immigrant native substitutability: role abilityethang/immnat.pdfaffect substitutability with...
TRANSCRIPT
1
Immigrant‐NativeSubstitutability:theRoleofLanguageAbility
EthanLewis*
DartmouthCollegeandNBER
December2012
Abstract.Wageevidencesuggeststhatimmigrantworkersareimperfectlysubstitutablefornative‐bornworkerswithsimilareducationandexperience.UsingU.S.CensusesandrecentAmericanCommunitySurveydata,Iasktowhatextentdifferencesinlanguageskillsdrivethis.Ifindtheyareimportant.Iestimatethattheresponseofimmigrants’relativewagestoimmigrationisconcentratedamongimmigrantswithpoorEnglishskills.Similarly,immigrantswhoarriveatyoungages,asadults,bothhavestrongerEnglishskillsandexhibitgreatersubstitutabilityfornative‐bornworkersthanimmigrantswhoarriveolder.InU.S.marketswhereSpanishspeakersareconcentrated,Ifinda“Spanish‐speaking”labormarketemerges:insuchmarkets,thereturntospeakingEnglishislow,andthewagesofSpanishandnon‐Spanishspeakersrespondmoststronglytoskillratiosintheirownlanguagegroup.Finally,inPuertoRico,wherealmostallworkersspeakSpanish,Ifindimmigrantsandnativesareperfectsubstitutes.Theimplicationsforimmigrantpovertyandregionalsettlementpatternsareanalyzed.
JEL:J24,J31,J61Keywords:immigration,language,wages,labormarkets
*IthankStevenRaphaelandDavidCard,theeditorsoftheRussellSagevolumeImmigration,PovertyandSocioeconomicInequalityinwhichafutureversionofthispaperwillappear,andmycolleaguesatDartmouthforhelpfulcommentsandChrisBachand‐Parenteforoutstandingresearchassistance.IalsothanktheRussellSageFoundationandDartmouthCollegeforfundingthisresearch.
2
StudieshavefoundthatthemassiveflowofimmigrantsintotheU.S.inthepastfewdecadeshas
hadlittlenegativeimpactontheaveragewagesofnative‐bornworkers(reviewsincludeBorjas,
1994,andFriedbergandHunt,1995;amorerecentarticleisOttavianoandPeri,forthcoming,
hereafter“OP”).However,manyofthesesamestudiestendtofindthatthenewarrivals
substantiallydepressedthewagesofpreviousimmigrantarrivals(Card,2001;OP).1Thefactthat
immigrantsseem,bythisevidence,nottofullycompeteinthenativelabormarketmaycontribute
totherelativelyhighratesofpovertyamongimmigrants,andunderstandingwhytheydonotmay
informpoliciesaimedatreducingimmigrantpoverty.
AplausibleexplanationforwhyimmigrantsdonotfullycompeteisthattheirlimitedEnglishskills
restrictthemtooccupationswhereEnglishskillsarelessimportant.Theabilitytospeaka
country’snativelanguageisassociatedwithhigherwages(e.g.,ChiswickandMiller,1995,2007;
Carliner,1996;DustmannandFabbri,2003;BleakelyandChin,2004;Ferrer,GreenandRiddell,
2006).Inaddition,less‐skilledimmigrantsintheU.S.appeartospecializeinoccupationswhich
requirerelativelylittlecommunication(PeriandSparber,2009),althoughfactorsbesides
immigrants’relativelackofEnglishskills(suchasagreaterwillingnesstoworkinmanualjobs)
couldcontributetothis.
Inthischapter,Iaskhowimmigrants’languageskillsaffecthowcloselytheycompetewithnative‐
bornworkers.Aswasdescribedintheintroduction,howcloselytwodifferentgroupsofworkers
(inthiscase,immigrantsandnatives)competeinthelabormarketisrevealedbyhowcloselytheir
wagesmovetogetherinresponsetochangesinthesizeofonegrouprelativetotheother.In
economicsjargon,twogroupsofworkersaredescribedas“perfectsubstitutes”iftheirwagesmove
insyncinresponsetochangesinrelativenumbers:fromthelabormarket’sperspectivetheyare
1InOP’spreferredspecification,forexample,thelargestestimatedimpactofimmigrationsince1990theycanfindontheaveragewagesofnative‐bornhighschooldropoutsis‐0.1percent,comparedto‐8.1percentforimmigrants.Muchofthemorerecentliterature(e.g.,Card,2001,2009)doesnotexamineimpactsonaveragewagesdirectly,butinsteadfocusesonthewagegapsbetweengroupsofnativeworkers.Evenmeasuredthisway,however,theimpactonless‐skillednativesisgenerallyfoundtobesmall.
3
identical.Incontrast,twogroupsare“imperfectsubstitutes”ifanincreaseinthesizeofonegroup
relativetoasecondgrouplowerthewagesofthefirstgrouprelativetothesecond.Themore
responsiverelativewagesaretorelativesupply,thelessdirectlythesetwogroupscompeteinthe
labormarket.IntheU.S.,immigrantsandnativeswithsimilareducationandexperienceare
imperfectsubstitutes(OP).
Putanotherway,then,thischapteraskswhetherimmigrantswithstrongEnglishlanguageskills
areclosersubstitutesfornativesthanimmigrantswithpoorEnglishlanguageskills.Specifically,it
asksiftherelativewagesofimmigrants(thatis,relativetonatives’wages)whospeakEnglishwell
respondmoretochangesintherelativetotalhoursworkedbyimmigrantsthandotherelative
wagesofimmigrantswhospeakEnglishpoorly.Toestimatethisrelationship,Irelyprimarilyon
variationacrossmetropolitanareas,usingdatafromthe2000U.S.CensusofPopulationand
AmericanCommunitySurveysfrom2007‐2009(Rugglesetal.,2010).Allcomparisonsaremade
withinbroadeducationgroups(highschoolorless,morethanhighschool)thatpreviousresearch
hasfoundrepresentdistinctlabortypes(Card,2009;GoldinandKatz,2008).Thesedatacontain
self‐reportsofEnglishlanguageskills.Aslanguageskillsmaybecorrelatedwithotherfactorsthat
affectsubstitutabilitywithnatives,suchaslegalstatus,Ialsomakecomparisonsbyageatarrival
andtimeintheU.S.,exploitingthefactthatimmigrantswhoarriveintheU.S.aschildrentendto
havefarbetterEnglishlanguageskillsasadults(BleakleyandChin,2004),and,similarly,that
EnglishskillstendtorisewithtimeintheU.S.Theideaisthatlegalstatus(andotherfactors)may
haveaweakerrelationshipwithEnglishskillsacrossthesegroups.
TheabovediscussioniswrittenasifallcommunicationmusttakeplaceinEnglish,butinsome
partsoftheU.S.,Spanishisalsoanimportantlanguage.AnextremecaseisPuertoRico,whereboth
“immigrants”(whoincludeforeign‐bornfromLatinAmericancountries,aswellasU.S.bornethnic
PuertoRicans)andnativesspeakSpanish.Asafurthertestofthelanguagehypothesis,Iask
4
whetherwagemovementsrevealPuertoRicanimmigrantsandnativestobeperfectsubstitutes.
SincePuertoRicoisusuallyconsideredtobeasinglelabormarket(e.g.,Borjas,2008),itisnot
possibletoexploitgeographicvariationinthesizeoftheimmigrantpopulation.Instead,this
analysisreliesonvariationovertimeandacrossdetailededucationandexperiencegroupsinthe
relativesizeoftheimmigrantpopulation,similartoOP,usingPuertoRicanCensusesfrom1970to
2000alongwiththe2007‐2009PuertoRicanCommunitySurveys(Rugglesetal.,2010).Ialso
performaparallelaggregateanalysisofthecontinentalU.S.
InlessextremecasesthanPuertoRico,Spanishcanbeadominantlanguage,likeinMiamiandLos
Angeles,orprevalentifnotdominant,likeinChicagoandSanFrancisco.Ialsoconsiderwhether
thedensityofSpanishspeakersaffectshowsubstitutableSpanishspeakingimmigrantsarefor
natives.Underwhatconditionsdoesaparallel“Spanish‐speaking”labormarketdevelopinthe
area?AkeyhypothesisfromtheoryisthatitrequiresnotjustalargenumberSpanishspeakers
overall,butasufficientlyrichdistributionofskillsamongSpanishspeakers.This,forexample,
distinguishesMiami‐‐wheremajorityofbothhighlyeducatedandlesseducatedworkersspeak
Spanish‐‐fromChicago,whereSpanishspeakersaredisproportionatelylesseducated.
Nearlyalloftheresultsareconsistentwiththeviewthatlanguageskillsareanimportantsourceof
imperfectsubstitutabilitybetweenimmigrantsandnatives.Theestimatesimplythattheincrease
inthelaborsuppliedbyimmigrants(relativetonatives)since1990intheaveragemetropolitan
areaisassociatedwithabouta6percentdeclineinthehourlywageofimmigrantswithpoor
Englishlanguageskills,butonlya2percentdeclineinthewagesofimmigrantswithstrongEnglish
languageskills(eachrelativetonativeswithsimilareducationandworkexperience).Ialsofind
thatthewagesofimmigrantswhoarrivedasyoungchildrenandlongagoarelesssensitiveto
immigrantrelativesupplythanthosewhoarrivedandolderagesandrecently.Alsoconsistentwith
5
languagedrivingimperfectsubstitutability,inPuertoRicoimmigrationisnotassociatedwitha
declineinthewagesofPuertoRicanimmigrantsrelativetoPuertoRico‐bornworkers.
WithintheU.S.,severalpiecesofevidencesuggestaparallelSpanish‐speakinglabormarket
emergesinareaswhereSpanishspeakersaresufficientlynumerous,and,inparticular,wherethe
skilldistributionofSpanish‐speakersissufficientlyrich.First,forSpanishspeakers,thewage
premiumforspeakingEnglishissmallinmarketswithaheavySpanish‐speakingpresence,
particularlythosewithaheavyeducatedSpanish‐speakingpresence.Second,aninfluxofless‐
educatedSpanishspeakersisassociatedwithalargerdeclineinthewagesofSpanish‐speaking
thannon‐Spanish‐speakingless‐educatedimmigrants.Third,aninfluxofeducatedSpanish‐
speakersisassociatedwithanincreasethewagesofSpanish‐speakingbutnotnon‐Spanish‐
speakingless‐educatedimmigrants.Thereasonthisthirdfactisrelevantisthatinstandardmodels
(liketheoneusedinChapter2)aninfluxofmoreeducatedworkersispredictedtoraisethewages
oflesseducatedworkersiftheyareinthesamelabormarket.Thus,thisthirdfactalsosuggests
Spanish‐andnon‐Spanishspeakersoperateinparallellabormarketsinthesamearea.
Thefindingsrefineourunderstandingoftheforceswhichaffectimmigrantpoverty.Previous
researchalreadysuggestedthatthewageimpactsofimmigrationwerebornedisproportionatelyby
immigrantsthemselves,(e.g.,Card2001;OP)andtheseresultssaythattheyareborneparticularly
stronglybyimmigrantswithpoorEnglishlanguageskills,whoareamongthepoorestimmigrants
(inrecentU.S.data,29percentofimmigrantswhodonotspeakEnglishwereinpoverty,compared
to15percentofimmigrantsoverall),andbySpanish‐speakingimmigrants.Theestimateshereare
consistentwiththeincreaseinimmigrationsince1990contributinganadditionalonetotwo
percentagepointstothepovertyrateofless‐educatedlow‐Englishimmigrants,andanadditional
twotofourpercentagepointstothepovertyratesofless‐educatedSpanish‐speakingimmigrants.
6
However,povertyratesamongtheseimmigrantgroupsfelloverthisperiod,sothisresultonlysays
thatimmigrantpovertyrateswouldhavefallenmorequicklywithoutadditionalimmigration.2
I. MotivationandBackground
Itisusefultobeginwithsomebasicfactsaboutimmigrants’andnatives’languageskillsandwages.
PanelAofTable1showsaverageEnglishskillsforimmigrantsandnatives.Itwasconstructedby
combiningdatafromthe2007,2008and2009AmericanCommunitySurveys(Rugglesetal.,2010)
fortheworking‐ageresidentsof136highimmigrationmetropolitanareasdescribedbelow.3These
metropolitanareascontainover80percenttheimmigrantslivingintheU.S.Table1alsoshows
separatemeansbywhetherornottherespondenthasanycollegeeducation.Throughoutthe
paper,thosewithahighschooldegreeorlesswillbereferredtoas“lesseducated”andwillbea
focusoftheanalysis,astheymakeupadisproportionateshareofthoseinpoverty.
ThefirstlineofPanelAshowsthatonly46percentofimmigrantsspeakEnglish“only”or“very
well.”Incontrast,amongthenative‐born,thisnumberisover98percent(lowerhalfoftable).Toa
usefulfirstapproximation,U.S.nativesarefluentinEnglish,whereasonlyhalfofimmigrantsare.
Thelatterrisesto68percentifyouincludeimmigrantswhosaytheyspeakEnglish“well”(butnot
“verywell”).BythisbroadermeasureofEnglishproficiency,abouthalfofimmigrantswithout
collegeareproficientinEnglish,and90percentofimmigrantswithcollegeeducationareproficient
inEnglish(columns2and3).
2Althoughpovertyratesroseamongimmigrantsoverallduringthisperiod–seeintroduction–povertyratesamonglesseducatednon‐Englishspeakingimmigrantsfellfrom35to30percentbetween1990and2008,andamongSpanishspeakersfrom25to22percent.3“Workingage”isdefinedasbeingbetweenage16and65andwithatleastoneyearof“potentialworkexperience”whichmeansbeingoldenoughtohavespenttimeoutsideofschoolgivennormalprogressionthroughschool.
7
Spanishspeakerswillalsobeafocusofthisstudy,andPanelBshowstheEnglishskillsofjustthose
immigrantsandnativeswhoreportspeakingSpanishathome.Spanish‐speakingimmigrantshave
belowaverageEnglishskills,withonly26percentfluentinEnglish;evenwithineducationcategory
theirEnglishskillsarebelowaverage.ThisismostlydrivenbyMexicanswhoarethelargest
immigrantgroupandwhohavepoorEnglishskills.(74percentofMexicanimmigrantsinthis
samplereportspeakingEnglishnotwellornotatall.)EvenU.S.‐bornSpanishspeakers,whomake
up7.7percentoftheworking‐agepopulation,haveimperfectEnglish.Only82percentclaimtobe
fluentinEnglish,andalmost7percentsaytheydonotspeakEnglishwelloratall.4
TheanalysisbelowalsoexploitsvariationinEnglishskillsbyimmigrants’ageatarrivalandtimein
theU.S.Figure1ashowsaveragelanguageskillsbythesecharacteristics.Thesharewhospeak
EnglishwellhasapositivemonotonicassociationwithtimeintheU.S.,risingfromhalfofthosewho
arrivedinthepastfiveyearsto90percentofthosewhohavebeenintheU.S.atleast40years(a
patternwhichmaypartlyreflectcohortdifferencesinEnglishskills).Figure1aalsoshowsthe
sharpdeclineintheshareofimmigrantswhospeakEnglishwell(asadults)inageatarrival:itis
higheramongthosewhoarrivedbeforeage10comparedtothosewhoarrivedatolderages.This
factwasexploitedinBleakleyandChin(2004)(BC)tostudytheeffectoflanguageskillsonwages.
Theyarguedthatthereisa“criticalperiod”atyoungageswhenchildrenareabletoeasilylearn
English.Below,Iaskifthereisasimilarkinkedrelationshipinthesubstitutabilityofimmigrants
fornativesbyageatarrival.5Figure1bshowsthatthiskinkedrelationshipismainlypresent
amongless‐educatedimmigrants;amongmoreeducatedimmigrants,therelationshipissmoother.6
4Thesenumbersmayunderstatenative‐bornSpanishspeakers’EnglishskillsifthereisatendencyforundocumentedMexicanstoclaimnative‐bornstatus.5NotethatIamnotfullyimplementingBC’smethodshere.AkeythingIamnotdoing,buttheydid,isdifferencingoutage‐at‐arrivalpatternsamongimmigrantsfromEnglish‐speakingcountriestoaccountforotherfactors,besideslanguage,associatedwithage‐at‐arrival.WageresponseestimatesforthesmallnumberofimmigrantsfromEnglish‐speakingcountriesweretoounreliabletoexploitthisapproach.6SplittingthesamplebyageatarrivalandeducationissomewhatproblematicgivenBC’sfindingthatarrivingasayoungchildtendstoraiseeducationalattainment.Theappropriateapproachinlightofthisistoaggregatetogethermore‐and
8
SuggestiveevidencethatEnglishskillsareimportantforeconomicwell‐beingisshowninTable2,
whichshowsmeanloghourlywagesandpovertybynativity,education,andEnglishandSpanish
languageskills.ThefirsttworowsofPanelAshowthatevenwithineducationcategory,
immigrantstendtoearnlessthannatives.Forexample,lesseducatedimmigrantsearnonaverage
18logpoints(about18percent)lessthanlesseducatednatives.Thenextrowshows,pertinentto
theideathatlanguageskillsmightmatterforthiswagegap,thewagesofimmigrantswhoarefluent
inEnglishareverysimilartothatofnatives,includingamongless‐educatedimmigrantsand
natives.Thereisalsoasteepwagegradientinself‐reportedEnglishskills,shownbelowthat.
Thesewagedifferencestranslatetodifferencesinpovertyaswell,showninpanelB.Whereas30
percentofless‐educatedimmigrants(andeven23percentofthosewithcollegeeducation)whodo
notspeakEnglishareinpoverty,only14‐16percentofless‐educatedimmigrantswhospeak
Englishwellorverywellareinpoverty,similartopovertyratesamongless‐educatednatives.
Finally,thebottomrowsofTable2showthatSpanishspeakersareworseoffthanthetypical
immigrant,somethingtheirpoorEnglishskills(Table1)likelycontributesto.Immigrantswho
speakonlySpanish,showninthebottomrowofthetable,havewagesandpovertyratesaboutthe
sameasthetypicalnon‐Englishspeaker.
Thewagegapbetweenlesseducatedimmigrantsandnativesisanalyzeddirectlyinmultivariate
regressionsinTable3.Column(1)repeatsthefindinginfromcolumn(2)ofTable2,thatless‐
educatednativesearnabout18percentlessthanless‐educatednatives.Column(2)ofTable3
showsthatasinglecontrolvariable‐‐adummyforspeakingEnglishonlyorverywell‐‐can
accountformostofthisgap.Thecoefficientonthiscontrolsuggeststhatthereisa21percentwage
premiumtospeakingEnglishfluently,afindingwhichisconsistentwithpreviousestimatesofthe
returnstospeakingEnglishintheU.S.labormarkets(e.g,ChiswickandMiller,1995,andCarliner,
less‐educatedworkers.Becausetherestoftheanalysisissplitbyeducation,however,Iwillsplititthiswayfortheestimatesbyage‐at‐arrival.
9
1996).ThislikelyoverstatesthecausaleffectofEnglishspeakingability,however.Column(3)
showsthattheadditionofsimpledemographicandskillcontrolsreducesthemagnitudeofthis
coefficient.EstimatesinBC’sstudyexploitingageatarrivalare,infact,consistentwithnocausal
effectofEnglishlanguageskillsonwages.Theycouldaccountforwagegapsacrossimmigrants
withvaryingEnglishlanguageskillsentirelywitheducationdifferencesacrossthesegroups.
ThelastthreecolumnsofTable3focusonthelargeminorityofAmericans‐‐bothimmigrantsand
natives‐‐whospeakSpanishathome.Column(4)showsthatthereisasomewhatsmallerreturn
toEnglishfluencyamongSpanish‐speakers,andthat,conditioningonEnglishfluency,thereisno
immigrant‐nativewagegap.MightEnglishskillsmatterlesswhentherearelargenumbersofother
Spanishspeakersinthesamelabormarket?Tofindout,column(5)addsaninteractionbetween
theEnglishfluencydummyandtheshareofthemetroarea’spopulationwhospeakSpanishat
home.Thecoefficientonthisinteractionisnegativeandsignificant,consistentwiththeideathat
Englishskillsbecomelessvaluable(forSpanishspeakers)asthesizeoftheSpanish‐speaking
populationincreases.Indeed,thiscontrolraisesthecoefficientonEnglishfluencydummybackto
itslevelinthefullsample;thatistosay,thefactthatSpanishspeakersaregeographically
concentratedfullyaccountsfortheirloweraveragereturntospeakingEnglish.Column(6)splits
theSpanish‐speakingshareintosharesamongthemore‐andless‐educatedpopulations.
Consistentwithwhatwashypothesizedintheintroduction‐‐thatitrequiresasufficientlyrich
distributionofskillstocreateaSpanish‐speakinglabormarket‐‐itisonlytheSpanish‐speaking
shareamongmoreeducatedworkerswhichisassociatedwithadiminishedimportanceofEnglish
languageskillsamonglesseducatedworkers.Totakeanextremeexample,inmarkets,likeMiami,
whereamajorityofcollegeeducatedworkersspeakSpanish(seeAppendixTableA1forother
examples)theseestimatespredicttherewillbenopremiumtoEnglishfluency.7Below,Ireassess
7Thoughthisresultisquitepreliminary,itdoesprovidesomecounterweighttotheevidencethatthetendencyforimmigrantstogeographicallyclusterisbadforthem(e.g.,Cutler,Glaeser,andVigdor,2008).Tobefairtothoseauthors,
10
howtherelativewagesofSpanish‐speakingimmigrantsisaffectedbythedensityofSpanishand
Englishspeakersinthelabormarket,usingtheestimationframeworkIdescribenow.
II. TheoryandDerivationofanEstimationEquation
AstartingpointforasimpletheoryofhowlanguageskillsmatterintheU.S.labormarketisthe
notionthatamongthosewithotherwisesimilarskills,thosewhocannotcommunicateinEnglish
well(hereafter,“speak”Englishforshort)imperfectlysubstituteforworkerswhodo;theformer
mightnotbeveryeffectiveinoccupationswhichrequirealotofcommunication,forexample(Peri
andSparber,2009).Forsimplicity,imagineworkerscanbesharplydividedintothosewhocanand
cannotspeakEnglish,indexedwithj=1andj=0,respectively.Ifthesetwotypesareimperfectly
substitutable,thewagepremiumtobeingabletospeakEnglishwilldeclineintherelativenumber
ofworkerswhospeakEnglishandwhodonot,whichIcapturewiththefollowingrelationship:
(1) 1010 lnln LLbaww
Ljisthenumberofworkersandwjisthewageoflanguagetypejworkers,andaandbarepositive
constants.bmeasuresthedegreeofimperfectsubstitutability:itwillbezeroifthosewhocanand
cannotspeakEnglishareperfectsubstitutes.8Inprinciple,(1)couldbeestimatedusingvariation
acrosslabormarketsandovertimeintherelativenumberofworkerswhospeakEnglish“only,”
“verywell”or“well”asaproxyforL1andtheremainingworkersasaproxyforL0.Inordertobe
though,theiranalysisisattheneighborhood,ratherthanmarketlevel.Immigrantsegregationandlanguageisolationarealsoanalyzedinchapter4ofthisvolume.8(1)fallsoutofacapital‐neutralsingle‐goodnestedCESproductionfunctionrepresentationoftheeconomy,wheretheinnermostnestcontainsworkerswithandwithoutEnglishskillsinthiscase,similartotheframeworkusedinchapter2.Inthisinterpretation,thecoefficientbrepresentsaninverse“elasticityofsubstitution”betweenlanguagetypes,and“a”representsademandshifterwhichisafunctionoffactorshareandproductivityparametersembeddedintheproductionfunction.Other,moregeneral,functionalformsarepossible,butthisCES‐derivedapproachiscommon(e.g.,OP;RaphaelandSmolensky,2008).
11
consistentwithpriorestimatesoftheimpactofimmigrationonwages,however,itisusefulto
translate(1)intosomethingwhichdirectlyinvolvesimmigrantsandnatives.9
Todoso,firstrecallthatnearlyallnativesreportspeakingEnglishfluently.Therelativenumberof
non‐Englishspeakersisdrivenalmostentirelybyimmigration,and,inpractice(demonstrated
below)movesalmostone‐for‐one(inpercentterms)withtherelativenumberofimmigrants.
Mathematically, NF LLcLL lnln 10 ,whereLFandLNare,respectively,thenumberof
foreign‐bornandnative‐bornworkers.Imposingthislinearapproximationon(1),wehavethat:
(1’) NF LLbaww lnln 10
Totranslatethelefthandsideintothewagesofimmigrantsandnatives,firstnotethatsinceall
nativesareassumedtospeakEnglishwecanimposewN=w1.(Inpractice,below,nativeswho
reportimperfectEnglishwillbedroppedfromthewagesample.)Asforimmigrants’wages,the
analysiswillexplorevariationacrossgroupsofimmigrantwithvaryingEnglishability.In
particular,Iwilllookacrossthetenyear“age‐at‐arrival”andfiveyear“yearsinU.S.”categories
showninFigure1(inadditiontodirectlyacrosstheEnglishlanguageskillcategoriesshownin
Tables1and2).Sosupposegindexesthesedifferentcategoriesofimmigrants,andfractiongof
groupgspeaksEnglish.IcanwritethemeanlogwageofgroupgimmigrantsaslnwFg=glnw1+
(1‐g)lnw0+g,wheregrepresentssourcesofimmigrant‐nativewagegapsotherthanEnglish
skills(forexample,ethnicdiscriminationorlegalstatus.)Translatingthisintoaimmigrant‐native
wagegap:
9Anotherpotentialreasontodosoisevidencethatself‐reportedmeasuresoflanguageskillsarenotreliable(DustmanandvanSoest,2001,2002;DustmannandFabbri,2003).Despitethis,theresultsbelowaresimilarlystrongwhenusingself‐reportedEnglishskillsdirectly,suggestingthattheyareatleastreliableenough(attheaggregatelevel)tobeuseful.
12
gg
gg
gggNFg
ww
ww
wwwww
10
10
101
ln1
lnln1
lnln1lnln
Substituting(1’)intothisproduces
(2) gNF
ggNFg LLbaww ln1ln
Theintercept,ag,isacombinationofconstants.(2)impliesthatamongotherwisesimilar
immigrantsandnatives,thesensitivityoftherelativewagesofforeignworkerstochangesin
foreignrelativesupplyisdiminishingintheshareofforeignworkersspeakEnglish.Forexample,
immigrantswhoarriveaschildrentendtohavebetterEnglishskillsthanthosewhoarriveas
adults,andso(2)impliesthattherelativewagesofthosewhoarriveaschildrenshouldrespond
lesstoimmigrantinflowsthanimmigrantswhoarriveasadults.
Thissimplemodelleavesoutseveralthings.First,accordingtothemodel,immigrantswithperfect
English,g=1,areperfectsubstitutesfornatives;theirrelativewagesareinsensitivetoimmigrant
relativesupply.Animportantsimplificationusedtoderive(2)wasthatothersourcesofimmigrant‐
nativewagegaps(g)areunrelatedtoimmigrantrelativesupply.Ifthisisnotthecase,theneven
fluentimmigrants’relativewagesmaybesensitivetothenumberofimmigrantsrelativetonatives.
Thismayalsobiasestimateof(2),anissuewhichwillbediscussedfurtherbelow.
Second,thismodelassumesthatEnglishskillsareequallyimportantforalljobs‐‐the“b”in(1)isa
constant.10AsitisplausibleEnglishskillsaremoreimportantinhigh‐skilljobs,theestimates
belowallowtheeffectstovarybytheeducationleveloftheworker.
10ThiswouldbeofgreatconcernonlyifthevariationinthewageresponseacrossimmigrantgroupsweredrivenbyvariationintheimportanceofEnglishratherthantheEnglishskillsoftheimmigrantsinthegroup‐‐thatis,variationin“b”noting.
13
Finally,thissimplemodelcanalsoonlypartlyaccommodatethefactthatinsomepartsoftheU.S.,
Spanish,notEnglish,isthedominantlanguage.InPuertoRico,wherebothimmigrantsandnatives
speakSpanish,(2)appliesandimpliesimmigrants’relativewagesshouldnotrespondtothe
relativenumberofimmigrants.InmarketswithamixofSpanish‐speakingandEnglish‐speaking
workers,itisnotclearwhatwillhappen,butsometheoriesaresuggestive.AccordingtoLang’s
(1986)theoryoflanguagediscrimination,wagesarelowerforSpanish‐speaking(orgenerally,non‐
Englishspeaking)immigrantlaborersthannativesbecausetheybearthecostoftrainingabilingual
supervisor;however,whereSpanishisspokenbyamajorityofworkers(asitisinsomeU.S.
markets),thesignofwagegapwithnativesreverses.InPeriandSparber(2009),less‐educated
immigrantsaresegregatedintomanualoccupationsbecauseoftheirinferior(English)
communicationskills.Thoughtheydidnotdiscussitinthepaper,onemightimaginethatwitha
largeenoughdensityoffellowSpanishspeakers,itmightbepossibleforSpanishspeakerstohave
accesstoafullrangeofoccupations.11Inboththeories,havingenoughskilled(Lang’s
“supervisors”)oreducatedSpanishspeakerswouldbeimportantforaseparate“Spanish”labor
markettoemerge,whichissupportedbypreliminaryevidenceinTable3.Thiswillbeevaluated
belowbyaddingtermsto(2)measuringthesizeoftheSpanish‐speakinglaborpoolbyeducation.
III. EstimationandIdentification
Themainestimatesof(2)willusevariationacrossskillgroupsandmetropolitanareasinthe
relativeaggregatehoursworkedoftheimmigrantpopulation,asfollows:
(3) gictNict
Fictggit
Nict
Fgict HHww lnln
11Inparticular,onemightassumetheoutputofPeriandSparber’sproductionfunctionmadewithSpanish‐speakingworkerswasperfectlysubstitutablefortheoutputmadewithEnglish‐speakingworkers.
14
whereiindexestwoeducationgroups(highschoolorless,morethancollege)andcindexes
metropolitanareas,andHrepresentstheaggregatehoursofthespecifiedgroup.gistheestimate
of‐(1‐g)bin(2)whichisexpectedtobenegativeandtobesmallerinmagnitudeforimmigrants
withstrongerEnglish‐languageskills.(3)willbeestimatedbothjointlyandseparatelyby
educationgroup.
Allestimatesof(3)includetime‐varyingeducationgroupcontrols,git,which,liketheslope,willbe
allowedtovaryacrossimmigrantgroups,g.Thesecontrolscaptureeconomy‐widechangesinthe
wagestructure.Thedependentvariablewillbecomputedasthedifferenceinthemeanlogwages
ofimmigrantsingroupgandnativeswiththesameeducation,i,inthesamemetropolitanarea,c,
andyear,t,or“cellict”forshort.12Toreducetheinfluenceofcompositionaldifferencesbetween
immigrantsandnativesonthisestimatedmeanwagegap,natives’meanwageswillbecomputed
usingweightsthatgivethem,onaverage,thesameeducationandexperienceastheimmigrantsin
groupgincellict.13(Inpractice,therawwagegapproducessimilarresults.)(3)willbeestimated
byordinaryleastsquares(thatis,unweighted)andstandarderrorswillbecomputedtoberobust
toarbitraryerrorcorrelationacrossobservationsonthesamemetropolitanarea.
Theerrortermin(3)capturesotherdeterminantsofimmigrants’relativewages.Therearetwo
broadreasonstoexpectthatitwillbecorrelatedwithimmigrants’relativehours,whichwilllead
estimatesofgtobebiased.Areaswithahighrelativedemandforsomeimmigrantsubgroup
wouldtendtosimultaneouslyhavehighrelativewagesandhoursfortheworkersinthisgroup,
therebygeneratingapositivecorrelationbetweentheerrortermandtheexplanatoryvariableand
12Thisusesthefactthat N
ictFgict
Nict
Fgict wwww lnlnln .
13Inparticular,thosewithhighschoolorlessaredividedintohighschoolcompletersandhighschooldropouts,andthosewithsomecollegeormorewillbedividedintothosewithandwithoutfouryeardegrees.Withinthesecells,workersarefurtherdividedintofive‐yearpotentialexperiencebandsupto40.Themeanofnativelogwagesarecomputedweightedbyp/(1‐p),wherepisthefractionofeachdetailededucationxexperiencexmetropolitanareaxyearcellthataregroupgimmigrants,amonggroupgimmigrantsandnativesinthatcell.(ThisweightisinteractedwiththeACSorCensussampleweight.)Cellswithnonativesornogroupgimmigrantsaredropped.
15
leadingtoslopeestimateslessnegative.14Astandardwaytoaddressthisistousepredictable
variationinthesizeofimmigrantinflowsbasedonthelabormarketsinwhichimmigrantsfrom
differentpartsoftheworldtendtocluster(MexicansinLA,RussiansinNewYork,etc.)This
approachisdescribedintheAppendix.Inpractice,ittendstoproducesimilarresultstothe
ordinaryleastsquaresestimatesthatarepresentedbelow,suggestingthatthistypeofbiasmaynot
belarge.Inaddition,insectionIV.C.,Iwillpresentestimatesof(2)thatrelyonlyonaggregate
variation‐‐thatis,acrosseducationxexperiencegroups,similartoOP‐‐ratherthanvariation
acrosslabormarkets.Thisalsoproducessimilar,iflessprecise,estimates.
(3)isalsolikelytobebiasedbecauseimmigrantswithpoorEnglishlanguageskillstendtobe
dissimilarfromU.S.nativesinotherwayswhichaffecttheirsubstitutabilitywithnatives.For
example,thelargestgroupofimmigrantswithpoorEnglishskillsisMexicans,manyofwhomreside
intheU.S.illegally.TheirlegalstatusmayconfineMexicanstoparticularoccupations,makingit
harderforthemtocompeteheadtoheadwithnatives.Ifso,immigrantrelativehoursmaybemore
negativelycorrelatedwithlow‐thanhigh‐Englishimmigrants’relativewages,but(atbest)only
partlybecauseoftheobserveddifferenceEnglishskills.Intheabsenceofaperfectwaytoidentify
immigrantswhodifferonlyinEnglishskillsandnototherfactors(legalstatusinparticularisnot
observable),theapproachItakeistoexaminevariationacrossdifferentimmigrantsubgroups,such
asbyageatarrival.AlthoughthevariationinEnglishskillsacrossthesesubgroupsisalsolikelyto
alsobecorrelatedwithdifferencesinotherfactorswhichaffectimmigrant‐nativesubstitutability,
thehopeisthattherelationshipisnotasstrongoratleastnotassystematicacrosstheseother
subgroups.Forexample,thehopeisthatage‐at‐arrivaldoesnothavea“kinked”relationshipwith
legalstatuslikeithaswithEnglishskills.(SeeFigure1b.)Thisapproachislikelytobeonlypartially
14Theseestimatesarealsowell‐identifiedonlyifthegeographicunitsinvolvedactasclosedeconomies.U.S.evidencesuggeststhatlocallabormarketsbehavelikeclosedeconomiesinthesensethatimmigrantsdonotappearto“displace”native‐bornworkerswiththesameskills(e.g.,CardandDiNardo,2000)nordotheyhavemuchimpactonindustrymix(Lewis,2004).
16
successful,sotheestimatesarestilllikelytostillbebiased,probablytowardsfindingthatlanguage
skillsareimportant.
III.B. Data
Datafortheregressionanalysiscomefromthe2000CensusofPopulationand2007,2008,and
2009CommunitySurveys.Thelatterthree(whicharefromRugglesetal.,2010)arecombinedinto
whatwillbereferredtoas“2008”data.Informationonhoursworkedandhourlywageswere
aggregatedtothemetropolitanareabyyearbybroadeducationgroup,asdescribedabove.
Includedinthecalculationofworkers’hourswereworkersage16to65,withpositivepotential
experience(oldenoughtobeoutofschoolgivenanormalprogressionthroughschool),livingin
oneofthe136metropolitanareasinthesample.15Thewagesampleisthesubsampleofthese
workerswhoarecurrentlyemployedwithpositivewageandsalaryearningsandzerofarmor
businessearningsinthepastyear.16Metropolitanareasweredefinedconsistentlyusing“PUMAs”
and1990metropolitanareaboundaries.
Table4presentsthe(unweighted)meansandstandarddeviationsofthesedata.Therelativelog
hoursofimmigrantsintheaverageareaisnegativebothoverallandbyeducationlevel,whichsays
immigrantsareonaverageaminorityofworkers;inthemeaneducation‐metro‐yearobservation
immigrants’hoursrepresentabout14.4percent(=e‐1.934x100)ofnatives’hours.Thereisalotof
variationacrossthesemetroareasintherelativehoursofimmigrants:thestandarddeviationis
around1,whichwillbeusefulforinterpretingthemagnitudeoftheregressionestimatesbelow.
Thetablealsoshowstheimmigrant‐nativewagegapisaround12.5percentintheaverage
metropolitanarea,andismuchlargerforlow‐Englishimmigrants,eventheonesmatchedto 15Also,thoselivingingroupquarterswereexcludedfromtheanalysis.16Tobeincludedinthewagesample,native‐bornworkersalsowererequiredtoreportspeakingEnglish“only”or“verywell.”Hourlywagesabove$200andbelow$2in1999dollarswerereplacedwiththesethresholds.
17
nativeswithsimilareducationandexperience(asthesegapsarecalculatedtodo).Thereisalotof
variationintheimmigrant‐nativewagegapacrossmetroareastobeexplained,whichIwillnow
trytodo.
IV. Results
IV.A. BasicU.S.Evidence
Table5presentsestimatesof(3).PanelAexaminestherelativewagesofallimmigrants.The‐0.04
estimateincolumn(1)saysthataoneunitincreaseinimmigrants’relativehoursisassociatedwith
a4percentdeclineinthewagesofimmigrantsrelativetonatives.Althoughaoneunitincreasein
theindependentvariablemeansapproximatelytriplingimmigrants’laborsupply,whichsounds
likealot,itisactuallyareasonablechangetoexamineasitisbothastandarddeviation(Table4)
androughlyequaltotheincreaseinthisvariablesince1990intheaveragemetropolitanarea.17In
anycase,the‐0.04responseissimilarinmagnitudetopreviousestimatesofthisrelationship,such
asOPorCard(2009).
PanelBofTable5producesseparateestimatesforthewagesofimmigrantswhoreportstrong
Englishlanguageskills‐‐immigrantswhosaytheyspeakEnglish“only,”“verywell,”or“well”‐‐and
whoreportpoorEnglishlanguageskills‐‐immigrantswhosaytheyspeakEnglish“notwell”or“not
atall.”Asexpected,therelativewagesofimmigrantswithpoorEnglisharemorenegatively
associatedtherelativepresenceofimmigrantlabor:theirwagesdecline5.7percentforaoneunit
increaseintheindependentvariable,comparedto2.2percentdeclineforimmigrantswithstronger
English.Thedifferencebetweenthetwocoefficientsisstatisticallysignificantattheonepercent
level,shownasthep‐valuelessthan0.01intherowbeneaththeseestimates.
17Intheaveragemetroarea‐educationcellinthesample,thisvariablerosefrom‐2.69to‐1.75between1990and2008.
18
Columns(2)and(3)showestimatesseparatelybybroadeducation.PanelAindicatesthat
educatedimmigrants’wagesaremoreresponsivethanlesseducatedimmigrants’,despitethefact
thateducatedimmigrantstendtoreportsubstantiallybetterEnglishlanguageskills(Table1).This
maynotbeinconsistentwiththeimportanceofEnglish,however;itisplausiblethatlanguageskills
aremoreimportantinjobsthatrequirecollegeeducation,aviewforwhichthereissomeevidence
(Berman,Lang,andSiviner,2003).Inaddition,thedifferenceinthewageresponsesofimmigrants
withpoorandstrongEnglisharesimilarformore‐andless‐educatedimmigrants:inPanelBthe
gapincoefficientsis3.6(=0.047‐0.011)percentagepointsforless‐and3.3(=0.078‐0.045)
percentagepointsformore‐educatedimmigrants.
PanelCbreaksoutestimatesbythemoredetailedEnglishlanguageskillsshowninTables1and2.
Goingtothisdetailrequirestheexaminationofasmallernumberofmetroareas,112,thatarelarge
enoughtoobserveallimmigrantsubgroups.(Inparticular,wageearningcollegegraduateswho
reportnotspeakingEnglishatallarerare.)Thewageresponseismonotonicinself‐reported
Englishlanguageskills,bothoverallandseparatelybyeducation,andthedifferencesacrossEnglish
categoriesarestatisticallysignificant.Interestingly,thereisasignificantnegativeresponseevenin
thetopEnglishcategory,whatIreferredtoas“fluent”above.Whilesomeoftheworkersinthis
categorymaynotbetrulyfluentinEnglish,areasonableinterpretationofthisisthatthereare
factorsbesidesEnglishskillsthatmakeimmigrantsimperfectlysubstitutablewithnatives.
Figure2aexamineshowthewageresponsevariesbyageatarrivalandtimeintheU.S.Itplots
coefficientestimates(andconfidenceintervals)fromestimatesof(3)‐‐thatis,stillusingthesame
variationinimmigrantrelativelaborbybroadeducation,metropolitanarea,andyear‐‐separately
forimmigrantsinthetenyearage‐at‐arrivalcategoriesandfiveyearyearsinU.S.categoriesshown
onthefigure’sx‐axis.PanelAshowsthattheresponseofwagesbyage‐at‐arrivalfollowthesame
“kinked”patternthatself‐reportedEnglishlanguageskillsdidinFigure1a.Whileitispossiblethat
19
otherunobservedfactorsthataffectimmigrants’substitutabilitywithnativesfollowthiskinked
pattern,thisreinforcesthedirectevidencefromEnglishlanguageskillsinTable5.Ontopofthis,
thekinkedrelationshipintheresponseappearstobelimitedtotheless‐educatedsubsamplewhere
itwasfound(Figures1band2b).ThereisalsoamonotonicrelationshipwithtimeintheU.S.
(Figure2a,panelB).
IV.B. TheEffectsofLanguageSupply
Alloftheregressionspresentedsofarhaveemployedthesameindependentvariable:thenatural
logoftheratioofimmigrants’aggregatehourstonatives’aggregatehours.InthissectionIinstead
employdirectmeasuresofthesuppliesoflanguageskills,includingasthenaturallogofthe
aggregatehoursofthosewhospeakEnglishnotwellornotatall(thosewith“poorEnglish”)to
thosewhospeakEnglishonly,verywell,orwell(thosewith“strongEnglish”).Inaddition,totry
andunderstandwhathappensinmarketswithalargenumbersofSpanishspeakers,Iwilladd
controlsfortherelativehoursofworkerswhospeakSpanishathome.
TheresultsareinTable6.Tokeepthingssimple,thetableexaminesonlyworkerswithahigh
schooleducationorless.Column(1)repeatstheestimatefromPanelAofTable5:itsaysthe
overallimmigrant‐nativelogwagegapdeclines3.4percentfora1unitincreaseinimmigrants’
relativehours.Column(2)replacesthisindependentvariablewiththeonemeasuringtherelative
hoursofthosewithpoorEnglishrelativetostrongEnglish.Thecoefficientisthesametotwo
decimalplaces.ThissupportstheargumentmadeinSectionIIthat,asapracticalmatter,the
relativesupplyofEnglishlanguageskillsmovesone‐for‐onewiththerelativesupplyofimmigrants.
Column(3)repeatstheestimatesofcolumn(2)forthesubgroupofimmigrantswhoreport
speakingSpanishathome.Theestimatedwageresponseisnearlyidenticalforthissubgroup.To
20
investigateiftheresponsesvarywiththedensityofSpanishspeakersinthemarket,column(4)
addscontrolsfortheaggregatehoursofworkerswhospeakSpanishathomerelativetoworkers
withstrongEnglish.Thismeasureisenteredseparatelyforcollegeandnon‐collegeworkers,
followingtheresultsinTable3.Column(4)presentsweakevidencethatthereisanadditional
depressingeffectofhavingalargenumberofSpanishspeakingnon‐collegeworkersonthewages
ofnon‐collegeSpanishspeakers,inadditiontotheimpacthavingalargenumberofnon‐English
speakers.However,thisisoffsetbythepositivewageimpactofgreaterdensityofcollege‐educated
Spanishspeakingimmigrants.
Agraphicalversionofthecolumn(4)relationshipisshowninFigure3.ItplotsSpanish‐speaking
relativehoursamongworkerswithmorethanahighschooleducationversusthosewithahigh
schooleducationorless–thetwovariablesinthelowerhalfofTable6–in2008data,anddotted
linesindicatethemedianofeachvariable.Thedashedlineisnotthefittedline(althoughitisvery
close);rather,itindicatesthedividinglinebetweenareaswithabove‐andbelowaveragewagesfor
less‐educatedSpanish‐speakingimmigrants(relativetonatives)accordingtotheestimatein
column(4)(ignoringtheimpactoftheothervariable,poor/strongEnglish).18Inareasabovethe
dashedline,lesseducatedSpanish‐speakingimmigrantstendtohaverelativelyhighwages:this
includestwoCubanenclaves,MiamiandJerseyCity,otherareasinFlorida,aswellasareasonthe
Texasborder(notlabeledongraph),andRiversideandLosAngeles.Interestingly,many“new”
immigrantdestinations(Singer,2004)‐‐likeAtlanta,Charlotte,Denver,Greensboro,Portland,and
Raleigh‐‐arebelowtheline,indicatingtheMexican(or,really,anyless‐educatedSpanish‐
speaking)immigrantsthereearnbelowaveragewages.Thisisareversaloftheseareas’position
notsolongago–asimilargraphfor1990wouldshowmostofthesenewdestinationsabovethe
line.Infact,themodelincolumn(4)appliedto1990(CensusofPopulation)datapredictsthat
18Specifically,itisalinewithaslopeof0.7=0.045/0.065–theratioofthetwolowercoefficientsincolumn4–thatgoesthroughthesamplemedianofthetwovariables,whichareshownwiththedottedlinesinthefigure.Theresidualscatterplot,conditioningoutpoor/strongEnglish,isshowninAppendixFigureA1andhasthesamequalitativepatterns.
21
MexicanimmigrantswouldhaveearnedarelativewagesevenpercenthigherinSinger’snew
destinationsthaninotherlargeareasthatwereherhistoricalimmigrantgateways.19Therelatively
highwagesmayhavebeenpartofwhatattractedMexicanstotheseareasstartinginthe1990s,
althoughotherresearchindicatesfasteremploymentgrowthwastheprimarydraw(Cardand
Lewis,2007).Overthepast20years,thewageadvantageofthesenewareasforMexicanshas
completelyerodedbecausetheSpanishinflowshavebeendisproportionatelyless‐skilled.20
TheestimatesinTable6mightadditionallybeinterpretedasindicatingthatasufficientlyrichskill
distributionofSpanishspeakersallowsaparallel“Spanish‐speaking”labormarkettoemerge
withinanarea.Toprovideadditionalevidenceforthis,column(5)showsanestimateofthesame
relationshipfornon‐Spanishspeakingimmigrantswho,inthisview,shouldnotbeaffectedbythe
densityofSpanishspeakers.ThewageboostfromcollegeeducatedSpanishspeakersisnotpresent
forthem.Alsoincolumn(5)thereisapositivecoefficientontheless‐educatedSpanish/English
hoursratiowhichnearlyoffsetsthecoefficientonpoorEnglish/Englishhours,indicatingthatan
inflowofless‐educatedSpanishspeakershaslittleoveralleffectonthewagesofnon‐Spanish
speakingimmigrants.ThisisconsistentwithSpanishspeakersworkingintheirown,parallellabor
marketwithinthearea.
IV.C. PuertoRico
19WhatIrefertoas“new”immigrantdestinationsintheSinger(2004)typographyincludeherthreecategoriesof“emerging,”“re‐emerging,”and“pre‐emerging.”Thepredictedvaluesarecomputedusingallthreevariablesincolumn(4).Anothermajorfactorinthewagedeterminationoflesseducatedworkersistheratioofcollegetonon‐collegeworkersinalabormarket(e.g.,Card,2009),butonthismeasure,newandolddestinationswereverysimilarin1990.TheactualaveragewageofSpanishspeakingimmigrantswas13percenthigherinnewthanolddestinationsin1990.20ThepatternsovertimearesimilarfornewrelativetotraditionalMexicandestinations.Idefinedas“traditional”Mexicandestinationsthetop20areasrankedonnumberofworkingageMexicansin1990.DrawingonCardandLewis(2007),Idefined15“new”MexicandestinationswithalargeresidualinflowrateofMexicansbetween1990and2008overwhatispredictedfromalinearregressionon1990Mexican(working‐agepopulation)share,andwhicharenottraditionalMexicandestinations.ThelargeareasamongthelatterareallnewdestinationsaccordingtotheSinger(2004)typography.EstimatesinTable6implythatMexicanswouldhaveearned3.5percentmoreinnewthanintraditionaldestinationsin1990,butby2008theywouldhaveearnedonepercentless.
22
AnotherreasontoexpectthataSpanish‐speakinglabormarketemergeswithasufficientdensityof
Spanishspeakers,isthatitmustinthelimitwhereeveryonespeaksSpanish.TheU.S.containsone
labormarketwherealmostallimmigrantsandnativesspeakSpanish:PuertoRico.21Accordingto
theviewthatlanguageskillsdriveimperfectsubstitutabilitywithnatives,PuertoRicanimmigrants
shouldbeperfectsubstitutesforPuertoRicannatives:theirrelativewagesshouldnotrespondat
alltoimmigrants’relativelaborsupply.Theapproachofexaminingamarketwhereimmigrants
andnativessharealanguagewaspreviouslytakenbyCastillo,GillessandRaphael(2009)(CGR),
whoexaminedCostaRica.
SincePuertoRicoisasinglelabormarket,theanalysiscannotexploitgeographicvariationinthe
sizeoftheimmigrantpopulation.Instead,IfollowtheapproachofCGRandOPandusevariation
overtimeandacrosseducation‐experiencecells,tobedetailedbelow,inthesizerelativesizeofthe
immigrantworkforce.
ThatPuertoRicohasanon‐trivialamountofimmigrationmaynotbewellknown.Figure4shows
theshareoftheworkingagepopulationthatwasbornoutsidePuertoRico–whatIdefineas
“immigrants”forthisanalysis‐‐between1970andthepresent,brokenoutseparatelybythefive
educationcategories‐‐fouryearscollege,1‐3yearscollege,highschoolcompletion,1‐3yearshigh
school,nohighschool‐‐usedintheanalysis.ItisbasedondatafromPuertoRicanpopulation
censuses,whicharetakeninparalleltotheU.S.Census,andthe2007‐9PuertoRicanCommunity
Surveys(Rugglesetal.,2010).MostofthosebornoutsidePuertoRicoarenottrulyimmigrantsbut
areotherU.S.citizens,mostlyofPuertoRicanheritage,whowereborninthecontinentalU.S.
Figure4showsuntilrecently,about15percentofcollegeeducatedworkerscamefromoutside
PuertoRico.Theless‐educated“immigrant”shareislowerbut,untilrecently,hadbeenrising.
21PuertoRicowasnotincludedintheearlieranalysis.In1990,whichisthelastyearthequestionwasasked,97percentofthosebornoutsideand99percentofthoseborninPuertoRicoindicatedthattheyspokeSpanish.Notethatthisisabroaderlanguagemeasurethanwasusedinthepreviousanalysis,whichonlyindicatedwhetherornottherespondentspokeSpanish“athome.”
23
TheanalysiswillexploitvariationacrossthecellsinFigure4,and,withinthese,acrossyearsof
potentialexperience,groupedintofiveyearcellsupto40.Usualweeklyhoursworkedisnot
availableinthe1970Census,sohourlywagesandaggregatehoursworkedwillbereplacedwith
weeklywagesandweeksworked.Inparticular,theestimationequationis:
(4) iktN
iktF
iktitktikNikt
Fikt WWWWww lnln
where Nikt
Fikt WWWWln representstheaggregaterelativeweeksworkedlastyearbynon‐Puerto
Ricanbornworkerswitheducationiandpotentialworkexperiencek.Estimatesof(4)controlfor
anexhaustivesetofeducationxexperience,experiencexyear,andeducationxyeardummies.22
Forcomparison,(4)willalsobeestimatedusingdataonthecontinentalU.S.,whichwillalsoserve
asanadditionalcheckonthecross‐cityresultspresentedabove.23
ResultsarepresentedinTable7,withPanelAforPuertoRicoandPanelBfortheContinentalU.S.
Column(4)showsanegativeandmarginallysignificantcoefficientof‐0.033,fortheU.S.,whichis
surprisinglysimilartotheestimatesinTable5giventhelargedifferenceinmethodology.The
estimateforPuertoRicoisnotzero,butpositiveandsignificant.Alikelyexplanationforthisthat
changesintheweeksworkedpartlyreflectslabordemand,whichmovesweeksworkedandweekly
wagesinthesamedirection.Toaddressthis,Iexploitthefactthatmuchofthevariationinthe
relativesizeoftheimmigrantworkforceinPuertoRicoisdrivenbychangesintherawnumbersof
22Allestimatesof(4)areunweighted,withstandarderrorscomputedtoberobusttoarbitraryerrorcorrelationwithineducationxexperiencecells.InthePuertoRicandata,weeksworkediscomputedforthoseaged16‐65whoarenotlivingingroupquartersandwhoareoldenoughtobeoutofschool.ThePuertoRicanwagesampleconsistsofthoseintheweekssamplewithpositivewageearnings,zerobusinessandfarmearnings,whoarenotcurrentlyenrolledinschoolandwhoreportanoccupation.23TheU.S.samplesaredefinedsimilarlytothePuertoRicanones(seepreviousnote),exceptforafewthings.First,tobeincludedinthewagesample,ratherthanreportinganoccupation,workersneedtoreportbeingcurrentlyemployed.(ThisdifferenceinmethodologyisduetothefactthatemploymentratesareverylowinPuertoRico.)Also,weeklywagesbelow10orabove10,000in1999dollarswereresettothesethresholds.Finally,thetwolowesteducationgroupsarecombinedinU.S.estimates,owingtotheverysmallnumberofnon‐highschooleducatednativesintheU.S.
24
U.S.‐bornethnicPuertoRicansandPuertoRican‐born.Asthesereflecteducationandfertility
decisionsmadelonginthepast,theyareunlikelytobesystematicallyrelatedtopresentdemand.24
ThesizeofthesetwopopulationsiscomputedusingthecombinationofPuertoRicanandU.S.data,
whichisnecessarybecauseonethirdofthePuertoRicanbornpopulationlivesintheU.S.When
relativeweeksworkedisreplacedbytherelativepopulationsize,incolumn(2),thecoefficientis
indeednonpositiveandisclosetozero.Column(3)showsthatthisvariablemovesalmostexactly
one‐for‐onewiththeweeksworkedvariable,arelationshipthatishighlysignificant.25
Therelationshipincolumn(2)isunfortunatelyverynoisy,andFigure5showswhy.Itshowsa
scatterplotunderlyingtherelationshipincolumn(2),aswellastheU.S.relationshipincolumn(4).
ThePuertoRicanwageestimatesaremuchnoisier,asshownbythelargemeansquarederror
(“vertical”variationaroundtheline)inthePuertoRicanfigurecomparedtotheU.S.,owingtothe
muchsmallernumberofobservationsinthePuertoRicandata.
TheremainingcolumnsofTable7examineseparatelytherelativewageresponse(inthe
continentalU.S.)ofimmigrantswithpoorandstrongEnglish,measuredinthesamewayasinPanel
BofTable5.26Thismeasureisnotavailableuntil1980,andsocolumn(5)showestimates
excluding1970,whicharesmallerinmagnitudeandlessprecise.BrokenoutseparatelybyEnglish
skills,incolumns(6)and(7),thenegativeresponseislimitedtoimmigrantswithpoorEnglish.
Thoughtheseestimatesarenotprecise,theyaresimilarinmagnitudetothedifferenceinresponse
betweenhigh‐andlow‐EnglishimmigrantsthatwasfoundinTable5,around3percentagepoints.
24AlthoughdemandconditionsinPuertoRicomayinfluencetheeducationalattainmentofthePuerto‐Ricanbornpopulation,thisinfluencemaybelimitedsinceasubstantialfraction(aboutonethird)ofPuertoRicansendupinthecontinentalU.S.labormarket.25Thisimpliesthataninstrumentalvariablesestimateoftheeffectofrelativeweeksworksonrelativeweeklywages,usingthepopulationvariableasaninstrument,wouldhavethesamecoefficientasappearsincolumn(2).26Thesecolumnsexcludenative‐bornworkerswhodonotreportspeakingEnglish“only”or“verywell.”
25
V. ImplicationsforPoverty
Whatcanwesayabouttheeffectsofimmigrationonpovertyinlightoftheseestimates?Afull
answertothisquestionrequiresestimatesoftheeffectofimmigrationonabroadersetofwage
outcomesthanwerestudiedinthischapter,butareexploredinotherresearch(Raphaeland
Smolensky,2008,2009).Chapter2explainsthisingreaterdetail,buttosimplify,themainthing
thepresentestimatesomitistheeffectimmigrationhasonwagesbyshiftingtheratioofnon‐
collegetocollegelabor.However,Card(2009)showsthisissmall(becausetheratioisrecently
similarforrecentimmigrantandnativeworkers).Ontopofthesepreviouslyestimatedeffects,the
estimatesinthischapterimplythatimmigrationhasalargerimpactonthewagesofimmigrants
withpoorEnglishlanguageskills,alreadyahighpovertygroup(Table2),anddependingon
immigrants’mixoflanguageskills,onSpanish‐speakers.
Furthermore,thewageresponsesofthesegroupscanberoughlytranslatedintoaneffecton
povertyrates,assumingallimmigrants’wagesareshifteddownbytheamountimpliedbythe
estimates.TheestimatesinbothTable7andTable5areconsistentwithaoneunitincreaseinthe
relativesupplyofimmigrantlabor‐‐again,roughlyequaltotheincreasesince1990‐‐loweringthe
wagesofimmigrantworkerswithpoorEnglishby3percentagepointsmorethanimmigrantswith
strongEnglish.Theresultingthreepercentdeclineinwages,assumingallimmigrants’incomeis
fromwages(whichisclosetotrue)would,withoutchanginghoursworked,beexpectedtodropthe
low‐Englishimmigrantsbetween100and103percentofthepovertylineintopoverty.In2008,
thisrepresentedaboutonepercentoflow‐Englishless‐educatedimmigrants.Ifyoualsoconsider
thefactthatthedeclineinwagesmightinducesomeimmigrantstoworkless,areasonable
approximationistodoublethisestimate.27Inshort,theriseinimmigrationsince1990mighthave
27Thiscomesfromassumingalaborsupplyelasticityofone(followingRaphaelandSmolensky,2009);thatistosay,thatathreepercentdeclineinhourlywagesisassociatedwithathreepercentdeclineinhoursworked,foratotalofasixpercentreductioninannuallaborincome.
26
addedonetotwopercentagepointstothepovertyrateofimmigrantswithpoorEnglish(compared
toimmigrantswithstrongEnglish).Usingtheestimatesincolumn(4)ofTable6,thechangeinthe
supplyofSpanish,English,andotherhoursbetween1990and2008isexpectedtohavelowered
thewagesofSpanishspeakingimmigrantsbyroughly6percent(relativetonatives),which
translatessimilarlyintoanincreaseintheirpovertyratesofSpanish‐speakingimmigrantsof
roughly2to4percentagepoints.28Inpractice,povertyratesamongless‐educatedimmigrants
declinedsince1990:from35to30percentamongnon‐Englishspeakersandfrom25to22percent
amongSpanishspeakers.Theestimatesinthispaperareconsistentwiththelargeimmigrantinflux
since1990attenuatingthisdecline.
VI. Conclusions
Onbalance,theestimatesinthispapersuggestthatwhileimmigrants’imperfectEnglishlanguage
skillsmaynotbetheonlyreasonthattheyareimperfectsubstitutesfornative‐bornworkersinthe
U.S.,theyareamajorreasonforthis.UsingseveraldifferentapproachesandsamplesIhavefound
thatthewagesofimmigrantswithpoorEnglishlanguageskillstendtorespondmorenegativelyto
agreaterpresenceofimmigrantsthandothewagesofimmigrantswithstrongEnglishlanguage
skills.Abottomlineisthat,atleastamongless‐educatedimmigrants,immigrationsince1990
mighthavepushedupthepovertyratesofimmigrantswithpoorEnglishbyonetotwopercentage
points(relativetoimmigrantswithstrongEnglish)andofSpanish‐speakingimmigrantbytwoto
fourpercentagepoints(relativetonatives).Theseimpactsareoverwhelmedbyadownwardtrend
inimmigrantpovertyrates,andaresmallerforSpanish‐speakingimmigrantsinmarketswhere
Spanishisprevalentinthecollege‐educatedworkforce.Theseimpactsmayalsohelpaccountfor
28Between1990and2008,ln(poor/strongEnglishhours)rose1.39,andln(Spanish/Englishhours)amongnon‐collegerose1.16andamongcollegeeducatedworkersrose0.58,fora‐0.061=‐0.033*1.39‐0.045*1.16+0.065*0.57changeintherelativelnwage,usingthecoefficientsfromTable6.
27
thenewspreadofMexicanimmigrantsto“new”immigrantdestinationsstartinginthe1990s,as
thelackoflabormarketcompetitionfromfellowSpanishspeakersinthosemarkets,comparedto
traditionalMexicandestinations,wouldhavemadethemarelativelyattractiveplacetosettle.29
29Chapter5includesadditionalanalysisofwhypovertyrateswerelowerin“new”immigrantdestinations.
28
ReferencesBerman,Eli,KevinLang,andErezSiniver.“Language‐SkillComplementarity:ReturnstoImmigrant
LanguageAcquisition.”LabourEconomics10(3):June2003,pp.265‐290.
Bleakley,andAimeeChin.“LanguageSkillsandEarnings:EvidencefromChildhoodImmigrants.”TheReviewofEconomicsandStatistics86(2):May2004,p.481–496.
Borjas,GeorgeJ.“TheEconomicsofImmigration.”JournalofEconomicLiterature32(4):December1994,p1667‐1717.
‐‐‐‐.“LaborOutflowsandLaborInflowsinPuertoRico.”JournalofHumanCapital2(1):Spring2008,pp.32‐68.
Card,David.“ImmigrantInflows,NativeOutflows,andtheLocalLaborMarketImpactsofHigherImmigration,”JournalofLaborEconomics19(1),January2001,pp.22‐64.
‐‐‐‐,“ImmigrationandInequality,”TheAmericanEconomicReview99(2),May2009,pp.1‐21.
Card,David,andJohnDiNardo,“DoImmigrantInflowsLeadtoNativeOutflows?”TheAmericanEconomicReview90(2),May2000,pp.360‐367.
Card,David,andEthanLewis.“TheDiffusionofMexicanImmigrantsDuringthe1990s:ExplanationsandImpacts.”InGeorgeJ.Borjas,ed.,MexicanImmigrationtotheUnitedStates.Chicago:UniversityofChicagoPress,2007.
Carliner,Geoffrey.“TheWagesandLanguageSkillsofU.S.Immigrants.”NBERWorkingPaper5763,September1996.
Castillo,Federico,J.KeithGilless,andStevenRaphael.“ComparingtheDomesticLaborMarketImpactsofaSouth‐NorthandSouth‐SouthMigration:TheCasesofCostaRicaandtheUnitedStates.”Mimeo,UCBerkeley,September2009.
Chiswick,BarryR.andPaulW.Miller.“TheEndogeneitybetweenLanguageandEarnings:InternationalAnalyses.”JournalofLaborEconomics13(2):April1995,pp.246‐288.
‐‐‐‐.“OccupationalLanguageRequirementsandtheValueofEnglishintheU.S.LaborMarket.”IZADPNo.2664,March2007.
Cutler,DavidM.,EdwardL.Glaeser,andJacobL.Vigdor.“WhenAreGhettosBad?LessonsfromImmigrantSegregationintheUnitedStates.”JournalofUrbanEconomics63(3):759‐74.
Dustmann,ChristianandArthurvanSoest.“LanguageFluencyandEarnings:EstimationwithMisclassifiedLanguageIndicators.”ReviewofEconomicsandStatistics83:2001,663‐674.
‐‐‐‐.“LanguageandtheEarningsofImmigrants.”IndustrialandLaborRelationsReview55(3):2002,pp.473‐492.
29
DustmannChristian,andFrancescaFabbri.“LanguageProficiencyandLabourmarketPerformanceofImmigrantsintheUK.”TheEconomicJournal113(489):July2003,pp.695‐717.
Ferrer,Ana,DavidA.GreenandCraigW.Riddell.“TheEffectofLiteracyonImmigrantEarnings.”TheJournalofHumanResources41:2006,380‐410.
Friedberg,RachelandJenniferHunt.“TheImpactofImmigrantsonHostCountryWages,EmploymentandGrowth.”JournalofEconomicPerspectives9(2):Spring1995,p.23‐44.
Goldin,ClaudiaandandLawrenceKatz.TheRaceBetweenEducationandTechnology,Cambridge,MA:HarvardUniversityPress,2008
Lang,Kevin.“ALanguageTheoryofDiscrimination.”TheQuarterlyJournalofEconomics101(2):May1986,pp.363‐382.
Lewis,Ethan,“HowDoLocalLaborMarketsintheU.S.AdjusttoImmigration?”FederalReserveBankofPhiladelphia,Mimeo,November2004.
Ottaviano,GianmarcoI.P.,andGiovanniPeri,“RethinkingtheEffectofImmigrationonWages,”Mimeo,UCDavis,March2010.AlsoforthcomingJEEA.
Peri,Giovanni,andChadSparber,“TaskSpecialization,ComparativeAdvantages,andtheEffectsofImmigrationonWages,”AmericanEconomicJournal:AppliedEconomics1(3),July2009,pp.135‐69.
Raphael,StevenandEugeneSmolensky.“ImmigrationandPovertyintheUnitedStates.”InstituteforResearchonPovertyDiscussionPaperno1347‐08,UniversityofWisconsin‐Madison,September2008.
‐‐‐‐.“ImmigrationandPovertyintheUnitedStates.”TheAmericanEconomicReview99(2),May2009,pp.41‐44.
Ruggles,Steven,J.TrentAlexander,KatieGenadek,RonaldGoeken,MatthewB.Schroeder,andMatthewSobek.IntegratedPublicUseMicrodataSeries:Version5.0[Machine‐readabledatabase].Minneapolis:UniversityofMinnesota,2010.
Singer,Audrey.“TheRiseofNewImmigrantGateways.”BrookingsInstitutionCenteronUrbanandMetropolitanPolicy,February2004.
30
.2.4
.6.8
1
0‐9 10‐19 20‐29 30‐39 40‐49 50‐59AgeatArrivalinU.S.
Only,VeryWell,orWell
OnlyorVeryWell
A.AgeatArrival
.2.4
.6.8
1
0‐4 5‐9 10‐1415‐1920‐2425‐2930‐40 40+YearsinU.S.
B.YearsinU.S.
DataSource:Combined2007,2008,and2009AmericanCommunitySurveys(Rugglesetal.,2010).Samplelimitedtoworkingageforeign‐bornlivingin136largemetropolitanareasandnotingroupquarters.
AllEducationLevels
Figure1a.ShareofImmigrantsWhoSpeakEnglish,byAgeAtArrivalandYearsinU.S.,2007‐9
31
.2.4
.6.8
1
0‐9 10‐19 20‐29 30‐39 40‐49 50‐59AgeatArrivalinU.S.
Only,VeryWell,orWell
OnlyorVeryWell
A.HighSchoolorLess
.2.4
.6.8
1
0‐9 10‐19 20‐29 30‐39 40‐49 50‐59AgeatArrivalinU.S.
B.MorethanHighSchool
DataSource:Combined2007,2008,and2009AmericanCommunitySurveys(Rugglesetal.,2010).Samplelimitedtoworkingageforeign‐bornlivingin136largemetropolitanareasandnotingroupquarters.
Figure1b.ShareofImmigrantsWhoSpeakEnglish,byAgeAtArrivalandBroadEducation,2007‐9
32
‐.15
‐.1‐.05
0
0‐9 10‐19 20‐29 30‐39 40‐49 50‐59AgeatArrivalinU.S.
SlopeEstimate
95%ConfidenceInterval
A.AgeatArrival
‐.15
‐.1‐.05
0
0‐4 5‐9 10‐1415‐1920‐2425‐2930‐40 40+YearsinU.S.
B.YearsinU.S.
DataSource:2007,2008,and2009AmericanCommunitySurveys(Rugglesetal.,2010)and2000CensusofPopulation.SeenotestoTable5fordetailsofdataconstructionandestimation.
AllEducationLevels
Figure2a.ResponseofImmigrantRelativeWagestoRelativeSupply,byAgeAtArrivalandYearsinU.S.,2000and2008
33
‐.2‐.15
‐.1‐.05
0
0‐9 10‐19 20‐29 30‐39 40‐49 50‐59AgeatArrivalinU.S.
SlopeEstimate
95%ConfidenceInterval
A.HighSchoolorLess
‐.2‐.15
‐.1‐.05
0
0‐9 10‐19 20‐29 30‐39 40‐49 50‐59AgeatArrivalinU.S.
B.MorethanHighSchool
DataSource:2007,2008,and2009AmericanCommunitySurveys(Rugglesetal.,2010)and2000CensusofPopulation.SeenotestoTable5fordetailsofdataconstructionandestimation.
Figure2b.ResponseofImmigrantRelativeWagestoRelativeSupply,byAgeAtArrivalandBroadEducation,2000and2008
34
BuffaloSt.LouisCleveland
Baltimore
Detroit
Minneapolis
Philadelphia Milwaukee
CharlotteGreensboro‐Winston‐Salem
Seattle
Tampa
Boston Atlanta
SaltLakeCity
Portland,OR
Nassau‐Suffolk
Raleigh‐Durham
WashingtonDCSacramento
FortLauderdale
Middlesex
WestPalmBeach
Orlando
Newark
Denver
Bergen‐Passaic
FortWorth
LasVegasChicago
NewYork
Oakland
Phoenix
SanDiego
SanFrancisco
Austin
Houston
Riverside
Dallas
SanJose
JerseyCity
Anaheim
LosAngeles
Miami
‐5‐4
‐3‐2
‐10
AmongWorkersWithMorethanHS
‐4 ‐3 ‐2 ‐1 0AmongWorkerswithHSorless
DataSource:2007‐9AmericanCommunitySurveys.Dottedlinesaremedians.PointsaboveslopedlinehaveaboveaveragewagesforSpanish‐speakingimmigrants,relativetonatives,accordingtoestimatesinTable6,column4.
Figure3.ln(Spanish/English)Hours,2008:morethanhighschoolvs.highschoolorless
35
NoHS
1‐3yrsHS
HSDegree
1‐3yrsColl
4+yrscoll
.05
.1.15
.2.25
1970 1980 1990 2000 2010
Datasource:1970‐2000U.S.Censuses,andcombined2007‐9PuertoRicanCommunitySurveys(Rugglesetal.,2010).Samplelimitedtopopulationaged16‐65,notlivingingroupquartersandoldenoughtobeoutofinschoolwithnormalprogression.
Figure4.Non‐PuertoRicanbornshareofPuertoRicanpopulation,byeducationandyear
36
SmHS,0
HS,0
SmC,0
CGRD,0
NoHS,5
SmHS,5
HS,5
SmC,5
CGRD,5NoHS,10
SmHS,10
HS,10
SmC,10
CGRD,10NoHS,15
SmHS,15
HS,15
SmC,15
CGRD,15
NoHS,20
SmHS,20
HS,20
SmC,20
CGRD,20
NoHS,25
SmHS,25
HS,25
SmC,25
CGRD,25
NoHS,30
SmHS,30
HS,30
SmC,30
CGRD,30
NoHS,35
SmHS,35
HS,35
SmC,35
CGRD,35
NoHS,40
SmHS,40
HS,40
SmC,40
CGRD,40
NoHS,0
SmHS,0
HS,0SmC,0
CGRD,0
NoHS,5
SmHS,5
HS,5SmC,5
CGRD,5
NoHS,10
SmHS,10HS,10SmC,10
CGRD,10
NoHS,15
SmHS,15
HS,15
SmC,15
CGRD,15
NoHS,20
SmHS,20
HS,20
SmC,20
CGRD,20
NoHS,25
SmHS,25
HS,25
SmC,25
CGRD,25
NoHS,30
SmHS,30
HS,30
SmC,30
CGRD,30
NoHS,35
SmHS,35
HS,35
SmC,35
CGRD,35
NoHS,40
SmHS,40
HS,40
SmC,40
CGRD,40
NoHS,0
SmHS,0
HS,0
SmC,0
CGRD,0
NoHS,5
SmHS,5
HS,5SmC,5
CGRD,5
NoHS,10
SmHS,10
HS,10
SmC,10
CGRD,10
NoHS,15
SmHS,15
HS,15
SmC,15
CGRD,15NoHS,20
SmHS,20
HS,20
SmC,20
CGRD,20
NoHS,25SmHS,25
HS,25
SmC,25
CGRD,25
NoHS,30
SmHS,30
HS,30
SmC,30
CGRD,30NoHS,35
SmHS,35
HS,35
SmC,35
CGRD,35NoHS,40
SmHS,40
HS,40
SmC,40
CGRD,40
NoHS,0
SmHS,0
HS,0
SmC,0
CGRD,0
NoHS,5
SmHS,5
HS,5
SmC,5
CGRD,5
NoHS,10
SmHS,10
HS,10
SmC,10
CGRD,10
NoHS,15SmHS,15
HS,15
SmC,15
CGRD,15
NoHS,20
SmHS,20
HS,20
SmC,20CGRD,20
NoHS,25
SmHS,25HS,25
SmC,25
CGRD,25
NoHS,30
SmHS,30HS,30
SmC,30
CGRD,30
NoHS,35SmHS,35
HS,35
SmC,35CGRD,35
NoHS,40
SmHS,40 HS,40
SmC,40
CGRD,40
NoHS,0
SmHS,0
HS,0
SmC,0
CGRD,0
NoHS,5
SmHS,5
HS,5
SmC,5
CGRD,5NoHS,10
SmHS,10
HS,10
SmC,10
CGRD,10NoHS,15
SmHS,15
HS,15SmC,15
CGRD,15
NoHS,20 SmHS,20
HS,20
SmC,20
CGRD,20
NoHS,25
SmHS,25
HS,25
SmC,25CGRD,25
NoHS,30
SmHS,30
HS,30 SmC,30
CGRD,30NoHS,35
SmHS,35
HS,35SmC,35
CGRD,35
NoHS,40
SmHS,40
HS,40
SmC,40
CGRD,40
SlopeCoef(se)=‐.002(.126)
NoHS‐<9yrsedSmHS‐9‐11yrsedHS‐12yrsedSmC‐13‐15yrsedCGRD‐>=16yrsed
‐.4‐.2
0.2
.4Residualln(Imm/NativeWeeklyWage)
‐.4 ‐.2 0 .2 .4 .6Residualln(PREthnics/PRNatives)
A.PuertoRico
HS,0
SmC,0
CGRD,0
HSDO,5HS,5
SmC,5
CGRD,5
HSDO,10HS,10
SmC,10CGRD,10
HSDO,15HS,15
SmC,15CGRD,15
HSDO,20
HS,20
SmC,20
CGRD,20
HSDO,25HS,25SmC,25
CGRD,25
HSDO,30
HS,30
SmC,30
CGRD,30
HSDO,35 HS,35SmC,35
CGRD,35HSDO,40
HS,40
SmC,40
CGRD,40
HSDO,0HS,0
SmC,0CGRD,0
HSDO,5
HS,5 SmC,5
CGRD,5
HSDO,10
HS,10SmC,10 CGRD,10
HSDO,15HS,15
SmC,15CGRD,15HSDO,20 HS,20SmC,20 CGRD,20
HSDO,25HS,25
SmC,25
CGRD,25HSDO,30
HS,30SmC,30CGRD,30
HSDO,35HS,35
SmC,35
CGRD,35
HSDO,40
HS,40SmC,40CGRD,40
HSDO,0HS,0
SmC,0
CGRD,0
HSDO,5
HS,5SmC,5
CGRD,5
HSDO,10
HS,10SmC,10
CGRD,10HSDO,15
HS,15SmC,15CGRD,15
HSDO,20
HS,20
SmC,20CGRD,20HSDO,25HS,25
SmC,25
CGRD,25
HSDO,30
HS,30SmC,30
CGRD,30HSDO,35
HS,35
SmC,35
CGRD,35HSDO,40
HS,40
SmC,40
CGRD,40HS,0SmC,0
CGRD,0
HSDO,5
HS,5SmC,5
CGRD,5
HSDO,10HS,10
SmC,10
CGRD,10
HSDO,15HS,15
SmC,15
CGRD,15
HSDO,20
HS,20
SmC,20
CGRD,20
HSDO,25HS,25SmC,25
CGRD,25HSDO,30
HS,30SmC,30
CGRD,30
HSDO,35HS,35
SmC,35CGRD,35
HSDO,40
HS,40
SmC,40CGRD,40
HS,0
SmC,0
CGRD,0HSDO,5
HS,5SmC,5
CGRD,5HSDO,10
HS,10
SmC,10
CGRD,10
HSDO,15
HS,15
SmC,15
CGRD,15
HSDO,20
HS,20SmC,20
CGRD,20
HSDO,25HS,25
SmC,25
CGRD,25
HSDO,30
HS,30
SmC,30
CGRD,30HS,35
SmC,35
CGRD,35
HS,40
SmC,40 CGRD,40
SlopeCoef(se)=‐.033(.017)
HSDO‐<12yrsedHS‐12yrsedSmC‐13‐15yrsedCGRD‐>=16yrsed‐.2
‐.10
.1.2
‐.3 ‐.2 ‐.1 0 .1 .2 .3Residualln(Imm/NativeWksWorked)
B.UnitedStates
Note:Scalesofthetwographsare(roughly)proportional.Labelsidentifyeducationxfiveyearexperiencescells.Thex‐variableinPanelAistheresidualln(ethnicPuertoRicanUS‐born/PuertoRican‐bornpopulation).(Aoneunitincreaseinthisvariableisassociatedwithaoneunitincreaseintherelativeweeksworkedofnon‐Puerto‐Ricanbornworkers‐‐seeTable7).Allvariablesareresidualsfromaregressiononanexhaustivesetofeducationxyear,experiencexyear,andeducationxexperiencedummies.RawdatasourcesareU.S.andPuertoRicanPopulationCensuses,andAmericanandPuertoRicanCommunitySurveys(Rugglesetal.,2010).Seetextfordetails.
Figure5.Responseofimmigrantrelativeweeklywagetoimmigrantrelativesupply:U.S.andPuertoRico
37
Subgroup:
Education Level: All Educ.
Levels
High School
or Less
More than
High School
All Educ.
Levels
High School
or Less
More than
High School
(1) (2) (3) (4) (5) (6)
Foreign‐Born, Share Speaks English…
Only/Very Well 0.464 0.278 0.679 0.261 0.177 0.516
Well 0.221 0.227 0.214 0.234 0.227 0.257
Not Well 0.211 0.318 0.088 0.317 0.365 0.173
Not At All 0.104 0.178 0.019 0.188 0.232 0.054
Native‐Born, Share Speaks English…
Only/Very Well 0.983 0.971 0.990 0.820 0.764 0.879
Well 0.011 0.017 0.007 0.113 0.139 0.086
Not Well 0.005 0.009 0.003 0.054 0.074 0.033
Not At All 0.001 0.003 0.000 0.013 0.023 0.002
B. Speaks Spanish At Home
Data Source: Combined 2007, 2008, and 2009 American Community Surveys (Ruggles et al. 2010). Sample limited to working‐age
population (age 16‐65 with positive years of potential work experience) residing in 136 large metropolitan areas and not in group
quarters. Panel B is further restricted to those who report speaking Spanish at home. Sample weights used to construct shares.
Table 1. Distribution Across English Speaking Categories, by Nativity, Education and Home Language
A. All Working Age
38
Outcome:
Education Level: All Educ.
Levels
High School
or Less
More than
High School
All Educ.
Levels
High School
or Less
More than
High School
(1) (2) (3) (4) (5) (6)
All Native‐Born 2.25 1.94 2.39 0.10 0.16 0.06
All Foreign‐Born 2.05 1.76 2.35 0.15 0.20 0.08Foreign‐Born Who Speak English:
Only/Very Well 2.31 1.91 2.46 0.09 0.14 0.06
Well 1.99 1.83 2.19 0.13 0.16 0.10
Not Well 1.70 1.67 1.87 0.21 0.22 0.17
Not at All 1.55 1.54 1.71 0.29 0.30 0.23
All FB Spanish Speakers 1.80 1.70 2.07 0.19 0.22 0.11 Speaks no English 1.54 1.53 1.66 0.30 0.30 0.24
B. Share of Group in PovertyA. Mean ln(hourly Wage), 1999$
Table 2. Mean ln Wages and Poverty Rates By Nativity, Education, and Language Skills, 2008
Data Source: Combined 2007, 2008, and 2009 American Community Surveys (Ruggles et al. 2010). In Panel B, sample is limited to working‐
age population (age 16 ‐ 65 and old enough to be out of school, given normal progression) residing in 136 large metropolitan areas and not in
group quarters. In Panel A, sample is limited to respondents from the Panel B sample that are currently employed and that had positive
hours worked, positive wage and salary earnings, and zero self‐employment and farm earnings in the past year. Wages are adjusted to 1999
dollars using the consumer price index, and wages exceeding 200 dollars and less than 2 dollars in 1999 dollars are replaced with these
thresholds.
Foreign‐Born
Native Born
Foreign‐Born who Speak Spanish At Home
39
Dependent Variable:
Subgroup:
(1) (2) (3) (4) (5) (6)
Immigrant ‐0.186 ‐0.041 ‐0.021 0.000 ‐0.003 ‐0.004
(0.009) (0.007) (0.007) (0.013) (0.013) (0.013)
Speaks English:
0.209 0.168 0.149 0.208 0.174
(0.008) (0.009) (0.008) (0.021) (0.019)
Among entire working‐age pop ‐0.212
(0.081)
0.069
(0.081)
‐0.306
(0.165)
Sample Size 724,737 724,737 724,737 173,590 173,590 173,590
R2
0.019 0.028 0.189 0.016 0.018 0.018
Other Controls?* No No Yes No No No
ln(Hourly Wage), Workers with HS Education or Less
Table 3: The Role of Language Skills in Immigrant‐Native Wage Gaps.
Spanish Speakers Only
Data Source: Combined 2007, 2008, and 2009 American Community Surveys (Ruggles et al., 2010). Sample limited to working‐age
respondents (age 16 ‐ 65 with positive years of potential work experience) that have 12 or fewer years of education, that reside one of
136 large metropolitan areas and not in group quarters, that are currently employed, and that had positive hours worked, positive wage
and salary earnings, and zero self‐employment and farm earnings in the past year. Wages exceeding 200 dollars or less than 2 dollars in
1999 dollars are replaced with these thresholds. Standard errors, in parentheses, computed to be robust to arbitrary error correlation
within metropolitan area. *Other controls are a quartic in potential work experience; years of education; years of interacted with
education below 9 years, born after 1950, and both; and dummies for education less than 9 years, born after 1950, female, black,
Hispanic, female*black, and female*hispanic.
Among those with HS or less
Among those with more than HS
Only or Very Well
Only or Very Well x Share of MSA who speak Spanish at home
40
Mean Standard Mean Standard Mean Standard
Deviation Deviation Deviation
(1) (2) (3) (4) (5) (6)
ln(imm hours/nat hours) ‐1.934 1.019 ‐1.581 1.113 ‐2.288 0.769
Immigrant‐Native Wage Gap
All Immigrants ‐0.125 0.079 ‐0.156 0.072 ‐0.093 0.073
High English Immigrants ‐0.076 0.066 ‐0.088 0.064 ‐0.063 0.065
Low English immigrants ‐0.360 0.201 ‐0.251 0.105 ‐0.469 0.215
Observations
Table 4. Regression Data Descriptive Statistics
272272
Data source: Combined 2007, 2008, and 2009 American Community Surveys (Ruggles et al ., 2010) and 2000 Census of
Population. Sample for constructing hours worked includes al l those aged 16‐65 who are old enough to be out of school
(given normal progress ion) and res iding in one of the 136 metropol i tan areas in the sample and not in group quarters . To be
in the wage sample requires being in the hours worked sample plus being employed, with pos i tive weeks and hours worked
las t year, nonzero wage and sa lary earnings , and zero bus iness and farm earnings ; for natives i t a lso requires speaking
Engl i sh "only" or "very wel l ." Hourly wages above $200 and below $2 in 1999 dol lars are reset to these thresholds . Data have
been aggregated to 136 metropol i tan areas x two education groups x two years (2000 and "2008," combining the three ACSs).
Table shows unweighted means and standard deviations .
544
All Education Levels More than High SchoolHigh School or Less
41
Education Levels: All
High School
or Less
More than
HS
(1) (2) (3)
A. All Immigrants ‐0.040 ‐0.034 ‐0.054
(0.003) (0.004) (0.006)
B. By Broad English Language Skills:
Speaks English Only, ‐0.022 ‐0.011 ‐0.045
Very Well, or Well (0.004) (0.004) (0.006)
Speaks English Not Well ‐0.057 ‐0.047 ‐0.078
or Not At All (0.006) (0.006) (0.016)
P‐value, Equal Coefficients 0.000 0.000 0.030
C. By Detailed English Language Skills (112 MSAs):
Speaks English Only or ‐0.020 ‐0.012 ‐0.036
Very Well (0.004) (0.004) (0.007)
Speaks English Well ‐0.028 ‐0.018 ‐0.047
(0.005) (0.006) (0.011)
Speaks English Not Well ‐0.049 ‐0.037 ‐0.071
(0.006) (0.007) (0.013)
Speaks English Not at All ‐0.083 ‐0.050 ‐0.144
(0.014) (0.008) (0.035)
P‐value, Equal Coefficients 0.000 0.000 0.005
Data Source: Combined 2007, 2008, and 2009 American Community Surveys (Ruggles et
al., 2010) and 2000 Census of Population. Table shows coefficient estimates from
regressions of the wage gap between specified immigrants and "similar" natives (see
below) on the natural log of the ratio of aggregate hours worked of immigrants and
natives, using variation across metropolitan areas, year (2000 or "2008") and the two
broad education of columns (2) and (3). All regressions control for year by education
effects. Standard errors, in parentheses, computed to be robust to arbitrary error
correlation within metropolitan area. Sample for constructing mean wages limited to
working‐age respondents (age 16 ‐ 65 and old enough to be out of school given normal
progression) , that reside one of 136 large metropolitan areas (112 in Panel C) and not in
group quarters, that are currently employed, and that had positive hours and weeks
worked, positive wage and salary earnings, and zero business and farm earnings in the
past year; for natives, sample is further limited to those who report speaking English
"only" or "very well." Hourly wages above $200 and below $2 in 1999 dollars were reset
to these thresholds. *The mean ln hourly wage of "similar" natives is computed by
weighting natives to have the same distribution across potential experience (in five year
bands) x education (four groups: high school dropouts, high school, some college, and at
least four years college) cells as the specified group of immigrants in the metropolitan
area and year.
Table 5: Response of the Difference in the Mean ln(Hourly Wage)
of Immigrants and Similar* Natives to Changes in Immigrant
Relative Aggregate Hours Worked
42
Dependent Variable:
Subgroup of Immigrants: Other
(1) (2) (3) (4) (5)
Ln (Immigrant Hours/ ‐0.034
Native‐Born Hours) (0.004)
‐0.034 ‐0.036 ‐0.033 ‐0.060
Hours) (0.004) (0.007) (0.028) (0.018)
ln(Spanish‐speaking Hours/Strong English Speaking Hours):
‐0.045 0.052
Education or Less (0.035) (0.024)
0.065 ‐0.015
High School Education (0.017) (0.017)
Metro x Year Observations 272 272 272 272 272
R2
0.300 0.298 0.157 0.224 0.110
Data Source: Combined 2007, 2008, and 2009 American Community Surveys (Ruggles et al., 2010) and 2000 Census of
Population. Wage sample limited to working‐age respondents (age 16 ‐ 65 with and old enough to be out of school, given
normal progression) that have 12 or fewer years of education (or a GED), that reside one of 136 large metropolitan areas and
not in group quarters, that are currently employed, and that had positive hours worked, positive wage and salary earnings,
and zero self‐employment earnings in the past year; for natives it also requires speaking English "only" or "very well." The
dependent variable is the difference in the mean ln hourly wage between the specified group of immigrants and similar
natives, where the mean ln hourly wage of "similar" natives is computed by weighting natives to have the same distribution
across potential experience (in five year bands) x education (high school dropouts or completers) cells as the specified group
of immigrants in the metropolitan area and year. Strong English‐speaking hours worked is the sum of hours worked by those
who report speaking English "Only" "Very Well," or "Well," while poor English‐speaking hours are the sum of hours worked
reported by those who speak English "Not Well" or "Not at All" among working‐age respondents. "Spanish‐speakers" are
respondents who report speaking Spanish at home. All regressions are unweighted and control for year effects. Standard
errors, in parentheses, computed to be robust to arbitrary error correlation within metropolitan area.
All Spanish‐Speaking
ln(Immigrant Wage/Native Wage), High School or Less
Table 6: Immigrants' Relative Wages and Language Supplies, 2000 and 2008
Ln(Poor/Strong English‐Speaking
Among workers with High School
Among workers with More than
43
Y = ln(for/ Subsample:
nat weeks) Poor English
(1) (2) (3) (4) (5) (6) (7)
ln(Foreign‐born weeks/ 0.150 ‐0.033 ‐0.021 ‐0.002 ‐0.031
Native‐born weeks) (0.061) (0.017) (0.019) (0.020) (0.049)
ln(Cont'l U.S. born ethnic Puerto ‐0.002 1.003
Ricans/Puerto‐Rican born) (0.126) (0.126)
Observations 224 224 224 180 144 144 144
R‐squared 0.489 0.457 0.919 0.935 0.942 0.913 0.976
Table 7: Continental U.S. and Puerto Rico: Aggregate estimates, 1970‐2000
Y=ln(foreign/native
Data Source: 1970, 1980, 1990, and 2000 Puerto Rican Censuses of Population and combined 2007, 2008, and 2009 Puerto Rican Community Surveys
(Panel A) and 1970, 1980, 1990, and 2000 U.S. Censuses of Population and combined 2007, 2008, and 2009 American Community Surveys (Panel B),
al l from Ruggles et al . (2010). Sample for independent variable includes workers age 16‐65 who are old enough to be out of school given
normal progress ion through school and not l i ving in group quarters . The sample used to compute the dependent variable, weekly wages , i s
this sample with the additional requirement of being currently employed (U.S.) or reporting an occupation (Puerto Rico), not enrol led in school ,
and without bus iness or farm income. In U.S. data (columns 4‐7), weekly wages exceeding $10,000 or below $10 in 1999 dol lars were reset at
these thresholds . In columns 5‐7, native‐born workers who did no report speaking Engl i sh "only" or "very wel l" were excluded from the wage
sample. Sample weights used to aggregate variables to the 5‐year experience x education cel l s used in the analys i s (see text). Standard errors
are ca lculated to be robust to arbi trary error correlation within education x experience cel l s .
weekly wage)
A. Puerto Rico B. Continental United States
Excluding 1970
Strong English
44
Appendix
Toaddressthepossibilitythatahighpresenceofimmigrantsinamarketreflectshighwagesin
thosemarkets‐‐whichwouldtendtobiasthecoefficientsinTables5and6towardszero,Iusean
instrumentalvariableforln(HF/HN)fromCard(2009),whichmeasurestheimmigrantinflowrates
predictedbythearea’slaggedoriginmixofimmigrants.Inparticular,itis:
1,
2,
tc
o
Fojttoc
jct P
Pfz ,
wherefoc,t‐2representsatwodecadelaginthefractionofU.S.immigrantsfromregionolivingin
metropolitanareac,whichapportions FojtP ,thenumberofimmigrants(nationally)arrivingfrom
oinskillgroupjinthepastdecade.(TheregionsarelistedinTableA2,alongwith FojtP figures.)
Thenumeratoristhusthepredictednumberofimmigrantarrivalsincelljc.Thisisconvertedtoa
predictedarrivalratebydividingbythearea’sbeginningofdecadepopulation,Pc,t‐1.The
assumptionbehindthisinstrumentisimmigrantspersistinlocatingincertainareasbecausethey
valuebeingnearsimilarimmigrants,andnotbecausetheseareashavepersistentlystrongerwage
growthforthattypeofimmigrant.
Thisinstrumentisastrongpredictorofimmigrants’relativehours.F‐statisticsontheinstrument
inthefirststageareinthe50‐100range.Inaddition,instrumentalvariablesestimatesusingthis
instrumentaresimilartotheOLSestimatespresentedinTable5.ThisisshowninAppendixTable
A3,whichisidenticalinstructuretoTable5butshowsIVestimates.Itshowsthesamepatternof
coefficients,withgreater(magnitude)wageresponsesforimmigrantswithworseEnglishlanguage
skills.ThedifferencesincoefficientsacrossEnglish‐speakingcategoriesaresimilarinmagnitudeto
theOLSestimates,andare,asinOLS,statisticallysignificant.
45
Seattle
Boston
St.Louis
SanFranciscoMinneapolis
NewYork
Atlanta
Portland,OR
Baltimore
Detroit
SanJose
SacramentoPhiladelphia
Newark
OaklandCharlotte
WestPalmBeach
Cleveland
Greensboro‐Winston‐Salem
Raleigh‐Durham
WashingtonDCChicago
Anaheim
FortLauderdale
Middlesex
Denver
Buffalo
Phoenix
Dallas
LosAngeles
FortWorth
Milwaukee
Miami
Bergen‐Passaic
Houston
SaltLakeCity
JerseyCity
SanDiegoLasVegasAustin
Nassau‐Suffolk
TampaRiverside
Orlando
‐1‐.5
0.5
11.5
AmongWorkersWithMorethanHS
‐1 ‐.5 0 .5 1AmongWorkerswithHSorless
DataSource:2007‐9AmericanCommunitySurveys.Pointsareresidualsofaregressionofln(Spanish/Englishhours)onln(poor/strongEnglishhours),separatelybyeducation(highschoolorlessonthex‐axis,morethanHSonthey‐axis).PointsaboveslopedlinehaveaboveaveragewagesforSpanish‐speakingimmigrants,relativetonatives,(conditionalonln(poor/strongEnglishhours)accordingtoestimatesinTable6,column4.
AppendixFigureA1.Residualln(Spanish/EnglishHours),2008:morethanhighschoolvs.highschoolorless
46
Area
High School
or Less
More than
High School Area
High School
or Less
More than
High School
Anaheim, CA 0.543 0.118 McAllen, TX 0.928 0.789
Aurora, IL 0.502 0.097 Miami, FL 0.684 0.612
Bakersfield, CA 0.507 0.198 Oxnard‐Ventura, CA 0.528 0.147
Brownsville, TX 0.816 0.670 Riverside, CA 0.505 0.213
El Paso, TX 0.846 0.678 Salinas, CA 0.673 0.193
Jersey City, NJ 0.526 0.270 San Antonio, TX 0.540 0.292
Laredo, TX 0.890 0.818 Santa Barbara, CA 0.589 0.151
Los Angeles, CA 0.633 0.216 Santa Cruz, CA 0.515 0.093
Table A1. Share Speaks Spanish at Home, by Broad Education, Selected Metropolitan Areas
Data source: s tacked 2007, 2008 and 2009 American Community Surveys (ACSs) (Ruggles et al ., 2010). Sample l imited
to working‐age population (age 16‐65 and old enough to be out of school , given a normal progress ion through
school ), and not l i ving in group quarters . Computed us ing ACS sample weights .
47
Origin Group:
High School
or Less
More than
High School
High School
or Less
More than
High School
(1) (2) (3) (4)
Mexican 1,944,656 292,542 2,618,328 296,963
Central American 517,066 105,261 493,669 92,671
South American 290,534 302,840 333,430 275,063
Caribbean (ex Cuban) 202,625 131,153 331,827 148,237
SE Asian (ex Filipino) 139,257 104,244 288,013 173,567
Chinese 135,836 220,608 158,375 302,729
Russian or E European 133,065 286,665 283,883 385,346
Sub‐Saharan African 129,245 178,315 129,346 173,449
South Asian 123,072 497,999 148,698 430,311
Cuban 89,306 56,648 109,769 56,659
Middle Eastern (ex Israeli) 88,988 165,310 94,684 137,885
Filipino 56,810 229,456 91,406 219,320
Commonwealth 51,432 189,733 74,478 264,485
Korean or Japanese 50,217 220,028 84,669 248,958
Southern European 27,374 46,875 34,168 48,243
Northern European* 9,521 63,872 55,668 169,033
2000‐2008 1990‐2000
Table A2. National Immigrant Arrivals, by Education, Origin, and Decade
* Includes Is rael i s . Data Source: Combined 2007, 2008, 2009 American Community Surveys
(columns (1) and (2), from Ruggles et al ., 2010) and 2000 Census of Population (columns (3) and
(4)). Sample l imited to working age population (age 16‐65 and old enough to be out of school ,
given normal progress ion through school ) and not l i ving in group quarters .
48
Education Levels: All
High School
or Less
More than
HS
(1) (2) (3)
A. All Immigrants ‐0.035 ‐0.030 ‐0.052
(0.005) (0.006) (0.007)
B. By Broad English Language Skills:
Speaks English Only, ‐0.018 ‐0.010 ‐0.044
Very Well, or Well (0.005) (0.005) (0.006)
Speaks English Not Well ‐0.043 ‐0.033 ‐0.074
or Not At All (0.009) (0.009) (0.016)
P‐value, Equal Coefficients 0.000 0.001 0.060
C. By Detailed English Language Skills (112 MSAs):
Speaks English Only or ‐0.022 ‐0.018 ‐0.036
Very Well (0.005) (0.005) (0.007)
Speaks English Well ‐0.015 ‐0.005 ‐0.046
(0.007) (0.008) (0.011)
Speaks English Not Well ‐0.030 ‐0.017 ‐0.069
(0.010) (0.011) (0.015)
Speaks English Not at All ‐0.057 ‐0.030 ‐0.137
(0.014) (0.013) (0.029)
P‐value, Equal Coefficients 0.000 0.009 0.001
Table A3: Instrumental Variables Estimates of the Response of the
Difference in the Mean ln(Hourly Wage) of Immigrants and Similar*
Natives to Changes in Immigrant Relative Aggregate Hours Worked
Data Source: Combined 2007, 2008, and 2009 American Community Surveys (Ruggles et
al., 2010) and 2000 Census of Population. Table shows coefficient estimates from
regressions of the wage gap between specified immigrants and "similar" natives (see
below) on the natural log of the ratio of aggregate hours worked of immigrants and
natives, using variation across metropolitan areas, year (2000 or "2008") and the two
broad education of columns (2) and (3). All regressions control for year by education
effects and are estimated by instrumental variables using the lagged origin mix
instrument described in the appendix. Standard errors, in parentheses, computed to be
robust to arbitrary error correlation within metropolitan area. Sample for constructing
mean wages limited to working‐age respondents (age 16 ‐ 65 and old enough to be out of
school given normal progression), that reside one of 136 large metropolitan areas (112
in Panel C) and not living in group quarters, that are currently employed, and that had
positive hours and weeks worked, positive wage and salary earnings, and zero business
and farm earnings in the past year; for natives, sample is further limited to those who
report speaking English "only" or "very well." Hourly wages above $200 and below $2 in
1999 dollars were reset to these thresholds. *The mean ln hourly wage of "similar"
natives is computed by weighting natives to have the same distribution across potential
experience (in five year bands) x education (four groups: high school dropouts, high
school, some college, and at least four years college) cells as the specified group of
immigrants in the metropolitan area and year.