4.384 4.773 ‐0.172 111Ethnicfractionalizationindex,0=homogeneous 0.321 0.321 0.010 111Percentshareofthelargestethnicgroupinregion 79.183 79.145 0.010 111PercentshareofJewishpopulation 3.571 0.920 3.328 111Percentshareofdeportednationalities 3.999 1.084 1.170 111Notes:Tosavespace,weonlyreportmeans,t‐testresultsforthemeandifference,andnumberofobservations.Thecomplete tablewith standard errors is presented inweb appendix. The data sources employ different levels ofregionalaggregation,whichexplainsthevaryingnumberofregions(N).Theregionalaveragesareweightedbytheshare of the region of birth in the combined RLMS‐ULMS sample. The sample includes the RLMS and ULMSrespondentssurveyedin2003‐2007andbornintheUSSR(within1939borders)in1936‐1958.TheUSSRregionsoutsidetheRussianandUkrainianSovietrepublicsareincludedinthistable,buttheircontributiontotheweightedaverageisinsignificantduetothesampleweights.Partiallyoccupiedregionsareconsideredtobeoccupiedifmorethan20percentoftheirterritoriesweresiegedoroccupied.Usingothercut‐offsforpartiallyoccupiedregionsordroppingtheseregionsentirelydoesnotaffecttheresultsinanysignificantway.Sources: The All‐Union Census of Population 1939 (www.demoscope.ru); The USSR Country of Socialism(Statisticalyearbook),1936(http://istmat.info/node/22521);HealthandHealthCareofUSSRWorkers,1937(http://istmat.info/node/22080).

Table4:DescriptiveStatisticsofHealthandNon‐CognitiveOutcomesbytheTreatmentStatus

Percentagemissing

Treatmentgroup

Controlgroup

Outcomes Badhealth 0.005 0.392 0.214Chroniccondition 0.005 0.758 0.649Chronicheartdisease 0.005 0.038 0.035Chronicrespiratoryillness 0.008 0.079 0.081Chronicdiseaseofdigestivesystem 0.012 0.376 0.330Chronicspinalillness 0.009 0.257 0.249Hypertension 0.011 0.533 0.532Surgeryinthelast12months 0.175 0.038 0.042Healthproblemsinthelast30days 0.177 0.604 0.497Residualheight 0.026 ‐2.516 ‐1.220Lowheight 0.026 0.398 0.302ResidualBMI 0.060 1.604 1.604LowBMI 0.060 0.101 0.122HighBMI 0.060 0.355 0.368Underweight 0.047 0.006 0.009Obese 0.047 0.296 0.317Eversmoked 0.003 0.256 0.384Consumealcoholicbeverages 0.005 0.342 0.497Consumedalcoholicbeverageslastmonth 0.264 0.472 0.670Lifesatisfaction 0.006 0.276 0.341Lifedissatisfaction 0.006 0.511 0.428Loser 0.742 1.952 1.940Proud 0.743 2.159 2.173Useless 0.742 2.147 2.153

Controls

Female 0.000 0.644 0.612Ethnicity,Russian 0.000 0.413 0.784Ethnicity,Ukrainian 0.000 0.458 0.093Ethnicity,Jewish 0.000 0.007 0.003Ethnicity,Deportednationalities 0.000 0.075 0.023Ethnicity,other 0.000 0.029 0.088

Birthplace,city 0.000 0.301 0.453Birthplace,township 0.000 0.058 0.083

Birthplace,village 0.000 0.640 0.463Mother’seducation,college 0.000 0.008 0.036

Mother’seducation,secondary 0.000 0.034 0.178Mother’seducation,missing 0.000 0.503 0.352

Father’seducation,college 0.000 0.010 0.048Father’seducation,secondary 0.000 0.036 0.160

Father’seducation,missing 0.000 0.508 0.392

Figure1:TheLengthofNaziOccupationbyUSSRMunicipality

Figure2:DefinitionsoftheTreatmentandControlGroups

Table5:PreliminaryResults

Appendix

Table A1: Variable Definitions

Generalnotes:1. The source for all  individual‐level variables is RLMS and ULMS. 2. Estimation sample covers 2003‐2007 time periods; variables are available for all years, unless noted 

otherwise  

VariableName DefinitionandSources

EconomicDevelopment Grossoutputinmillionsofrubles

per1000people,1935TheUSSRCountryofSocialism(Statisticalyearbook),1936(http://istmat.info/node/22521)

Grossoutputinmillionsofrublesper1000workers,1935

TheUSSRCountryofSocialism(Statisticalyearbook),1936(http://istmat.info/node/22521)

Powercapacityinthousandkilowattper1000people,1935

TheUSSRCountryofSocialism(Statisticalyearbook),1936(http://istmat.info/node/22521)

Electricityproductioninmillionkilowatt‐hoursper1000people,1935

TheUSSRCountryofSocialism(Statisticalyearbook),1936(http://istmat.info/node/22521)

Employmentparticipationrate,%,asofJan.1,1936

TheUSSRCountryofSocialism(Statisticalyearbook),1936(http://istmat.info/node/22521)

Averagewageofstaffandworkers,rubles,March1936

TheUSSRCountryofSocialism(Statisticalyearbook),1936(http://istmat.info/node/22521)

Health Care  

 

Numberofhospitalbedsper1000peopleasofJan.1,1936

HealthandHealthCareofUSSRWorkers,1937(http://istmat.info/node/22080)

Numberofdoctorsper1000peopleasofJan.1,1936

HealthandHealthCareofUSSRWorkers,1937(http://istmat.info/node/22080)

Numberofbirthcentersandinpatientclinicsformaternalandinfanthealthper100,000peopleasofJan.1,1936

HealthandHealthCareofUSSRWorkers,1937(http://istmat.info/node/22080)

Numberofpediatrictreatmentclinicsper100,000peopleasofJan.1,1936

HealthandHealthCareofUSSRWorkers,1937(http://istmat.info/node/22080)

Coefficientofinfantmortality‐numberofinfantdeathstothenumberofborn,1939

HealthandHealthCareofUSSRWorkers,1937(http://istmat.info/node/22080)

Percentshareoffemale

populationCensus1939

Percentshareofurbanpopulation

Census1939

Crudebirthrate‐numberofinfantsper1000people

Census1939

Child‐womanratio‐numberofchildrenunder5per1000womenaged15‐49

   

Percentshareofilliteratepopulation(age15+)

   

Numberofpeoplewithsecondaryeducationper1000people

   

Numberoffemaleswithsecondaryeducationper1000people

   

Numberofpeoplewithhighereducationper1000people

   

Numberoffemaleswithhighereducationper1000people

   

Ethnicfractionalizationindex,0=homogeneous

   

Percentshareofthelargestethnicgroupinregion

   

PercentshareofJewishpopulation

   

Percentshareofdeportednationalities

   

Outcomes   Badhealth =1ifarespondentevaluateshishealthasbadorverybad

(verygood,good,andaverageareotheroptions)Allyears

Chronicillness =1ifarespondenthasoneofthefollowingchronicillnesses:1) heart disease; 2) illness of the lungs; 3) liver disease; 4)kidneydisease;5)gastrointestinaldisease

RLMS:2000‐2013ULMS:all

Hypertension =1ifadoctorhastoldarespondentthathehashighblood

pressure

Lifedissatisfaction =1ifarespondentisunsatisfiedorfullyunsatisfiedwithhis

life

ControlVariables

Mother’seducation(secondary) =1ifmotherhassecondaryeducationorhigherNote:Availablein2006and2011surveys;extrapolatedtootheryears based on individual panel id for cases of consistentreportingofmother’seducation. 

Mother’seducationismissing =1ifmother’seducationismissing Jewishethnicity =1if Deportednationalities    Otherethnicity Locationofbirth‐Village =1ifrespondentwasborninavillage,derevnia,kishlak,aul

Locationofbirth‐Township = 1 if respondent was born in a city or an urban‐typesettlement

Locationofbirth‐missing =1iflocationofbirthismissing

Federaldistricts(dummies)1)BelarusandMoldova;2)Armenia,Georgia,Azerbaijan;3)Kazakhstan,Tajikistan,Uzbekistan,

AppendixA:LocationCodingandDataLinking(Incomplete)

A.CreatingtheMasterListofMunicipalitieswithIDThedatamanagementinthisprojectiscomplex,asitrequireslinkingnumerousdatasets

withoutthematchingidentifiers. Westartwiththelistofgeographicunitsprovidedbythe2011GADMdatabaseofGlobalAdministrativeAreas(www.gadm.org),aspatialdatabaseofthelocationoftheworld’sadministrativeareas.AdministrativeareasintheGADMdatabaseincludecountries,regionsandlowerlevelsubdivisionssuchascitiesanddistricts.WeselectallcountriesoftheformerSoviet Union and exclude uninhabited geographic areas (such as water bodies) and within‐citydistricts.

The geographic unit in GADM is defined at the three levels of aggregation: 1) countries(ADM0),2)regionsandoftencountrycapitals(ADM1),and3)countiesandcities(ADM2).However,thelevelofaggregationintheGADMdatabaseisnotconsistentacrosscountries.Forexample,theleveltwoinRussiaandUkraineisslightlylargerthanaU.S.county,whileinEstoniatheleveltwocorrespondstotheneighborhoodorsmallnumberofnearbyvillages. Twoborderinggeographicunits,PylvacountyandPechorskiydistrict,havesimilarpopulationsizeof25,000people,butthefirstoneisclassifiedatleveloneinEstonia,whilethesecondoneisatleveltwoinRussia.Tomakethegeographiccodesconsistent,weusetheleveltwoaggregations(ADM2)forlargercountriessuchasBelarus,Kazakhstan,Russia,Ukraine,andUzbekistan,andleveloneaggregation(ADM1)forlesspopulouscountriessuchasBalticcountries,Moldova,Armenia,etc.Werefertotheselectedlevelsofaggregationasmunicipality.

EachmunicipalityinourmasterlistisassignedauniquemunicipalityID.Iftheareaofalarge‐tomedium‐sizedcityisincludedintotheareaofcountyintheGADMclassification,thenanewlinewithuniqueIDforthecityisaddedtothemasterlist.Altogetherwehave3,857municipalities,ofwhichalmost80percentarelocatedinRussiaandUkraine;seeTable1forthedistributionofUSSRmunicipalitiesbygeographicarea.20

B.CodingthebirthplaceWeusetheuniquemunicipalityIDfromthemasterlisttogeocodetheplaceofbirthofthe

RLMSandULMSrespondentswhoarebornintheUSSRandwhoreporttheircityorcountyofbirth.Therewereafewproblemsassociatedwithgeocodingofbirthplaces.First,respondentsoftenreporttheirlocationofbirthusingtheoldnamestheseplaceshadin1930s,1940s,and1950s.WehadtofindcurrentnamesoftheirplacesofbirthastheyappearinGADMdata.

To code location of birth of RLMS and ULMS respondents (location of birth id), we useconstructedmunicipalityid.Wedropindividualswhosecountryofbirthisunknown(XXindividuals)andwhowerebornoutsidetheSovietUnion(XXobservations).

Finally,wemergedGADMmunicipalitynameandidwithindividualdatafromRLMS‐ULMScombinedfile.Therewereafewproblemsassociatedwiththismerging.First,surveyrespondentsreportedoldnamesoftheirplacesofbirth(namestheseplaceshadin1930s,1940s,and1950s)andwehadtofindcurrentnamesoftheirplacesofbirth(astheyappearinGADMdata).Second,someregionsofthesamecountryhadmultiplemunicipalitieswiththesamenameandifarespondentwasborninoneofthem,wecouldnotidentifywhichmunicipalityitwas.Therefore,wehadtochooserandomlyfromthelistofmunicipalitiesinagivenregionofbirth.Finally,ifrespondentdidnotreporthismunicipalityofbirth,weassumedthatthedurationofoccupationofhismunicipalityofbirthwasthesameashisregionofbirthcapital.

                                                            20ThistableexcludesmunicipalitiesforwhichtheoccupationstatusisambiguoussuchasKaliningradregion,KlaipedainLithuania,SouthernSakhalin,andTranscarpathia,whichwereacquiredbytheSovietUnionin1945asaresultofWWII.

We use the same GADM classification of Global Administrative Areas to geocode therespondents’answerstoopen‐endedquestionsonbirthplace;thegeocodingprocessisdescribedinwebAppendix1.ThegeocodedindividualobservationsinRLMS‐HSEandULMSarethenlinkedtoexternaldatabases,includingtheoneonthedatesofoccupationandliberation.Table1providesthedistribution of municipalities in two survey samples. The number of municipalities whererespondentsareborn is fairly large (more than1,400), andmostofmunicipalitiesare located inRussia and Ukraine. In the combined RLMS‐ULMS sample, about half of municipalities wereoccupied,andofthose,47percentwereundertheGermanoccupationbetween2and3years.IntheextendedRLMSsample,theshareofoccupiedmunicipalitiesissmaller(about30percent),butthevarianceinthenumberofdaysunderoccupationissignificant;theoccupationlastedforlessthan6monthsin26percentofthe“treated”municipalities,2to3years–in32percent,and3yearsandmore–in13percent.

C.LinkingexternaldatasetsTomergeGADMmunicipalitieswithourexternaldataonpre‐treatmentcharacteristicsand

demographic characteristics of population from Census 1939, we created four codes, namely,municipality id in 1935, 1939, 1945, and 1959. The main problem associated with mergingcontemporaneousGADMcodeswithcharacteristicsofmunicipalitiesin1935,1939,1945,and1959ischangesinthegeographicboardersofregionsandmunicipalitiesaswellasfrequentchangesofmunicipalitynames.Therefore,wecheckthehistoriesofthesechangesandcreatemunicipalityidsfor1935,1939,1945,and1959bothinGADMandexternaldatasetsformerging.

D.OccupationoftheSovietUnion

InourcombinedRLMS‐ULMSsample,wecoded2,135uniquemunicipalitiesofbirthlocatedinall15republicsoftheSovietUnion.Foreachofthesemunicipalitieswecollectedinformationonwhether theywere occupied duringWWII and if theywere occupied how long their occupationlasted.InformationonthedatesofoccupationofSovietmunicipalitiesbytheGermanforcesandtheirliberationbytheSovietarmywascollectedusingthehistoriesofWWIIandmapsdisplayingchangesinthebattlefieldovertime.OurprimarysourceofinformationontheoccupationofSovietcitiesisDudarenkoetal.(1985)‐oneofthemostcomprehensivesourcesonGermanoccupationoftheSovietUnionpreparedbyateamofresearchersfromtheCentralArchiveoftheMinistryofDefense,theInstituteofMilitaryHistoryof theMinistryofDefenseand theCentralNavalArchivesofRussianFederation.Dudarenkoetal.(1985)providesinformationonthebeginningdateofoccupationandthedateofliberationof727citiesandtownsoftheSovietUnionincludinginformationonpartiallyoccupiedcitiesaswellas thecurrentnamesandthenamesof thecitiesusedduringWWII if thenamesdiffer.FormunicipalitiesnotlistedinDudarenkoetal.(1985),suchasruralmunicipalities,weusedotherrespectedsources.21Iftheexacttimingofoccupationcouldnotbedeterminedbasedon the available sources, we use the beginning of occupation and the liberation of the nearestgeographic municipality (located within X miles) with known information. We calculated thedurationofoccupationasthedifferencebetweenthebeginningofoccupationandtheliberationindays. If a municipality was occupied more than once, we calculated the sum of all occupationperiods.22Incasesofmissinginformation,wemakethefollowingassumptions.IfthemunicipalityofbirthofRLMS‐ULMSrespondentisunknown,weassumethathismunicipalityofbirth’soccupation

                                                            21 MemoryBook(1993,1998),Newsletter(1995),RomanenkoandLovchakova(2002),historyofAkhtubinsk,Alekseevka, Biryuch, Borisovka, Dubrovka, Gubkin, Ivnya, Kletnya, Klimovo, Komarichi, Navlya, Pogar,Rakitnoe,Rognedino,Veydelevka,Volokonovka. 22 Forexample,BelgorodinRussiawasoccupiedfromOctober1941toFebruary1943(473days)andfromMarch1943toAugust1943(140days).Therefore,inourdataBelgorod’stotaloccupationlasted613days. 

lastedaslongastheoccupationofthecapitaloftheregionofbirth(Xindividuals).Ifrespondentwasborninasmallcountry,suchasBelarus,Estonia,Latvia,Lithuania,orMoldova,andhisregionandmunicipalityofbirthareunknown,weassumethattheoccupationofhismunicipalityofbirthisequaltothelengthofoccupationofacapitalcity(Xindividuals).