Forschungsinstitut zur Zukunft der ArbeitInstitute for the Study of La - PDF document

Forschungsinstitut zur Zukunft der ArbeitInstitute for the Study of Labor Does Emigration Bene�t the Stayers?Evidence from EU EnlargementIZA DP No. 6843 Does Emigration Benet the Stayers? Evidence fromEU EnlargementBenjamin ElsnerIZA and IIIS IZA Discussion Paper No. SeptemberBSTRACTDoes Emigration Benefit the Stayers?Evidence from EU EnlargementAround 9% of the Lithuanian workforce emigrated to Western Europe after the enlargement of the European Union in 2004. I exploit this emigration wave to study the effect of emigration on wages in the sending country. Using household data from Lithuania and work permit and census data from the UK and Ireland, I demonstrate that emigration had a significant positive effect on the wages of stayers. A one percentagepoint increase in the emigration rate predicts a0.67% increase in real wages. This effect, however, is only statistically significant for men.JEL Classification:F22, J61, R23Keywordsmigration, labor mobility, EU enlargementCorresponding author:Benjamin ElsnerIZAP.O. Box 724053072 BonnGermanymail: elsner@iza.org I am grateful to Gaia Narciso for all her support and encouragement. I would also like to thank the editor, three anonymous referees, Catia Batista, Karol Borowiecki, John FitzGerald, Ulrich Gunter, Julia Anna Matz, Corina Miller, Mrdjan Mladjan, Alfredo Paloyo, Todd Sorensen, Pedro Vicente and Michael Wycherley, as well as the seminar participants at the 6th ISNE conference in Limerick/IE, the 3rd RGS doctoral conference in Bochum/GER,the 24th Irish Economic Association annual conference in Belfast/UK and the TCD Development Working Group for helpful comments. The Lithuanian and Irish Statistical Offices were very helpful in providing the data. I gratefully acknowledge funding from theIrish Research Council for the Humanities & Social Sciences (IRCHSS) and the Department of Economics at Trinity College Dublin. The paper has been conditionally accepted at the Journal of Population Economics. All errors are mine. 1IntroductionMigrationaectsbothsendingandreceivingcountries.Whileavastliteraturedocumentstheimpactofmigrationonwagesandemploymentinthereceivingcountries,thereisonlysparseevidenceonitsimpactonthesendingcountries.1InthispaperIexploittheemigrationwavefromLithuaniaaftertheenlargementoftheEuropeanUniontostudytheeectofemigrationonwagesinthesendingcountries.WithEUenlargementin2004,LithuanianworkerswereallowedtomigratewithoutrestrictionstotheUnitedKingdom(UK),IrelandandSweden.Between2004and2007,around9%oftheworkforcetookthisopportunityandemigratedtotheUKandIreland.ThelargeemigrationwavecausedbyachangeintheinstitutionalframeworkmakesLithuaniaanidealcasestudyofasendingcountry.Toidentifytheeectofemigrationonwages,Iusetheskill-groupapproachproposedbyBorjas(2003).Thisapproachclusterstheworkforceinanumberofskillgroupsdenedbygender,education,andworkexperienceandcomparesemigrationratesandrealwageswithineachskillgroupbeforeandafterEUenlargement.UsingmicrodatafromLithuania,andworkpermitandcensusdatafromtheUKandIreland,Ishowthatemigrationhasasignicantpositiveeectonthewagesofstayers.Groupswithlargeremigrationrateshadhigherwageincreases.A10%increaseintheemigrationratepredictsanaverageincreaseinrealwagesof6.6%.Thispositiveeect,however,isonlystatisticallysignicantformenbutnotforwomen.Giventhatemigrationwastriggeredbyanexogenouschangeinmigrationlaws,theresultscanbeinterpretedascausal.Thepositiveeectofmigrationonwagesisconsistentwithasimplesupply-and-demandframework.Migrationdecreaseslaborsupply,which-givenadownward-slopinglabordemandcurve-leadstoanincreaseinwages.Theabsenceofastatisticallysigni- 1SeeKerrandKerr(2011)andClemens(2011)forreviewsoftheliteratureontheeconomiceectsofmigrationonreceivingandsendingcountries.2 canteectforwomenissurprising,giventhatwomenaccountedfor40%ofallemigrants.Potentialexplanationsareapositiveself-selectionoffemaleemigrants,orendogenousresponsesinlaborsupply,i.e.womenwhohadnotbeenworkingpreviouslylledthejobofwomenwhoemigrated.TheinstitutionalarrangementsintheEuropeanUnionallowmetoovercomedataconstraintsthatareinherentinthestudyofsendingcountries.Sendingcountriestypicallydonotkeeprecordsofemigrants,whichmakesitdiculttoquantifythenumberofemigrants.WithEUenlargement,workersfromthenewmemberstateswereonlyallowedtomigratetotheUK,Ireland,andSweden.Therefore,itispossibletocalculatethenumberofLithuanianemigrantsfromthecensusandworkpermitdataofthesecountries.Tobecertain,identicationfacesseveralchallenges.Onechallengeisomittedvariablebias.Wagesaredeterminedbynumerousfactorsotherthanmigration,forexampleFDIinows,trade,orunemployment.Ifthesefactorsareomittedfromthemodel,theresultsmaybebiased.Totacklethisproblem,Iaddarichsetofdummyvariablesandinteractiontermstotheregression,whichaccountforchangesinthereturnstoeducationandexperience,anddierencesintheage-earnings-proleacrosseducationgroups.Inaddition,IcontrolforFDI,exports,andunemploymentattheregionallevel.Theresultsarenotsensitivetotheinclusionofthesevariables,however.Anadditionalchallengeisself-selection.Averagewagesmayincrease,simplybecauseworkersfromthelowerendofthewagedistributionhaveleftthecountry.GiventhedataonemigrantsfromtheUKandIreland,itisnotpossibletoassessdirectlywhetheremigrantswithinaskillgroupwerenegativelyselected.Aninspectionofthewagedis-tributioninLithuaniabeforeandafterEUenlargement,however,doesnotindicateanegativeselection.Moreover,asthereceivingcountrieshaveonaveragehigherskillrequirements,selectionshouldbepositive,andtheresultswouldbedownward-biased.Thispaperaddstotheliteratureonthewageeectofemigration,asitshowsthat3 emigrationincreaseswagesintheshortrun.Previousliteraturehaslookedatlong-standingmigrationmovements.Usingthesameapproachasthisstudy,Mishra(2007)andAydemirandBorjas(2007)showthatemigrationfromMexicototheUShasledtoalong-runincreaseinwagesinMexico.Boutonetal.(2011)ndsimilarresultsforMoldova.Thispaper,bycontrast,exploitsasuddenemigrationshocktoshowthatemigrationincreaseswagesevenintheshortrun.2TheEUenlargementwasoneoftherareoccasionsinwhichhigh-incomecountriesopenedtheirbordersforworkersinmiddle-incomecountries.Theresultsofthisstudyarethereforeofinterestformiddle-incomecountriesthatmayfaceasimilarsituationinthefuture.IftheUS,forexample,openeditsbordersforworkersfromSouthAmerica,itwouldbehelpfulforpolicymakersinthesendingcountriestoknowwhatfractionofthepopulationtheycanexpecttoemigrate,andwhatconsequencesthisemigrationwavehasonthelabormarket.2EUEnlargementandMigrationTheEUenlargementinMay2004wasamilestoneintheprocessofEuropeanintegration.15yearsafterthefalloftheIronCurtain,8formersocialistcountriesfromCentralandEasternEuropebecamemembersoftheEuropeanUnion.AtthetimeofEUenlargement,thenewmemberstateswerestillintheprocessofeconomictransition.ComparedtoWesternEurope,economicoutputinthenewmemberstateswasconsiderablylower,whichalsotranslatedintosubstantialwagedierentials.In2004,wagedierentialswerehighestinLatviaandLithuania,whereworkersearnedonaverage30%ofthePPP-adjustedwageintheUK.3Aswagedierentialsareamajordrivingforceofinternationalmigration,themigra- 2Inarecentpaper,Gagnon(2011)usestheemigrationwavefromHondurasafterHurricaneMitch,andndswageeectsthataresimilartothoseinthispaper.3OwncalculationsfromEurostat.4 tionpotentialinthenewmemberstatesbeforeEUenlargementwassubstantial.Studiesthatestimatedthemigrationpotentialfromthenewmemberstatesbeforetheenlarge-mentpredictedthatbetween3%(BauerandZimmermann,1999;BoeriandBrücker,2001)and5%(Sinn,2004)ofthepopulationofthenewmemberstateswouldmigratewithin15years.WithfreedomofmovementbeingacoreprincipleoftheEuropeanUnion,theen-largementwouldhaveallowedworkersfromthenewmemberstatestoworkinanyotherEUcountry.Policymakersintheoldmemberstates,however,fearedthatalargeimmi-grationwavefromEasternEuropecoulddepresswages,increaseunemployment(ZaicevaandZimmermann,2008),andimposeaburdenonthewelfarestate,anddecidedtogivetheoldmemberstatestheoptiontorestrictaccesstotheirlabormarketsuntil2011.OnlytheUK,Ireland,andSwedenopenedtheirlabormarketsin2004.GiventherestrictionsinotherpotentialdestinationcountriesaboveallGermanyandFranceandthegoodeconomicconditionsintheUKandIreland,itwasnosurprisethatthesetwocountrieswerethedestinationforthemajorityofworkersfromEasternEurope.Between2004and2007theUKissuedaround770,000andIrelandaround400,000workpermitstoworkersfromthenewmemberstates,whileonly19,000workerswenttoSweden(Wadensjö,2007).Elsner(2011)showsthatthemagnitudeoftheemigrationwavewasparticularlylargeinLithuania.9%ofallLithuanianworkersreceivedaworkpermitintheUKandIrelandinLatviaandSlovakiathesharewas6%,inPoland5%.4Mostoftheemigrantswereyoung,andhadamediumtohigheducationlevel(ZaicevaandZimmermann,2008).Anumberofstudieshaveevaluatedtheeconomicconsequencesofthismigrationwave.5Moststudiesonthereceivingcountriesdidnotndtheeectsoftheimmigration 4HungaryandtheCzechRepublic,onthecontrary,hadoutowsoflessthan1%.5SeeConstant(2011)forareviewofthemostrecentliteratureandKahanecandZimmermann(2009)foracollectionofcountrystudiesonEUenlargement.Barrelletal.(2010)illustratethemacroeconomicconsequencesofmigrationonthesendingandreceivingcountries.5 waveonwagesandemploymenttobelarge(Barrett,2009;BlanchowerandShadforth,2009).Onthesideofthesendingcountries,theevidenceispurelydescriptive.Kacz-marczyketal.(2009)andHazansandPhilips(2009)illustratethatwagesinPolandandtheBalticStatesincreasedwhileunemploymentdecreasedafterEUenlargement.Thispaperextendstheexistingliterature,asitpresentsarsteconometricevaluationoftheeectofthepost-enlargementmigrationwaveonthesourcecountries.3DataandDescriptiveStatisticsToanalyzetheeectofemigrationonwages,onewouldideallyliketouseamicro-datasetthatcontainsinformationonbothemigrantsandstayers.Suchadataset,however,isusuallynotavailableforthesendingcountries.Inmostcountries,emigrantsarenotobligedtode-register,whichmakesitdicultforthesendingcountriestokeepreliablerecordsontheiremigrants.FollowingMishra(2007),IusedatafromthetwomaindestinationcountriesIrelandandtheUKtocalculatethenumberofLithuanianemigrantsfordierentgroupsofworkersandmatchthemwithstayersfromthesamegroups.Theremainderofthissectiondescribesthedatasetsusedinthisstudyandexplainsthecalculationofthenumberofemigrants.63.1LithuanianHouseholdBudgetSurveyThecoredatasetofthisstudyistheLithuanianHouseholdBudgetSurvey(HBS),whichisavailablefortheyears2002,2003,2005and2006.TheHBSisanannualsurveyof7,000-8,000households;itisrepresentativeattheindividuallevelandcontainsinformationonincomeandexpenditure,aswellasindividualcharacteristicssuchassex,age,educationandplaceofresidence.TheHBSdoesnotcontaininformationonoccupations,industries,orsectors. 6TheentiresectionondataissimilartoElsner(2011),whichusesthesamedatasources.6 Thesamplecontainsemployeesaged18-64workingintheprivatesector.Iexcludepublicsectorworkersbecausewagesinthepublicsectoraretypicallydeterminedbysenioritypayandnotbysupplyanddemand.Inaddition,Idropworkerswithzeroornegativedisposableincome,pensioners,self-employedworkersandworkerswhosemainincomecomesfromtheirownfarm.Thevariableincomefromemployment,deatedbytheHCPI,givesinformationonrealmonthlygrosswages.AswecanseeinTable1a),realwagesincreasedbyaround40%between2002to2006.Alongwiththewagelevel,thestandarddeviationofwagesincreased.Apotentialconcernwithhouseholdbudgetsurveysisover-orunder-reportingofincome,whichcanbiastheresults.Toassessthedegreeofmisreportingbias,Icomparetheself-reportedrealwagesfromtheHBSinTable1a)withthewagesfromtheliveregisterfromtheLithuanianStatisticalOceinTable1b).Itisreassuringthatbothsourcesreportsimilaraveragerealwages,sothatmisreportingshouldnotbiastheresults.[InsertTable1here]3.2IrishCensusToobtainthestocksofLithuanianmigrantsinIrelandandtodeterminethemigrants'skilldistributionIusedatafromtheIrishcensusin2002and2006.TheIrishcensusiscarriedoutevery4-5yearsandcoverstheentirepopulationthatispresentinIrelandinthecensusnight.Forthe2002and2006censuses,theCentralStatisticsOce(CSO)ofIrelandprovidedatabulationofthenumberofLithuaniansbytheireducationalattainment,genderandage.Table2reportsthecharacteristicsofLithuanianmigrantsinIrelandin2002and2006.Mostmigrantshadanuppersecondaryeducationandwereintheir20s.Thenumberof7 menin2006was30%higherthanthenumberofwomen.ThedierenceinthenumberofLithuaniansinIrelandbetween2002and2006showsthatthemajoritymusthavemigratedtoIrelandaroundorafterthetimeofEUenlargement.Notably,theeducationdistributiondidnotchangesignicantlyovertime,eventhoughthestockofmigrantsin2006was10timeshigherthanin2002.ComparingthemigrantsinTable2b)tothestayersinTable1a),wecanseethatthemigrantswereonaverageyoungerandlesseducatedthanstayers.Theshareofworkerswithalowersecondaryeducationislargeramongmigrants,whiletherearerelativelylessmigrantswithanuppersecondaryorathird-leveleducation.Migrantswereonaverage12yearsyoungerthanstayers.[Table2abouthere]3.3IrishandUKWorkPermitDataToobtainthetotalnumberofLithuanianemigrants,IuseworkpermitdatafromtheUKandIreland.Whilethecensusdatareectsalowerboundtothenumberofmigrants,theworkpermitdataisanupperboundofthemigrationowsfromLithuaniatotheUKandIreland.TheworkpermitdatacaptureseverypersonwhocomestotheUKandIrelandandwantstotakeupemployment,beitforapermanentpositionorforatemporaryjob.ThenumberofworkerswholefttheLithuanianworkforcepermanentlyshouldthereforebelowerthanthenumberofworkpermits.Figure1showsthenumberofworkpermitsgrantedtoLithuaniansbetween2002and2007.Intotal,thenumberofLithuanianmigrantstotheUKandIrelandamountedto150,000.Aswecansee,themigrationwavesetinwithEUenlargementin2004andreacheditspeakin2005.AsameasureofthenumberofworkpermitsIusenationalinsurancenumbers(NINo)8 fortheUKandpersonalpublicservicenumbers(PPS)forIreland.7Theworkpermitstatisticsreectactualmigration,becauseworkersonlyreceiveaworkpermitiftheyarephysicallypresentinthedestinationcountry.Toobtainaworkpermit,aworkerhastoreportinpersontotheSocialWelfareOceinIrelandortheDepartmentforWorkandPensionsintheUKandproduceaproofofaddress.Ifaworkermovesback-and-forthbetweenLithuaniaandeithertheUKorIreland,shekeepsherworkpermit,sothatrepeatedmigrationdoesnotcausedoublecounts.8[Figure1abouthere]3.4CalculationofEmigrantNumbersFromthecensusandworkpermitdataInowconstructmeasuresforthenumberofemigrantsbygender,education,experience,andyear.ForthebaselinespecicationIuseacombinationofalldatasources,asthecensusislikelytounder-estimate,andtheworkpermitdataislikelytoover-estimatethenumberofemigrants.Moreover,onlytheIrishcensuscontainsinformationontheskilldistributionofmigrants,whiletheUKandIrishworkpermitdataonlycontainsinformationontheinowsperyear.CensusdatafromtheUKisnotavailableforthetimearoundEUenlargement,asthecensuswascarriedoutin2001and2011.9 7ForfurtherinformationaboutPPSandNINOnumbers,seehttp://www.welfare.ieandhttp://www.direct.gov.uk.In2004theUKintroducedaWorkerRegistrationScheme(WRS)forworkersfromthenewmemberstates.ComparedtothedatafromtheWRS,NINooerstheadvan-tagethatitprovidesinformationonimmigrationbefore2004.TheWRSandNINonumbersafter2004aresimilar.8Doublecountsareonlypossibleifworkersreceivedaworkpermitinbothdestinationcountries.Althoughtheredoesnotseemtobeanyevidenceoflargenumbersofworkersregisteringinbothcountries,Iamawarethatdoublecountingcoulddownward-biastheestimates.9OtherUKdatasets,theLabourForceSurveyandtheEuropeanCommunityHouseholdPanelhavefewobservationsonimmigrantsineachround,andtheygroupimmigrantsfromEasternEuropebyregion,notbycountry.9 Toconstructmeasuresforthenumberofemigrants,ItaketheskilldistributionofLithuanianmigrantsfromtheIrishcensusandmultiplyitwithaweightingfactorwhichaccountsformigrantstotheUK.ThecalculationoftheshareofemigrantsisbasedontheassumptionthattheskilldistributionofLithuanianimmigrantsinIrelandisthesameastheskilldistributionofLithuaniansintheUK.AsshownbyElsner(2011),thisassumptionisjustied,astheeducationandagedistributionofmigrantsfromthe8NewMemberStatesinIrelandandintheUKisalmostidentical.Inaddition,HazansandPhilips(2009)showthateventhoughmigrantsfromLatvia,LithuaniaandEstoniaworkindierentsectorsinIrelandandtheUKinIrelandmoreinconstructionandtrade,intheUKmoreinagricultureandservices-theireducationandageproleisthesameinbothcountries.TomakeuseofallavailableroundsoftheHBSIconstructmeasuresfortheemigrationratesin2003and2005fromthecensusesin2002and2006,assumingthattheskilldistributionofmigrantsarrivingin2003isthesameasin2002,andlikewisethattheskilldistributionofmigrantsin2005isthesameasin2006.Table2suggeststhattheeducationdistributionhasbeenconstantbetween2002and2006,whichimpliesthattheeducationdistributionhasneitherchangedbetween2002and2003,norbetween2005and2006.Theagedistribution,ontheotherhand,haschangedbetween2002and2006;thecohortsarrivingafter2002havebeenonaverageyoungerthanthecohortsbefore2002.Nevertheless,giventhat2002and2003arebothbefore,andthat2005and2006arebothafterEUenlargement,sothatitisplausibletoassumethatworkerscomingin2003hadroughlythesameagedistributionasthosecomingin2002,andworkersarrivingin2005hadthesameagedistributionasthosearrivingin2006.Fort=(2002;2006),thenumberofemigrantsMtghjisMtghj=IEtghj1+NINOt PPSt:(1)10 IEtghjisthenumberofLithuaniansinIrelandinagender(g)-education(h)-experience(j)cellattimet.NINOtandPPStarethenumbersofBritishandIrishworkpermitsissuedtoLithuaniansinyeart.Therstterminparentheses(1inthiscase),accountsforthenumberofmigrantsintheIrishcensus.Thesecondterm,NINOt PPSt,accountsformigrantstotheUK.If,forexample,in2006thenumberofworkpermitsintheUKwas50%higherthanthenumberofworkpermitsinIreland,thisfactoris1.5.Fortheyear2003Itaketheskilldistributionfrom2002andweightitwiththeinowsof2003.Analogously,fortheyear2005Iusetheskilldistributionfrom2006.Thenumberofemigrantsfor2003and2005areM2003ghj=IE2002ghjPPS2003 PPS2002+NINO2003 PPS2002(2)M2005ghj=IE2006ghjPPS2005 PPS2006+NINO2005 PPS2006:(3)Therstterminparentheses,PPS2003 PPS2002andPPS2005 PPS2006accountsforthechangesininowsbetween2002and2003,andbetween2005and2006.10AsinEquation(1),thesecondterminparenthesesrepresentsthenumberofmigrantstotheUK.TocalculatetheemigrationratemperskillgroupandyearIdividethenumberofemigrantsfromEquations(1)to(3)bythepopulationinLithuaniaofthesamegroup,mghjt=Mtghj Pipighjt:(4)ThepopulationofskillgroupghjinyeartisthesumofthesamplingweightspighjtofallworkersiintheLithuanianHBSthatbelongtothisgroup.11 10NINO2003 PPS2002actuallyconsistsoftwofactors:NINO2003 PPS2003,whichaccountsforthesizeofmigrantowstotheUKrelativetoIrelandandPPS2003 PPS2002,accountingforthechangeinmigrationowstoIrelandfrom2002to2003.Bymultiplicationofthosetwoterms,PPS2003cancelsout.11Thesamplingweightpghijtistheinverseprobabilitythatobservationiisincludedinthesample.11 Onemightbeconcernedthatthecalculatedemigrationratemayover-estimatetheactualchangeinlaborsupply,incasemigrantsfromothercountrieshadcometoLithuaniaandtakenupthejobsoftheworkerswholeft.Infact,theLithuanianimmigrationstatisticsshowanincreaseinthenumberofimmigrantsbetween2002and2006.Acloserlook,however,indicatesthatthisincreasewasinlargepartsdrivenbyreturnmigrantsfromtheUK.12Theshareofemigrantscouldalsobeunder-estimated,ifworkersmovedtoothercountriesbesidestheUKandIreland,inparticularcountriesclosetoLithuania,suchasRussia,Germany,andScandinavia.Ofallthesecountries,however,onlytheGermanstatisticsshowaconsiderableincreaseininowsfromLithuania.AsshownbyBrenkeetal.(2009),theannualinowofLithuaniansincreasedfrom2,775beforeEUEnlarge-menttomorethan4,000from2004onwards.Russia,Sweden,Denmark,andFinlanddidsomewhatsurprisinglynotseealargeinowofLithuanians.Between200and300migrantsannuallymovedtoRussia,Sweden,andDenmark,whiletheannualinowinFinlandwaslessthan100.NorwayhasseenincreasedimmigrationfromLithuaniainrecentyears,butthelargeinowonlyoccurredafter2007.13Insum,LithuaniansdidmigratetocountriesotherthantheUKandIreland,buttheymovedtothesecountriesinnumbersthataresmallcomparedtothemigrationwavetotheUKandIreland.4EmpiricalFrameworkThetheoreticalunderpinningsfortheempiricalstrategyarederivedfromasimplesupply-and-demandmodelofalabormarket.Emigrationdecreasesthelaborsupply,whichshiftsthelaborsupplycurveinwards.Givenaconstant,downward-slopinglabordemandcurve,emigrationmakestheremainingworkersamorescarceresource,andleadstoanincrease 12Source:StatisticsLithuania.13Sources:statisticalocesoftherespectivecountries.ForDenmarktheowshavebeencalculatedfromthedierenceinstocks.Tablescanbeproduceduponrequest.12 inwages.4.1TheSkillGroupApproachToidentifytheaverageeectofemigrationonwages,Iusetheskill-groupapproachpro-posedbyBorjas(2003),whichconsidersemigrationratesandwagesatthenationallevelandexploitsthevariationinbothvariableswithinskillgroupsovertime.Ifemigrationindeedincreasedwages,weshouldobservehigherwageincreasesingroupswithahighershareofemigrants.Askillgroupisdenedbytheobservablecharacteristicseducationandworkexperi-ence.Theworkforceconsistsof27skillgroups3educationand9experiencegroups.The3educationgroupsarelowersecondaryeducation(atmost10yearsofschooling),uppersecondaryeducation(11-14yearsofschooling),andthird-leveleducation(atleast15yearsofschooling).14Ahighernumberofeducationgroupswouldbedesirable,asitwouldallowformorevariationinemigrationandwagesacrosseducationgroups.Theavailabledata,however,imposesaconstraintonthenumberofeducationgroups.Thedatasetsfromthesendingandreceivingcountriesdierintheirclassicationofeducationgroups;theHBScontains12,theIrishcensusonly5categories.Choosing3broadeducationgroupsmakesitpossibletoconsistentlymatchemigrantsandstayerswiththesameeducationlevel.Eacheducationgroupisdividedinto9experiencegroups:0-4years,5-9years,10-14years,...,40+yearsofworkexperience.Theworkexperienceiscalculatedastheexposuretothelabormarket,i.e.thetimesincenishingeducation,experience=age-education-6.Thevalueforeducationis10yearsforlowersecondary,12yearsforuppersecondary,and15yearsforthird-leveleducation. 14SeeAppendixC.2foradetaileddescriptionoftheeducationaltracks.13 4.2EmpiricalModelTheempiricalmodelisaregressionofindividualwagesontheshareofemigrantsintheindividual'sskillgroup,estimatedfrompooledcross-sectionaldata.Thebaselineestimatingequationiswight=mght+Xi0ght+year+educ+exper+"ight:(5)wightisthelogrealwageofindividualiwitheducationg,experiencehinyeart=2002;2003;2005;2006.mghtistheemigrationrateforindividuali'sskillgroup.Thecoecientofinterest,,denotesthepercentagechangeinrealwagesassociatedwitha1percentage-pointchangeintheemigrationrate.Thedummyvariablesyear,educ,andexperabsorbchangesinaveragewagesovertime,anddierencesinaveragewagesacrosseducationandexperiencegroups.Xightisavectorofindividualcontrolvariables,whichincludegender,maritalstatus,whetherindividualihaschildrenunder18,andwhethershelivesinacity."ightisanerrorterm.Becausemght,theregressorofinterest,isagroupvariabledenedbyeducation,experienceandtime,Iclusterthestandarderrorsattheyear,education,andexperiencelevel.ThemodelinEquation(5)hastheadvantagethatitusesalownumberofdegreesoffreedom,butitpotentiallycomesatthecostofomittedvariablebias.Theyear,educ,andexperdummiesreducethisbias,buttherecouldbefactorsthathaveanimpactonwagesoverandabovewhatisabsorbedbythedummies.Examplesarechangesinthereturnstoeducationorexperience,ordemandshifterssuchasFDIorexports.Toaccountforthesefactors,Iextendthebaselinemodelwiththeinteractionterms(yeareduc),(yearexper),and(educexper).(yeareduc)and(yearexper)absorbchangesin14 thereturnstoeducationandexperience;(educexper)accountsfordierencesintheage-earningsproleacrosseducationgroups.Theinclusionofinteractiontermshastheadditionaladvantagethatitabsorbscross-wageeects.Iftheunderlyingtheoreticalmodelhasaheterogeneousworkforcewithseveralskillgroups,theimpactofemigrationdependsonthedemographiccharacteris-ticsoftheemigrantscomparedtothestayers(CardandLemieux,2001;Borjas,2003).Emigrationinoneskillgroupaectsthemarginalproductofallothergroups,andhasalargerwageimpactongroupsthatareclosesubstitutes.Aftercontrollingforcross-wageeects,measurestheown-wageeect,i.e.theaverageeectoftheemigrationofworkersfromaspecicskillgrouponthewagesofthatsamegroup.4.3IdentificationIssues4.3.1Sourcesofvariation:skillgroupsvs.occupationsvs.geographyTheskill-groupapproachovercomesidenticationproblemsinherentinthemigrationliterature,byfocusingonmigrationandwagesatthenationallevel.Alargenumberofstudieshaveusedgeographicvariationofmigrationandwagestoidentifytheimpactofimmigrationonthewagesofnatives.15Thesmallandinsignicanteecttypicallyfoundinthesestudiescanbetheresultofunobservedadjustmentinlocallabormarketsoroftheendogenouslocationchoiceofmigrants.Ifmigrantslocateinareaswithmoreexiblelabormarkets,theymaybeabsorbedwithoutdepressingthewagesofnatives,orimmigrationcantriggertheoutowsofnatives(Card,2001).Inaddition,ifmigrantslocateinareasthatexperienceaneconomicboomandhighwages,aspuriouspositivecorrelationbetweentheshareofimmigrantsandwagesmayappear.Theskillgroup 15SeeFriedbergandHunt(1995)andKerrandKerr(2011)forareviewofthisliteratureandLonghietal.(2010)forameta-analysis.15 approach,bycontrast,eliminatestheendogeneityinthelocationchoiceofmigrants.Endogeneitybiascouldonlyariseifmigrantswereabletochoosetheirskillgroup,butthisisnotpossibleasworkersgenerallymaketheireducationdecisionbeforetheyentertheworkforce.Somestudiesovercomethebiasresultingfromendogenouslocationchoicebyexploit-ingvariationinmigrationratesandlabormarketoutcomeswithinoccupationsatthenationallevel(Card,2001;Friedberg,2001).Iftheoccupationispredeterminedbytheimmigrants'educationandtraining,andifimmigrantscannoteasilyswitchtooccupa-tionswithhigherwagegrowth,itispossibletoestimateacausaleectofimmigrationonwagesandemployment.Althoughthewithin-occupationsapproachcanprovideacleanidentication,itre-quiresinformationontheoccupationbeforeemigration,whichisnotavailableforLithua-nianworkersinIrelandandtheUK.Theonlyavailableinformationisthemigrants'currentoccupationafteremigration.InthecontextofEUenlargement,however,itisnotpossibletousethisinformationtoinfertheoccupationbeforeemigration.AsshownbyKahanecetal.(2009,p.20),Drinkwateretal.(2009)andSaleheenandShadforth(2006),immigrantworkersfromthenewmemberstateswereoverrepresentedintypicallow-skilledoccupations,althoughtheireducationlevelwasonaveragehigherthanthelevelofnatives.Theskillgroupapproach,bycontrast,clusterstheworkforceinbroadercategoriesandmakesemigrantsandstayerscomparable.4.3.2EndogeneityissuesThemarginaleectofemigrationonwagesonlyhasacausalinterpretationifemigrationisexogenous.Ideally,onewouldrunanexperiment,inwhichtheemigrationrateisrandomlyassignedacrossskillgroups.AftercontrollingforallotherfactorsinEquation(5),theaveragechangeinwagescouldthenbeexclusivelyattributedtoemigration.As16 realitydoesnotpermitsuchexperiments,identicationhastorelyonquasi-experimentalvariationinemigrationrates.IdenticationinthisstudyisbasedonanexogenouschangeinmigrationlawsaftertheEUaccessionofLithuaniain2004.OnlywhenthecountryjoinedtheEuropeanUnionwereworkersactuallyallowedtoemigrateandtakeadvantageofthehigherwagesinWesternEurope.AsFigure1shows,fewLithuaniansmigratedtoIrelandandtheUKbefore2004,whilethelargemigrationwavebeganin2004.Usingthevariationinemigrationratesandrealwageswithinskillgroupsfrom2002to2006,themodelinEquation(5)comparestheemigrationratesandwagesforeachskillgroupinthetwoyearsbeforeandthetwoyearsafterEUaccession.Theincreaseinemigrationrateswascausedbyanexogenouspolicychange.Therefore,thechangesinrealwages,overandabovethedummiesandinteractionterms,canbeattributedtoemigration.ApotentialconcernabouttheexogeneityofEUenlargementisthatworkerscouldhaveanticipatedtheliftingofmigrationbarriersandaccumulateddestination-specichumancapital.Inthelead-uptoEUenlargement,workersinLithuaniacouldhaveindeedanticipatedthattheywereallowedtoemigrate,asthecountrybeganitsaccessionnegotiationsin1999.Yetthedestinationsformigrationonlybecameclearin2003,whentheoldmemberstatesdecidedontemporaryrestrictionsoftheirlabormarkets.Germany,forexample,onlydecidedinspring2004thatitwouldkeepitslabormarketsclosedforworkersfromthenewmemberstates(DeutscherBundestag,2004).Whileintheorythecausalityrunsfrommigrationtowages,thedirectionofcausalityislessclearempirically.Wagescanbeapushfactorformigration,aslowwagescreateanincentiveforworkerstoemigrate.Inthiscasetherelationbetweenmigrationandwagesshouldbenegative,asskillgroupswithlowwagesshouldhavehighemigrationrates.IntheLithuaniancase,however,reversecausalityshouldnotconfoundtheresults.Theemigrationwavewastriggeredbythecountry'sEUaccession,andworkersfromallskill17 groupsemigrateddespiteconsiderablewageincreases.Moreover,iftheestimateofispositive,reversecausalitycanatmostdownward-biastheresult.Equation(5)onlyidentiesthewageeectiflabordemandisconstant.Shiftsofthelabordemandcurve,unlesscontrolledfor,canbiastheestimates.Onesuchdemandshifteriscapitaladjustment.BasedontheideaofaSolow(1956)-typeframework,emi-grationleadstoadecreaseinthecapitalstock,whichosetthewageeectofemigrationinthelongrun.Thispaper,bycontrast,studiesashort-runeect,sothatcapitalad-justmentsshouldnotaecttheresults.Moreover,itisunlikelythatrmsdecreasetheircapitalstockinaperiodofhigheconomicgrowth,asLithuaniaexperiencedinthe2000s.OnemightbeconcernedthattheLithuanianeconomyunderwentstructuralchangesaroundthetimeofEUenlargementInparticular,EUenlargementdidnotonlychangethemigrationlaws;Lithuaniagainedaccesstoafree-tradeareaandreceivedEUstructuralfunds,whichmaycauseanincreaseinlabordemand.IfEUenlargementchangedthetradeandinvestmentpatterns,wewouldexpectashiftinthelevelofexportsandFDI,orachangeinthetrendofbothvariables.TheaggregatedatadoesnotsuggestthatEUaccessionhasledtosubstantialshiftsinthetradeandinvestmentpatterns.AswecanseeinFigure2,noneofthesevariablesshowsastructuralbreakafterEUenlargement.16[Figure2abouthere]Theoveralltimetrendinthetradeandinvestmentpatternsandofotherfactorsthataectwages,suchasTFPgrowthisaccountedforbytheyeardummiesinEqua-tion(5).Inaddition,ifafactorshiftslabordemandforhigh-skilledworkersmorethan 16Between2004and2006LithuaniareceivedEUstructuralfundsofEUR1.5bn,whichis8%ofthecountry'srealGDPin2004.Thelargestshareofthefunds,whichwerespreadacross3,500projects,wentintoinfrastructureprojects(EuropeanCommission,2007).18 forlow-skilledworkers,orforyoungworkersmorethanforoldworkers,theinteractions(yeareduc)and(yearexper)absorbthesedierentialdemandshifts.TheonlydemandshiftsIcannotcontrolforwithinteractionterms,areskillgroup-specicdemandshifts,becauseaninteraction(yeareducexper)wouldcompletelysaturatethemodel.4.3.3Self-selectionofmigrantsAsitisonlypossibletoobservethewagesofworkerswhodecidenottomigrate,self-selectionarisesasapotentialsourceofbias.Negativeself-selectionofmigrantsleadstoanupward-biasintheestimates.Ifmostemigrantsareselectedfromthelowerendofthewagedistribution,theaveragewageoftheremainingworkersincreases.Yet,thisincreaseisnotcausedbyadecreaseinlaborsupply,butbyachangeinthecompositionoftheworkforce.Analogously,ifmostemigrantsareselectedfromtheupperendofthewagedistribution,theestimateswillbedownward-biased.Theselectionofmigrantscanoccuralongtwodimensions:betweenandwithinskillgroups.WhenwecomparetheeducationdistributionofstayersinTable1andofmigrantsinTable2,wecanseethat,betweenskillgroups,emigrantswerenegativelyselected.Negativeselection,however,doesnotbiastheresults,asthedummiesandinteractiontermsinEquation(5)accountforit.Selectionwithinskillgroupsaselectionpatternthatcannotbeobservedfromthesummarystatisticscanbeasourceofbias.Itisdiculttodeterminethedirectionandsizeofthisbias,asthedatahasnoinformationoncounterfactualwages,i.e.thewagesemigrantswouldearnhadtheystayedinLithuania.ThestandardizedwagedistributioninLithuaniabeforeandafterEUenlargementdoesnotgiveevidenceofselectionbias.Ifmigrantswereonaveragenegativelyselected,wewouldexpecttheprobabilitymasstoshifttotheright.AswecanseeinFigure3,theshapeofthewagedistributionisalmost19 identicalin2002and2006.17[Figure3abouthere]Moreover,giventhedierenceintheeconomicsituationbetweenLithuaniaandIre-landandtheUK,itisunlikelythatmigrantsareonaveragenegativelyselected.First,migrantsare,bydenition,moremobilethanstayers.Ifmobilityispositivelycorrelatedwithability,migrantsshouldbeonaveragemoreskilledthanstayers,andearnhigherwages.Second,becauseoftheforeignlanguagerequirements,andbecauseofminimumwages,theskillrequirementsareonaveragehigherintheUKandinIrelandthaninLithuania.Mostjobs,inparticularintheservicesector,requireuencyinEnglishandagoodknowledgeofBritishorIrishculture.Inaddition,theminimumwagesintheUKandIrelandareconsiderablyhigherthaninLithuania,whichcreatesanadditionalhurdleforlow-skilledmigrants.Onlythemoreproductivemigrantsgetajobthatpaysthematleasttheminimumwage.18AstheUKHomeOce(2009)shows,morethan80%ofimmigrantsfromtheaccessioncountrieswereociallyemployed,sothattheminimumwageisbindingforthemajorityofimmigrants.Third,sincetherewaslittlemigrationfromLithuaniatoIrelandandtheUKpriortoEUaccession,migrantscouldnotrelyonlargemigrantnetworksthatwouldsupporttheminndingajobandfacilitateassimilation.Assuggestedbytheliteratureonmigrantnetworks(Carringtonetal.,1996;McKenzieandRapoport,2010),smallnetworksareusuallyassociatedwithapositiveselectionofmigrants. 17Figure4inAppendixC.4plotsseparatewagedistributionsformenandwomen.Formen,therehavebeensomechangestotheleftofthemean,butnosubstantialshiftsintheprobabilitymass.Bycontrast,forwomentheprobabilitymassmovedtotheleftofthemean,indicatingapositiveselection.18In2004,minimumwageswereEUR7inIrelandandGBP4.85intheUK.20 Closelyrelatedtotheissueofself-selectionisthequestionwhethersomeoftheworkerswereunemployedbeforetheyemigrated.Ifthiswasthecase,emigrationcouldhavedecreasedunemploymentandinthemostextremecasehavenoeectonwages.Infact,Figure2showsthatunemploymenthadbeenfallingbetween2002and2006.WhileIcannotexcludethatemigrationplayedaroleinreducingunemployment,theunemploymentratedoesnotshowastructuralbreakafterEUaccession.Eventheemigrationof9%oftheworkforcedidnotcauseasuddendropintheunemploymentrate.Ifbeingunemployedisassociatedwithlowerability,andifmigrantsareonaveragepositivelyselectedwithinskillgroups,thenmostofthemigrantsshouldbeemployedatthetimeofemigration.WhiletheLithuanianunemploymentdataisnotdetailedenoughtocalculateunemploymentratesperskillgroup,itispossibletocontrolforunemploymentattheregionallevel,whichIdoinarobustnesscheckinAppendixC.1.Moreover,fortheunlikelycasethatmanyemigrantswereunemployedrightbeforeemigration,theestimatesofthewageeectwouldbedownward-biased,asthecalculatedemigrationratewouldbehigherthantheactualone.5EmpiricalAnalysis5.1EstimationresultsTable3presentstheresultsoftheestimatedimpactofemigrationontherealwagesofstayers.Thewageeectformenandwomen,reportedinColumn(1),indicatesthatemigrationpredictsasignicantincreaseinwages.Aonepercentage-pointincreaseintheemigrationrateincreasesrealwagesonaverageby0.67%.[InsertTable3here]21 Whilethiseectmaybelargeandstatisticallysignicantonaverage,thewageeectscandierbetweenmenandwomen.Toanalyzethedierenceinthewageeectbetweenmenandwomen,Iinteracttheemigrationratewithadummyformen.Ascolumn(2)shows,thecoecientoftheinteractiontermindicatesalargeandstatisticallysignicantdierenceinthewageeectofemigrationformenandwomen.Foraonepercentage-pointincreaseintheemigrationrate,thewagesofmenincreasedonaverageby1.1%,whilethemarginaleectforwomenissmallerandstatisticallyinsignicant.Fromcolumns(5)and(6)wecanseethattheseresultsalsoholdifthesampleissplitbetweenmenandwomen.InColumn(3)IcontrolforFDIinows,exports,andunemploymentattheregionallevel.Eachofthesefactorscanconfoundtheanalysis,iftheyaectwagesoverandabovewhatitabsorbedbythedummyvariables.Thethreevariablesaremeasuredattheregionallevel,sothatthewageofapersoncanbematchedwiththeFDI,unemployment,andexportsintheregionthepersonislivingin.Itisreassuringthatthemostobviouspotentialconfoundingfactors,FDI,exports,andunemployment,donotchangetheresultsofthemoreparsimoniousspecicationinColumn(1).Next,Iincludeaninteractionofregionandyeardummiesintothebasicmodeltoensurethatnootherfactorsaectwagesattheregionallevel.Theregion*yearinter-actionsabsorballeconomicfactorsthataectaregionovertimebutareunrelatedtoemigration.Theresultsofthisspecication,displayedinColumn(4),arenotdierentfromthepreviousresult.[InsertTable4here]Anobviousproblemwithcontrolsattheregionallevelisthatthedemandshiftersarethesameforallskilllevels.If,forexample,thedemandshiftislargerforhigh-skilled22 thanforlow-skilledworkers,thischangeinreturnstoeducationcannotbecapturedwiththecontrolsofthebasicmodel.Toaccountforchangesinreturnstoeducation,Ire-estimatethebasicmodelwithaninteractionofyearandeducationdummies.AswecanseeinColumn(1)ofTable4,theestimatedwageeectisthesamewhenweaccountforchangesinreturnstoeducation.Inasimilarfashion,thereturnstoexperiencecanchangeovertime.Technologicalprogress,forexample,canbenetyoungworkersmorethanoldworkers.Toaccountforchangesinreturnstoexperience,Iincludeaninteractionofyearandexperiencedummies.Column(2)ofTable4indicatesthatchangesinreturnstoeducationexplainpartofthewageincreases.Thepointestimatesare0.3lowercomparedtothebenchmarkcase.Partoftheinitialresultscanalsobedrivenbydierencesintheage-earningsproleacrosseducationgroups.ThebasicmodelinEquation(5)estimatesaseparateinterceptforeveryeducationlevel,everyexperiencelevel,andeveryyear.Thedierenceinwagesforoldandyoungworkers,however,maybelargerforhigh-skilledworkersthanforlow-skilledworkers,orviceversa.Aninteractionofeducationandexperiencedummiesabsorbsthedierenceintheage-earningsprolebetweeneducationgroups.TheresultsinColumn(3)ofTable4suggestthattheage-earningsprolesdierinfactbyeducationlevel.Takingthemintoaccountincreasesthepointestimatesformenandwomenby0.3.Theinclusionofinteractiontermschangestheestimates,whichsuggeststhatreturnstoeducation,returnstoexperience,anddierenceinage-earningsprolesexplainpartofthewagechanges.Toseehowtheinteractionsjointlyaecttheresults,IincludetwointeractionsatatimeinColumns(4)-(6)inTable4.Theresultsaremixed,withresultssimilartothebaselinecaseinColumns(5)and(6),andnostatisticalsignicanceandlowpointestimatesifyear*educationandyear*experienceareincluded.Column(7)displaystheestimateswithallthreeinteractionsincluded.InthisspecicationthesameasinBorjas(2003)andMishra(2007)theonlypossiblevariationiswithinskillgroupsover23 time.Despitethelargenumberofregressors,Indalargeandstatisticallysignicantpositiveeectofemigrationonthewagesofmen,andastatisticallyinsignicanteectonthewagesofwomen.5.2DiscussionoftheresultsTheresultsshowthatemigrationhasapositiveimpactonwagesonaverage,whichisconsistentwithasupply-and-demandframework.Emigrationleadstolaborshortages,whichgivenadownward-slopinglabordemandcurvecausesanincreaseinrealwages.EUenlargementincreasedtheworkers'bargainingpowervis-a-vistheiremployers,whichenabledthemtonegotiatehigherwages.Theestimatedeectisstatisticallyandeconomicallysignicant.Themarginaleectof0.67meansthataonepercentage-pointincreaseintheemigrationrateincreasesrealwagesonaverageby0.67%,whichisinlinewithElsner(2011),whoestimatesthedemandelasticitywiththesamedatainastructuralmodel.If5%oftheLithuanianworkforceemigratedpermanently,themodelpredictsthatwagesincreaseby3.3%over5years.GivenaveragewagesinLithuaniaincreasedby40%overthesameperiod(seeTable1),emigrationcanexplain8%oftheoverallwageincreases.Ifwefocusonthemarginaleectformen,emigrationevenexplains16%ofthewageincreases.Thedierenceinthewageeectsforwomenandmenisstriking.Thereareseveralpotentialexplanationsfortheabsenceofasignicanteectforwomen.Oneexplanationisfemalelaborforceparticipation.ThedatafromtheUKandIrelandmayover-estimatethenumberofwomenthathavelefttheLithuanianworkforce,becauseeithertheemigrantwomenwerenotpartoftheworkforce,orbecausetheywerereplacedbywomenwhowerepreviouslynotpartoftheworkforce.InbothcasestheactualnumberofemigrantswouldbesmallerthanthenumberintheBritishandIrishdataandwecouldexpectalowerimpactofmigrationonwages.Yetgiventhatthroughoutthe24 2000sfemalelaborforceparticipationhasbeenashighastheparticipationofmen,thisexplanationseemslessplausible.Thedierentialeectcouldperhapsbeduetothesectoralcompositionofemigration,ifwomenareover-representedinsectorswithmorerigidwages.Whilethedataonemigrantsdoesnotyieldanyinformationonthemigrants'occupationpriortomigration,wecanatleastlookattheshareofwomenworkingindierentsectorsinLithuania.Inthe2000swomenwereover-representedinservices,whilemenwereover-representedinagriculture,construction,andmanufacturing.Theearningsincreasesforservices,however,weresimilartotheincreasesconstruction,andwerelargerthaninmanufacturingandagriculture.Insum,thesectoralcompositioncannotexplainthedierenceinthewageeects.19AnotherexplanationcouldbethatEUenlargementgaveahigherbargainingpowertomenthantowomen.Ifmenarethemainearnersofthefamily,itiseasierformenthanforwomentousetheoptiontoemigrateasacrediblethreatwhennegotiatingtheirsalaries.Yetanotherexplanationisself-selectionofemigrantwomen.Ifwomenwereonav-erageselectedfromtheupperendofthewagedistributionthentheaveragewageoftheremainingwomendecreases.Theshiftofthewagedistribution,showninFigure4intheonlineappendix,supportsthisexplanation.6ConclusionInthispaperIstudytheeectofemigrationonthewagesofstayers.Accordingtoasimplesupply-and-demandframework,emigrationreduceslaborsupplyandcausesanincreaseinrealwages.UsingtheemigrationwavefromLithuaniaafterEUenlargement,Itestthishypothesis. 19Sources:StatisticsLithuania.Tableavailableonrequest.25 WithEUenlargement,workersfromLithuaniawereallowedtoemigratetotheUKandIreland;around9%oftheLithuanianworkforceemigratedafterthecountryjoinedtheEuropeanUnion.Iexploitthisexogenouschangeinmigrationlawsandtheresultingemigrationwavetoidentifytheeectofemigrationonwages,usingvariationwithinde-mographicgroupsovertime.Theestimatedimpactofemigrationonwagesissignicant.Aone-percentagepointincreaseintheemigrationrateincreasesrealwagesonaverageby0.66%.Thiseect,however,isonlysignicantformen,notforwomen.Themagnitudeoftheeectislargerthaninpreviousstudies(Mishra,2007;AydemirandBorjas,2007),whichlookedatthelong-runeect.TheresultsofthisstudyindicatethatemigrationcanhavealargereectintheshortrunthaninthelongrunTheresultscaninformpolicymakersabouttheeectsofalargeemigrationwaveonthelabormarketsinthesendingcountries.Thereareanumberofmiddle-incomecountriesthatcouldfaceasimilaremigrationwave,oncetheirworkersareallowedtoemigrate.ExamplesareEUcandidatecountrieslikeCroatia,Serbia,Montenegro,orTurkey,whichexhibitlargewagedierentialsvis-á-visWesternEurope.Thisstudyopensseveralavenuesforfutureresearch.Asmoremigrationdatabecomesavailable,itisimportanttocheckthevalidityoftheresultsforalargernumberofcountries.Whiletheimmigrationliteraturehasfoundverysmalleectsofmigrationonwagesinthereceivingcountries,thelimitedevidenceonthesendingcountriesshowsthattheeectscanbesignicant.Tobecertainthatthiseectisnotonlylimitedtoasmallnumberofcountries,werequireevidencefrommorecountries.EUenlargementoccurredduringaneconomicboominWesternEuropesothatwork-ersfromEasternEuropecouldeasilyndjobsafteremigration.Withthenancialcrisis,startingin2008,theprospectsformigrantsinmostofWesternEuropehavebecomelesspositive,andmanymigrantsarereturningtotheirhomecountries.ThesetwostatesoftheEuropeaneconomyboombefore2008,followedbyacrisiscouldbeusedto26 identifytowhatdegreemigrationandreturnmigrationisdrivenbywagedierentialsanddierencesintheemploymentrates.Moreover,inlookingatworkersthatemigratedimmediatelyafterEUenlargementitwouldbeinterestingtoinvestigatewhichworkersstayedandwhichworkersreturnedtotheirhomecountries,andwhatdeterminedthetimingofthedecisiontoreturn.27 ReferencesAydemirA,BorjasG(2007)Cross-CountryVariationintheImpactofInternationalMigration:Canada,MexicoandtheUnitedStates.JournaloftheEuropeanEconomicAssociation5(4):663708BarrellR,FitzGeraldJ,RileyR(2010)EUEnlargementandMigration:AssessingtheMacroeconomicImpacts.JournalofCommonMarketStudies48(2):373395BarrettA(2009)EUEnlargementandIreland'sLabourMarket.In:KahanecM,Zimmer-mannKF(eds.)EULaborMarketsafterPost-EnlargementMigration,Springer-Verlag,Berlin.145161BauerTK,ZimmermannKF(1999)AssessmentofPossibleMigrationPressureanditsLabourMarketImpactFollowingEUEnlargementtoCentralandEasternEurope.IZAResearchReport3BlanchowerDG,ShadforthC(2009)Fear,UnemploymentandMigration.TheEconomicJournal119:136182BoeriT,BrückerH(2001)EasternEnlargementandEU-Labour-Markets.WorldEco-nomics2(1):4968BorjasGJ(2003)TheLaborDemandCurveISDownwardSloping:Re-examiningtheImpactofImmigrationontheLaborMarket.TheQuarterlyJournalofEconomics118(4):13351374BoutonL,PaulS,TiongsonER(2011)TheImpactofEmigrationonSourceCountryWages:EvidencefromtheRepublicofMoldova.WorldbankPolicyResearchWorkingPaper576428 BrenkeK,YukselM,ZimmermannKF(2009)EUEnlargementunderContinuedMobilityRestrictions:ConsequencesfortheGermanLaborMarket.In:KahanecM,Zimmer-mannKF(eds.)EULaborMarketsAfterPost-EnlargementMigration,Springer-VerlagCardD(2001)ImmigrantInows,NativeOutowsandtheLaborMarketImpactofHigherImmigration.JournalofLaborEconomics19(1):2264CardD,LemieuxT(2001)CanFallingSupplyExplaintheRisingReturntoCollegeForYoungerMen?ACohort-BasedAnalysis.TheQuarterlyJournalofEconomics116(2):70546CarringtonWJ,DetragiacheE,VishwanathT(1996)MigrationwithEndogenousMovingCosts.TheAmericanEconomicReview86(4):909930ClemensMA(2011)EconomicsandEmigration:Trillion-DollarBillsontheSidewalk?JournalofEconomicPerspectives25(3):83106ConstantAF(2011)Sizingitup:LaborMigrationLessonsoftheEUEnlargementto27.IZADiscussionPaper6119DeutscherBundestag(2004)EntwurfeinesGesetzesüberdenArbeitsmarktzugangimRahmenderEU-Erweiterung.Drucksache15/2672DrinkwaterS,EadeJ,GarapichM(2009)PolesApart?EUEnlargementandtheLabourMarketOutcomesofImmigrantsintheUK.InternationalMigration47(1):161190ElsnerB(2011)EmigrationandWages:TheEUEnlargementExperiment.IZADiscus-sionPaper6111EuropeanCommission(2007)TheEuropeanStructuralFunds(2004-2006),LietuvaFriedbergRM(2001)TheImpactofMassMigrationontheIsraeliLaborMarket.TheQuarterlyJournalofEconomics116(4):1373140829 FriedbergRM,HuntJ(1995)TheImpactofImmigrationonHostCountryWages,Em-ploymentandGrowth.JournalofEconomicPerspectives9(2):2344GagnonJ(2011)"StaywithUs?"TheImpactofEmigrationonWagesinHonduras.OECDWorkingPaper300HazansM,PhilipsK(2009)ThePost-EnlargementMigrationExperienceintheBalticLaborMarkets.In:KahanecM,ZimmermannKF(eds.)EULaborMarketsafterPost-EnlargementMigration,Springer-Verlag,Berlin.255304HomeOce(2009)AccessionMonitoringReport,May2004-December2008,A8Coun-tries.HomeOceReportKaczmarczykP,MioduszewskaM,ZyliczA(2009)ImpactofthePost-AccessionMigra-tiononthePolishLaborMarket.In:KahanecM,ZimmermannKF(eds.)EULaborMarketsafterPost-EnlargementMigration,Springer-Verlag,Berlin.219253KahanecM,ZaicevaA,ZimmermannKF(2009)LessonsfromMigrationafterEUEn-largement.In:KahanecM,ZimmermannKF(eds.)EULaborMarketsafterPost-EnlargementMigration,Springer-Verlag,Berlin.345KahanecM,ZimmermannKF(eds.)(2009)EULaborMarketsAfterPost-EnlargementMigration.Springer-Verlag,BerlinKerrSP,KerrWR(2011)EconomicImpactsofImmigration:ASurvey.FinnishEco-nomicsPapers24(1):132LonghiS,NijkampP,PootJ(2010)JointimpactsofImmigrationonWagesandEm-ployment:ReviewandMeta-Analysis.JournalofGeographicalSystems12:355387McKenzieD,RapoportH(2010)Self-SelectionPatternsinMexico-U.S.Migration:TheRoleofMigrationNetworks.ReviewofEconomicsandStatistics92(4):81182130 MishraP(2007)EmigrationandWagesinSourceCountries:EvidencefromMexico.JournalofDevelopmentEconomics82:180199SaleheenJ,ShadforthC(2006)TheEconomicCharacteristicsofImmigrantsandTheirImpactonSupply.BankofEnglandQuarterlyBullettinQ4:373385SinnHW(2004)EUEnlargement,Migration,andtheNewConstitution.CESifoWorkingPaper1367SolowRM(1956)AContributiontotheTheoryofEconomicGrowth.TheQuarterlyJournalofEconomics70(1):6594WadensjöE(2007)MigrationtoSwedenfromtheNewEUMemberStates.IZADiscus-sionPaper3190ZaicevaA,ZimmermannKF(2008)Scale,DiversityandDeterminantsofLabourMigra-tioninEurope.OxfordReviewofEconomicPolicy24(3):42845231 ATables32 Table1SummaryStatisticsLithuania 2002200320052006 a)LithuanianHBSObservationsmen2,3222,4112,4262,314women1,6281,7251,6161,560Educationlowersecondary8.81%10.42%10.76%9.91%uppersecondary69.01%69.17%67.62%67.48%third-level22.18%20.41%21.62%22.61%MonthlyEarnings(LTL)men1,1851,2521,4401,688(856)(913)(981)(1,134)women9409881,1891,303(684)(686)(890)(985)b)LithuanianStatisticalOceMonthlyEarnings(LTL)men1,1731,2271,4201,676women9981,0291,1671,356 Note:a):Summarystatisticsforallemployeesbetween18and64years.Educationgroups:lowersecondaryeducation(10yearsorlessofschooling),uppersecondaryeducation(morethan10yearsofschooling,butnonishedthird-leveleducation),third-leveldegree(atleast15yearsofschoolingandB.Scequivalent).Percentagesofeducationaldistributionrelativetoallmenandwomeninagivenyear.MonthlyearningsaredeatedbytheHCPI.Standarderrorsofmonthlyearningsinparentheses.b)monthlyearningsareaveragegrossmonthlyrealearningsinLTL.33 Table2SummaryStatistics,Irishcensus 20022006 Observationsmen97812,085women9049,293Educationlowersecondary16.6%20.1%uppersecondary63.4%62.3%third-level20.0%17.56%Age3.5%2.8%20-2953.3%60.7%30-3926.0%24.6%40-4923.3%9.4%50+3.9%2.5% Note:ThistabledisplaysthesummarystatisticsoftheIrishcensus.Educationgroups:lowersecondaryeducation(10yearsorlessofschooling),uppersecondaryeducation(morethan10yearsofschooling,butnonishedthird-leveleducation),third-leveldegree(atleast15yearsofschoolingandB.Scequivalent).Percentagesofeducationandagedistributionrelativetoallmenandwomeninagivenyear.34 Table3ThewageeffectofemigrationDependentvariable:logrealwage (1)(2)(3)(4)(5)(6)Sample:allallallallmenwomen Emigrationrate0.665**0.3910.4260.4011.245***0.283[0.2937][0.3132][0.3154][0.3236][0.2950][0.3910]Emigration*male0.799***0.793***0.777***[0.2936][0.2912][0.2852]YeardummiesyesyesyesyesyesyesEducationdummiesyesyesyesyesyesyesExperiencedummiesyesyesyesyesyesyesFDI,unemp.,exportsnonoyesnononoYear*regionnononoyesnonoObservations997099709970997067713199AdjustedR20.34630.34680.35680.36380.33710.3222 Robuststandarderrorsinbrackets***p0.01,**p0.05,*p0.1Note:ThistableshowstheOLSresultsfortheeconometricmodelinEquation(5),aregressionoflogrealwagesontheemigrationrate,interactionsoftheemigrationratewithadummyformen(emig*male),avectorofpersonalcharacteristics.Standarderrorsareclusteredatthetime-education-experiencelevel.Allobservationsareweightedwithsurveyweights.FDIstocks(inlogs),unemploymentrateandexports(inlogs)aremeasuredattheregionallevel.Year*regionisaninteractionofyearandregiondummies.35 Table4EstimationofthewageeffectwithadditionalcontrolsDependentvariable:logrealwage A.Men(1)(2)(3)(4)(5)(6)(7) Emigrationrate1.117***0.825**1.497***0.2921.372***1.464***0.833***[0.3218][0.3998][0.3145][0.3953][0.3127][0.3265][0.2649]Observations6771677167716771677167716771AdjustedR20.33740.33750.33820.33820.33850.33870.3392 B.Women Emigrationrate0.310-0.0390.622*-0.0120.6420.817*1.035[0.4166][0.4986][0.3217][0.4632][0.3859][0.4569][0.6657]Observations3199319931993199319931993199AdjustedR20.32250.32910.33020.33020.33050.33640.3375 Controlsyear*educyesnonoyesyesnoyesyear*expernoyesnoyesnoyesyeseduc*expernonoyesnoyesyesyes Robuststandarderrorsinbrackets***p0.01,**p0.05,*p0.1Note:Thistableshowstheestimationresultsoftheeectofemigrationonwageswithanumberofadditionalcontrols.Year*educaccountsforchangesinreturnstoeducation.Year*experaccountsforchangesinreturnstoexperience.Educ*experaccountsforfundamentaldierencesinwagesbetweenexperiencegroupswithinaneducationgroup.Furthermore,allregressionsincludeyeardummies,educationdummies,experiencedummies,andavectorofpersonalcharacteristics.Standarderrorsareclusteredatthetime-education-experiencelevel.Allobservationsareweightedwithsurveyweights. 36 Figure1LithuanianImmigrantstotheUKandIreland,2002-2007Notes:NumberofLithuanianimmigrantstotheUKandIrelandbetween2002and2007,asmeasuredbythenumberofworkpermits(PPSNumberinIreland,NationalInsuranceNumbersintheUK).Sources:IrishDepartmentofSocialandFamilyAairs,UKDepartmentforWorkandPensions.BFiguresforthebodyofthePaper37 Figure2FDI,exports,GDPpercapita,andtheunemploymentrateinLithuania,2002-2006Notes:ThegraphshowsthetimeseriesforexportstotheEU,FDIinowsfromtheEU,realGDPpercapitaandtheunemploymentrate.Allvariablesarenormalizedto100(rstquarterin2002).NoneofthevariablesshowsastructuralbreakaroundEUenlargement.Source:StatisticsLithuania38 Figure3StandardizedwagedistributioninLithuania,2002and2006Notes:ThegraphshowsaKerneldensityplotofthelogrealwagesin2002and2006.ThisplotallowsforacomparisonofthewagedistributionbeforeandafterEUaccession.Itshowsthattheshapeofthedistributionhasonlychangedslightly,despitetheemigrationof9%oftheworkforce.Tomakethedistributioncomparableacrossyears,wagesarestandardizedtotheirz-scores,i.e.thewageofanindividualminusthemeanwage,dividedbythestandarddeviationofwages,zi=(wi�w)=w.Themeanofthedistributioniszero.Source:LithuanianHouseholdBudgetSurvey39 COnlineAppendixC.1RobustnesschecksThecalculationoftheemigrationratesisbasedonanumberofassumptions.Table5demonstrateshowtheresultschangewhentheassumptionsaredropped.PanelA)showstheresultsforthebaselinemodelinEquation(5);inpanelB)Iaddarichsetofinteractionterms.Columns1)and2)showthesensitivityoftheresultswithrespecttochangesinthecellsize.Thecoecientsarelowerfor2-yearcellsandlargerfor10-yearcells.Paneli)displaystheestimatesformenandwomentogether.Thecoecientisstatisticallysignicantfor2-yearcellsbutnotfor5-yearcells.Thestatisticalsignicanceoftheeectformenisnotaectedbythecellsize.InColumn3)Idropthedatafor2003and2005,asIdonothavepreciseemigrationdatafortheseyears.WerstlookatpanelA):Thecoecientformenandwomenjointlyislargerthaninthebaselineandstatisticallysignicantatthe5%level.Theinteractionoftheemigrationrateandthemaledummyinii)issimilartothebaseline,andsignicantatthe10%level.InthesaturatedmodelinpanelB)noneofthecoecientsisstatisticallysignicant.Column4)displaystheresultsforIrishdataonly.Thisexerciseclearlyunderestimatesthenumberofemigrants,asaround60%ofallLithuanianemigrantswenttotheUK.Asaconsequence,thecoecientsaresignicantlylargerthaninthebaselinescenario.[InsertTable5here]40 C.2EducationGroupsTheLithuanianeducationsystemoersavarietyofeducationaltracksanddegrees.20Iaggregatethedierenteducationlevelsintothreebroadeducationgroupsfortworeasons:Firstly,theIrishcensusonlyincludesvedierenteducationgroups(primaryandlower,lowersecondaryschool,uppersecondaryschool,third-level-nodegreeandthird-leveldegree),sothatamatchingoftheeducationalattainmentofemigrantsandstayersisonlypossibleifbroadereducationgroupsareconsidered.Secondly,insomecasesdierenteducationaltracksinLithuanialeadtocomparabledegrees.Forexample,thebasicschool,whichstudentsnishattheageof16,andthestageIofvocationaltraining.Bothofthosetracksleadtoabasicschoolleavingcerticate.Thus,studentsholdingeitherofthosecomparabledegreescanbeseenasclosesubstitutesonthelabormarketandshouldbeequallyaectedbytheemigrationofworkerswithcomparablecharacteristics.Tables1and2showthedistributionoftheeducationlevelsintheLithuanianHBSaswellasintheIrishcensus.Ithereforedenetheeducationgroupsasfollows:Lowersecondaryschoolandless,uppersecondaryschoolandthird-leveldegree.LowerSecondarySchoolandLessPeoplewith10yearsofschoolingorless.AstheLithuanianHBScontainsveryfewobservationswithprimaryschooleducationorless,Imergethesewiththecategorylowersecondaryschool.Therefore,intermsoftheLithuanianclassication,thiscategoryincludeshighschooldropouts,workerswhoonlynishedprimaryschool,thosewithabasicschoolleavingcerticate(usuallyobtainedattheageof16)andthosewhopursuedstageIofvocationaltraining,whichalsoleadstoabasicschoolleavingcerticate.IntheIrishcensus,thisgroupconsistsofprimaryschoolandlessandlowersecondaryschool. 20http://www.euroguidance.ltprovidesanoverviewoftheLithuanianeducationsystem.41 UppersecondaryschoolThiscategoryincludesallworkershavingadegreehigherthanabasicschoolleavingcerticate(i.e.atleast11yearsofschooling),butdonotholdadegreethatwouldallowthemtoenteramasters'programmeatauni-versityinLithuaniaorabroad.ThedominantdegreeinthiscategoryistheLithuanianA-level,usuallyobtainedattheageof18.TheotherdegreesofthiscategoryarestagesII,IIIandIVofvocationaltrainingandcerticatesfromnon-universitythird-levelinsti-tutions.IntheIrishcensus,thiscategorycontainsallworkerswithanuppersecondaryschooldegreeorathird-leveleducationthatdoesnotleadtoauniversitydegree.Third-leveldegreeAllworkerswithatleast15yearsofschoolingandadegreethatenablesthemtoapplyforauniversitymasters'degreeinLithuaniaorabroad.Workerswithamasters'oraPhDdegreearealsoincludedinthiscategory.42 C.3Tablesforonlineappendix43 Table5RobustnesschecksDependentvariable:logrealwage (1)(2)(3)(4)2yrcells10yrcells2002&2006Irishdata A)withoutinteractionsi)allEmigrationrate0.535***0.873*1.020**2.089**[0.194][0.454][0.419][0.996]ii)men/womenEmigrationrate0.2420.6450.7641.236[0.230][0.444][0.476][0.995]Emig*male0.634**1.055***0.842*2.761***[0.274][0.348][0.434][0.745]B)withinteractionsi)allEmigrationrate0.417*0.3340.5761.332[0.218][0.694][0.648][1.341]ii)men/womenEmigrationrate0.2171.442**0.6901.663*[0.226][0.808][0.704][0.952]Emig*male0.532*1.625**0.6372.773***[0.272][0.638][0.601][0.826] Robuststandarderrorsinbrackets***p0.01,**p0.05,*p0.1Note:Thistabledisplaysthecoecientsforaseriesofrobustnesschecks:1)2-yearexperiencecells,2)10-yearexperiencecells,3)onlydatafrom2002and2006,4)onlyIrishdata.Emig*maleisaninteractiontermoftheemigrationrateandamaledummy.Yeardummies,educationdummies,experiencedummiesandpersonalcharacteristicsarecontrolledfor.PanelA)areestimatesofthebaselinemodelinEquation(5).PanelB)enhancesthebaselinemodelbytheinteractiontermsyeareduc,yearexper,andeducexper.Robuststandarderrorsinbrackets.Standarderrorsareclusteredattheeducation-experience-yearlevel.Signicancelevels:***p0.01,**p0.05,*p0.1.44 C.4Figuresforonlineappendix Figure4StandardizedwagedistributionformenandwomeninLithuania,2002and2006Note:ThegraphshowsaKerneldensityplotofthelogrealwagesin2002and2006formenandwomen.Tomakethedistributioncomparableacrossyears,wagesarestandardizedtotheirz-scores,i.e.thewageofanindividualminusthemeanwage,dividedbythestandarddeviationofwages,zi=(wi�w)=w.Themeanofthedistributioniszero.Source:LithuanianHouseholdBudgetSurvey45