Explaining Job Polarization

AER文章

ExplainingJobPolarization:Routine-BiasedTechnologicalChangeandOffshoring

ByMAARTENGOOS,ALANMANNINGANDANNASALOMONS Thispaperdocumentsthepervasivenessofjobpolarizationin16West-ernEuropeancountriesovertheperiod1993-2010.Itthendevelopsandestimatesaframeworktoexplainjobpolarizationusingroutine-biasedtechnologicalchangeandoffshoring.Thismodelcanexplainmuchofbothtotaljobpolarizationandthesplitintowithin-industryandbetween-industrycomponents(JEL:J21,J23,J24)

The‘Skill-BiasedTechnologicalChange’hypothesis(SBTC)–seeAutorandKatz(1999),GoldinandKatz(2008,2009)andAcemogluandAutor(2011)forexcellentoverviews–arosefromtheobservationthatdemandisshiftinginfavorofmoreed-ucatedworkers.Inspiteofitssuccessinexplainingmanydecadesofdata,however,SBTCcannotexplaintherecentphenomenonofjobpolarizationasdocumentedbyAu-tor,KatzandKearney(2006,2008)andAutorandDorn(2013)fortheUSandGoosandManning(2007)fortheUK.JobpolarizationhasalsobeendocumentedforGermany(Spitz-Oener2006;Dustmann,LudsteckandSchönberg2009)andthereareindicationsitispervasiveinEuropeancountries(Goos,ManningandSalomons2009;Michaels,NatrajandVanReenen2013).The rstcontributionofthispaperistoshowthatjobpolarizationispervasiveacrossadvancedeconomiesbyshowingitholdsin16Western-Europeancountries.

Themainhypothesesputforwardtoexplainjobpolarizationarethatrecenttechnologi-calchangeisbiasedtowardsreplacinglaborinroutinetasks(whatwecallroutine-biasedtechnologicalchange,RBTC)andthatthereistaskoffshoring(itselfpartiallyin uencedbytechnologicalchange),andthatbothoftheseforcesdecreasethedemandformiddlingrelativetohigh-skilledandlow-skilledoccupations(Autor,LevyandMurnane2003;Au-tor,KatzandKearney2006,2008;GoosandManning2007;AutorandDorn2013).Thesecondcontributionofthispaperistodevelopandestimateamodel–thathasitsrootsinthecanonicalmodel rstdevelopedbyKatzandMurphy(1992)fortheanalysisofSBTC–toquantifytheimportanceofRBTCandoffshoringinexplainingjobpolarization.OurestimatessuggestthatRBTCismuchmoreimportantthanoffshoring.Weshowthatthismodelexplainsnotjustoveralljobpolarizationbutalsoitswithin-industryandbetween-industrycomponentsthatarebothempiricallyimportant.Withineachindustrythereis

DepartmentofEconomics,UniversityofLeuven,Naamsestraat69,3000Leuven,Belgium,

maarten.goos@kuleuven.be.Manning:CentreforEconomicPerformanceandDepartmentofEconomics,LondonSchoolofEconomics,HoughtonStreet,LondonWC2A2AE,UnitedKingdom,a.manning@lse.ac.uk.Salomons:UtrechtUni-versitySchoolofEconomics,AdamSmithHall,InternationalCampusUtrecht,Kriekenpitplein21-22,3584ECUtrecht,TheNetherlands,a.m.salomons@uu.nl.Acknowledgements:WethankDavidAutor,LarryKatz,ThomasLemieux,StephenMachin,GuyMichaels,JohnVanReenen,UlrichZierahn,theeditor,refereesandnumerousseminarpartici-pantsforexcellentsuggestions.Theauthorsdeclarethattheyhavenorelevantormaterial nancialintereststhatrelatetotheresearchdescribedinthispaper.

1

Goos:

AER文章

2THEAMERICANECONOMICREVIEWMONTHYEAR

ashiftawayfromroutineoccupationsleadingtowithin-industryjobpolarization.ButRBTCalsoleadstosigni cantbetween-industryshiftsinthestructureofemployment.Ontheonehand,anindustryaffectedbyRBTCwilluselessemploymenttoproduceagivenlevelofoutputwhichwillcauseoccupationalemploymentsharestopolarizeevenifoutputsharesdonot.Ontheotherhand,industriesintenseinroutinetaskswillseealargerdecreaseinrelativecostsandoutputpricesleadingtoashiftinproductdemandtowardstheseindustries(aswas rstpointedoutbyBaumol1967).Weshowthat,inourdata,thiseffectattenuatesbetween-industryjobpolarizationbutdoesnotoverturnit.Theremainderofthispaperisorganizedasfollows.SectionIIdescribesthedata.Sec-tionIIIshowshowtheoccupationalemploymentstructurebetween1993and2010in16WesternEuropeancountriesispolarizingbothwithinandbetweenindustries.SectionsIVpresentsoureconomicmodel.SectionV nallyshowsthatthemodelcanexplainalargepartofthechangesinemploymentsharesacrossoccupations,boththetotalchangeandthesplitintowithin-industryandbetween-industrycomponents.

I.Data

Inthissectionwedescribethedatasourcesofourmeasuresofemployment,therou-tinenessandoffshorabilityofoccupationsaswellasourmeasuresofindustryoutputandcosts.1

A.Employment

OurmaindatasourceforemploymentistheharmonizedindividuallevelEuropeanUnionLabourForceSurvey(ELFS)forthe17-yearperiod1993-2010.TheELFScon-tainsdataonemploymentstatus,weeklyhoursworked,2-digitInternationalStandardOccupationalClassi cation(ISCO)codesand1-digitindustrycodesfromtheClassi ca-tionofEconomicActivitiesintheEuropeanCommunity(NACErevision1).Through-outthispaper,weuseweeklyhoursworkedasthemeasureofemployment,althoughourresultsarenotaffectedbyusingpersonsemployedinstead.Outofthe28countriesavailableintheELFS,weexclude11newEUmembercountriesandIcelandbecauseoflimiteddataavailability.BecauseELFSdataforGermanyonlystartin2003,weusedatafromtheGermanFederalEmploymentAgency’sSIABdatasetwhichisa2%ran-domsampleofsocialsecurityrecordsfortheperiod1993-2008instead.Insum,dataforthefollowing16Europeancountriesisusedintheanalysis:Austria,Belgium,Den-mark,Finland,France,Germany,Greece,Ireland,Italy,Luxembourg,theNetherlands,Norway,Portugal,Spain,SwedenandtheUnitedKingdom.

Table1listsoccupationsorderedbytheirmeanwagerank.2Column(1)ofTable1givesthe1993employmentshareofeachoccupationafterpoolingemploymentacrossthe16countries.Foreachoccupation,column(2)ofTable1givesthepercentagepoint

1SeeOnlineAppendixAfordetailsonthemeasuresdiscussedinthissection.

2Torankoccupations,weusewagesfromtheEuropeanCommunityHouseholdPanel(ECHP)andEuropeanUnion

StatisticsonIncomeandLivingConditions(EU-SILC),sincetheELFSdoesnotcontainanyearningsinformation.SeeOnlineAppendixAformoredetailsonourwagemeasures.

AER文章

VOL.VOLNO.ISSUEEXPLAININGJOBPOLARIZATION3

changeintheemploymentsharebetween1993and2010;wereturntothesenumbersbelow.

-TABLE1ABOUTHERE-B.RoutinenessandOffshorabilityofOccupations

FollowingtherecommendationofAutor(2013)thatresearchersuse,asfarasispossi-ble,off-the-shelfmeasuresforthecontentofoccupations,ourmeasureoftheroutinenessofanoccupationistheRoutineTaskIntensity(RTI)indexusedbyAutorandDorn(2013)andAutor,DornandHanson(2013)3mappedintoourEuropeanoccupationalclassi ca-tionandnormalizedtohavezeromeanandunitstandarddeviation.TheresultingRTIindexisreportedincolumn(3)ofTable1.RTIishighestat2.24forof ceclerks(41)andlowestat-1.52formanagersofsmallenterprises(13).

WhiletheliteratureseemstobesettlingonusingtheRTImeasureasthebestwaytocapturetheimpactofrecenttechnologicalprogress,thereis,asyet,nosimilarconsensusmeasureaboutanoccupation’soffshorabilitytocapturetheimpactofoffshoring.WeuseameasuretakenfromBlinderandKrueger(2013).4UsingtheindividuallevelPrince-tonDataImprovementInitiative(PDII)dataset,BlinderandKrueger(2013)reportthreemeasuresofoffshorability:oneself-reported,oneacombinationofself-reportedques-tionsmadeinternallyconsistent,andthelastonewhichisbasedonprofessionalcoders’assessmentoftheeasewithwhicheachoccupationcouldpotentiallybeoffshored.Blin-derandKrueger(2013)concludethattheirthirdmeasureispreferred.Forouranalyses,weconvertthispreferredmeasureintoourEuropeanoccupationalclassi cationandnor-malizeittohavezeromeanandunitstandarddeviation.Theresultingvaluesarereportedincolumn(4)ofTable1.Themostoffshorablearemachineoperatorsandassemblers(82)at2.35andtheleastoffshorablearedriversandmobileplantoperators(83)at-1.Thecorrelationcoef cientbetweentheRTIandoffshorabilitymeasureis0.46andstatisticallysigni cantsuggestingthatroutinenessandoffshorabilitymaybelinked.5

3ThisRTImeasureisbasedonAutor,LevyandMurnane(2003)andAutor,KatzandKearney(2006,2008).The ve

originalDOTtaskmeasuresofAutor,LevyandMurnane(2003)arecombinedtoproducethreetaskaggregates:theMan-ualtaskmeasurecorrespondstotheDOTvariablemeasuringanoccupation’sdemandfor‘eye-hand-footcoordination’;theRoutinetaskmeasureisasimpleaverageoftwoDOTvariables,‘setlimits,tolerancesandstandards’measuringanoccupation’sdemandforroutinecognitivetasks,and‘ ngerdexterity’,measuringanoccupation’suseofroutinemotortasks;andtheAbstracttaskmeasureistheaverageoftwoDOTvariables:‘directioncontrolandplanning’,measuringmanagerialandinteractivetasks,and‘GEDMath’,measuringmathematicalandformalreasoningrequirements.FromthesethreemeasurestheRoutineTaskIntensity(RTI)indexisconstructedasthedifferencebetweenthelogofRoutinetasksandthesumofthelogofAbstractandthelogofManualtasks.

4OnlineAppendixAprovidessomeadditionalevidenceinsupportofusingthismeasure.Inparticular,weusedataonactualinstancesofoffshoringbyEuropeancompaniesrecordedinfactsheetscompiledintheEuropeanRestructuringMonitor.Thesefactsheetscontainconsiderabledetailaboutoffshoringeventsincludingwhichoccupationsarebeingoffshored.Weprocessedthesefactsheetstoconstructanindexofactualoffshoringbyoccupation.WethenregressthesemeasuresofactualoffshoringbyoccupationonBlinderandKrueger’s(2013)preferredmeasureofanoccupation’soffshorabilityto ndthattheyarestronglyandpositivelycorrelated.Moreover,weshowthatcompetingindicesofanoccupation’soffshorabilitythathavealsobeenusedintheliteraturedonothavethesameexplanatorypower.

5ThisisinlinewithBlinderandKrueger(2013)who ndthatoffshorabilityisconceptuallydistinctfrom,thoughrelatedto,anoccupation’sroutineness.Theauthorsarguethatitislikelythatjobsthatcanbebrokendownintosimple,routinetasksareeasiertooffshorethanjobsrequiringcomplexthinking,judgment,andhumaninteraction.However,a

AER文章

4THEAMERICANECONOMICREVIEWMONTHYEAR

Buttherearealsosomedifferences.Forexample,physical,mathematicalandengineer-ingprofessionals(21)isrankedasnon-routinebutatthesametimehighlyoffshorablewhereasof ceclerks(41)areveryroutinebutnotveryoffshorable.

C.

IndustryOutput,PricesandCosts

Measuresofindustryoutput,industrymarginalcostsandrelativeoutputpricesaretakenfromtheOECDSTANDatabaseforIndustrialAnalysis.Eachofour16countriesexceptIrelandisincludedinSTAN.Thisdatacoverstheperiod1993-2010forall15ofthesecountries.STANusesanindustrylistforallcountriesbasedontheInternationalStandardIndustrialClassi cationofallEconomicActivities,Revision3(ISICRev.3)whichcoversallactivitiesincludingservicesandiscompatiblewithNACErevision1usedintheELFS.

FromSTANweuseproduction,de nedasthevalueofgoodsorservicesproducedinayear,whethersoldorstocked.Toobtainoutput,productionisde atedusingindustry-country-yearspeci cpriceindicesalsoavailablefromSTAN.Toobtainameasureofin-dustryaveragecosts–whichinourmodelpresentedbelowwillalsobeindustrymarginalcosts–weuseSTAN’snetoperatingsurplusdatainadditiontoproductionandoutput,andtakethedifferencebetweenproductionandnetoperatingsurplusanddividethisdifferencebyoutput.

II.

JobPolarization

ToprovideasnapshotofchangesintheEuropeanjobstructure,columns(1)and(2)ofTable1showtheemploymentsharesofoccupationsandtheirpercentagepointchangesbetween1993and2010afterpoolingemploymentforeachoccupationacrossour16Europeancountries.6Thistableshowsthatthehighest-payingmanagerial(12,13),pro-fessional(21to24)andassociateprofessional(31to34)occupationsexperiencedthefastestincreasesintheiremploymentshares.Ontheotherhand,theemploymentsharesofof ceclerks(41),craftandrelatedtradesworkers(71to74)andplantandmachineoperatorsandassemblers(81to83),whichpayaroundthemedianoccupationalwage,havedeclined.SimilartopatternsfoundfortheUSandUK,therehasbeenanincreaseintheemploymentsharesforsomelow-paidserviceworkers(51,ofwhichthemaintaskconsistsofprovidingservicesrelatedtotravel,cateringandpersonalcare;butnot52,ofwhichthemaintaskconsistsofsellinggoodsinshopsoratmarkets)andsomelow-paidelementaryoccupations(91,whichareserviceelementaryworkersincludingcleaners,domestichelpers,doorkeepers,porters,securitypersonnelandgarbagecollectors;but

widevarietyofcomplextasksthatinvolvehighlevelsofskillandhumanjudgmentcanalsobeoffshoredviatelecom-municationdevices.Consequently,BlinderandKrueger(2013)concludethattheirmeasureofoffshorabilityismoregermanetotheoffshoringissuethanisthequestionofroutinenessbutthatthetwocriteriadooverlap.AsourRTIandoffshorabilitymeasuresareundoubtedlyimperfect,andtheextentofmeasurementerrorinthemmayalsobedifferent,onehastobecarefulinone’sclaimsabouttheabilitytodistinguishbeyondreasonabledoubttheseparateeffectsofRBTCandoffshoring.

6Insteadofpoolingcountries,reportingunweightedaverageemploymentsharesandtheirchangesacrosscountrieswouldnotqualitativelychangethenumbersincolumns(1)and(2)ofTable1.

AER文章

VOL.VOLNO.ISSUEEXPLAININGJOBPOLARIZATION5

not93,whichmainlyincludeslow-educatedlaborersinmanufacturingperformingsim-pletasksincludingproduct-sortingandpackingandunpackingproduce).ThisisanindicationthattheexistingevidencetodatethatthereisjobpolarizationintheUS,UKorGermanyisnotanexceptionbutrathertherule.

Theremay,ofcourse,beheterogeneityacrosscountriesintheextentofjobpolariza-tion.Foreaseofexposition,Table2groupstheoccupationsintothethreegroupslistedinTable1:the4lowest-paidoccupations,9middlingoccupationsandthe8highest-paidoccupations.Wethencomputethepercentagepointchangeinemploymentshareforeachofthesegroupsineachcountrybetween1993and2010.Table2con rmsthatem-ploymentpolarizationispervasiveacrossEuropeancountries–theshareofhigh-payingandlow-payingoccupationshasincreasedrelativetothemiddlingoccupationsineachcountry.7

-TABLE2ABOUTHERE-The naltwocolumnsofTable1decomposetheoverallchangeinemploymentsharesreportedincolumn(2)ofTable1intoawithin-industry(column(5))andbetween-industry(column(6))component.Forthegroupof8highest-paidand4lowest-paidoccupationsthewithin-industryandbetween-industrycomponentsarepositive,andforthegroupof9middlingoccupationstheyarebothnegative.Whatthisshowsisthatover-alljobpolarizationcanbesplitintoawithin-industryandbetween-industrycomponent,eachofwhichistoolargequantitativelytobeignored.

Insum,thephenomenonofjobpolarizationispervasiveacrossadvancedeconomies.Moreover,overalljobpolarizationisdrivenbyjobpolarizationbothwithinandbetweenindustries.Thepurposeofthemodelpresentedinthenextsectionistoexplaintotaljobpolarizationand,moreambitiously,itssplitintowithin-industryandbetween-industrycomponents.

III.

AFrameworkToUnderstandJobPolarization

Ourmodelisatwo-stageset-upformodellingtheproductionprocessinwhichoutputindifferentindustriesisproducedfromasetoftasks,andwhereeachtaskisproducedusingatechnologythatcombineslaborfromaspeci coccupationandotherinputs.TocaptureRBTCandoffshoring,thecostoftheseotherinputschangesovertimeaccordingtotheroutinenessandoffshorabilityofatask.Thistypeoftwo-stageset-uphasalsobeenusedinrecentworkby,amongothers,Teulings(1995,2005),AutorLevyandMur-nane(2003),Autor,KatzandKearney(2006),GrossmanandRossi-Hansberg(2008),CostinotandVogel(2010),AcemogluandAutor(2011)andAutorandDorn(2013).Inrelationtothisliteratureourapproachcanbethoughtofaspositionedbetweentheveryempiricalanalysesthateitherestimateonlywithin-industryeffectssocansaynothingaboutthebetween-industryeffectsthatwedo ndareimportant(e.g.Michaels,Natraj

7Onemightbeconcernedthatourresultsaresensitivetothestartandendyears.However,OnlineAppendixA

presentstheemploymentsharesforallyearsshowingthatchangesareclosetolineartrends.