Research Institute of Industrial Economics

P.O. Box 55665

SE-102 15 Stockholm, Sweden

info@ifn.se

www.ifn.se

IFN Working Paper No. 954, 2013

Hierarchies and Entrepreneurship

Joacim Tåg, Thomas Åstebro and Peter Thompson

HierarchiesandEntrepreneurship

JoacimTåg*

ResearchInstituteofIndustrialEconomics(IFN)

ThomasÅstebro**

HECParis

PeterThompson***

GeorgiaInstituteofTechnology

May2016

Weestablishacorrelationbetweenthehierarchicalstructureofafirmandthelikelihoodofbusinesscreationamongitsformeremployees,usingasampleof16millionobservationsofSwedishworkersandanovelproxyforhierarchiesbasedonoccupationdata. Condi‐tionalonfirmsizeandmanyothervariables,employeesinfirmswithmorelayersarelesslikelytoenterentrepreneurship,tobecomeself‐employed,andtoswitchtoanotherem‐ployer.Theeffectsoflayersaremuchstrongerforbusinesscreationthanforjob‐switchingandtheyarestrongerforentrepreneurshipthanforself‐employment.Wediscusstwopo‐tentialexplanationsforthedistinctivehierarchyeffectwefind.Partoftheeffectcouldbetobeduetopreferencesortingbyemployees,andpartduetoemployeesinfirmswithfewerlayershavingabroaderrangeofskills.Onetestshowingthattheprobabilityofentrepre‐neurshipincreaseswiththeirpriorrankinanorganizationisconsistentwithabilitysort‐ingandinconsistentwithpreferencesorting.

Keywords:entrepreneurship;employeemobility;hierarchy,rank,smallfirmeffect.JELcodes:L26,D20,J20,M50.

*ResearchInstituteofIndustrialEconomics(IFN),Box55665,SE‐10215Stockholm,Sweden.E‐mail:joacim.tag@ifn.se.JoacimgratefullyacknowledgesfinancialsupportfromVinnova,theJanWallanderandTomHedeliusFoundation,andtheMarianneandMarcusWallenbergFoundation.**HECParis,1ruedelaLiberation,78351Jouy‐en‐JosasCedex,France.Email:astebro@hec.fr.ThomasgratefullyacknowledgesfinancialsupportfromHECFoundationandtheHECLeader‐shipCenter.***SchellerCollegeofBusiness,GeorgiaInstituteofTechnology,800W.PeachtreeStreetNW,Atlanta,GA30308,USA.Email:peter.thompson@scheller.gatech.edu.WethankTinoSanandajiandparticipantsinseminarsinAmsterdam,Bergen,Cambridge,Copenhagen,Leuven,Lund,Madrid,Milano,andUppsalaforexcellentcommentsandsuggestions.WealsothankSe‐bastianJävervallforoutstandingresearchassistance.

2

1.Introduction

Comparedwiththeircounterpartsinlargefirms,workersinsmallfirmsaremorelikelytoseparatefromtheiremployers(AndersonandMeyer,1994;LazearandShaw,2008)andthosethatleavesmallfirmsaremorelikelytobecomeentrepreneursorself‐em‐ployedthanthoseleavinglargefirms(Wagner,2004;DobrevandBarnett,2005;Gom‐pers,LernerandScharfstein,2005;Elfenbein,HamiltonandZenger,2010;Chen,2012).

Onecandidateexplanationforthis“smallfirmeffect”isthesegregationofworkertypesinto large and small firms. Indeed, Elfenbein et al. (2010) use self‐reported data onwhether individualshaveadesire tobecomeentrepreneurs toshowthat individualswiththisdesireareoverrepresentedinsmallfirms,andaremorelikelytosubsequentlyestablishabusiness.

Smallfirmsattractindividualswithpreferencesandabilitiesforentrepreneurshipnotjustbecausetheyaresmaller,butbecausetheyaredifferent.Inparticular,largefirmsaregenerallymorehierarchicalandbureaucratic,characteristicsthatmayinducethemtoassignlessvaluetoskillsappropriateforentrepreneurshipandthatrepelworkerswithapreferenceforautonomyandworkvariety.Unfortunately,researchontheeffectoforganizationalbureaucracyontransitionstoentrepreneurshiphasbeenlimitedtodatebyitsrelianceonpotentiallycrudeproxies,notablyfirmsizeandage(e.g.,DobrevandBarnett,2005;Sørensen,2007;Kacperczyk,2012).1Butthefactthatfirmsizeisinverselyrelatedtoentrepreneurshipmightbeduetoahostofreasons.Forexample,smallandyoungfirmsaremuchmorelikelytogooutofbusinessthantheircounterparts(Haltiwanger,JarminandMiranda,2013),thusmechanicallycreatinganegativecorre‐lationbetweenbothfirmsizeandfirmagewithentrepreneurship.

Inthispaper,whilecontrollingforfirmsizeandageeffects,ourobjectiveistousetwowell‐establishedmeasuresofthehierarchicalstructureofthefirminordertoexaminetheircorrelationwithentrepreneurship.First,weusethenumberoflevelsofdecision‐makinginthehierarchy.Second,wemeasurethespanofcontrol.Thesetwomeasuresrepresenttwooutofthreebasicstructuralmechanismsusedtohelpcoordinatedividedtasks(GriffinandMoorehead,2014,p.438)andwhichappearwithgreatersizeandageoffirms.2Entrepreneurshipisarareevent,andanalysingitscorrelationwiththehier‐archicalstructureofthefirmrequiresanextensivedatasetonthehierarchicalstructureoffirms.Typically,suchdatasetshavenotbeenavailablewhichhasrestrictedanalysis

1 Defending the choice of using firm size andage as representativemeasure of bureaucracy,Sørensen(2007)writes:“Anorganization’sdegreeofbureaucratizationisnotdirectlyobserva‐ble.Moreover,constructingandcollectingspecializedmeasuresofhierarchy,rolespecialization,androutinizationinthelargesamplesneededtocapturetransitionstoentrepreneurshipispro‐hibitivelydifficult.”(p.395)2Thethirdmechanismisdepartmentalization–themannerinwhichdividedtasksarecombinedandallocatedtoworkgroups(GriffinandMoorehead,2014,p.438).Ourdatadonotallowustocapturethismechanismwithanyprecision.

3

ofthehierarchicalstructureof firmstosmallsamplesizes. Incontrast,oursampleisdrawn from the Swedishmatched employer‐employee dataset, and consists of over240,000firm‐levelobservationsand16millionindividual‐levelobservationsovertheperiod 2001–2008. Following the pioneering work of Caliendo and Rossi‐Hansberg(2012)andCaliendo,MonteandRossi‐Hansberg(2015),weuseemployees’jobtitlestoclassifythemintooneofuptofourranksintheirorganization(CEOs,seniorstaff,su‐pervisors,andproductionworkers).Wethenmeasurethehierarchalstructureofthefirmbycountingthenumberofdistinctranksthatarerepresentedinthefirm,whichwerefertoasthenumberoflayers.Wealsocountthenumberofemployeesateachrankandcomputethespanofcontrolastheratioofthenumberofemployeesatalowerrankto thenumberat thenexthigher rank, andaverage thismeasureacross all adjacentranksintheorganization.Ourdataallowustocreatethesetwomeasuresforallfirmswithatleasttwoemployeesinaneconomy.3Oursamplebehavesasonewouldexpectfromameaningfulmeasureofapyramidalhierarchy:higherrankscontainfewerem‐ployees and payhigherwages than lower ranks, and employee transitions aremostlikelytobetoanadjacentrank.

Section2describesourdata.InSection3,wedemonstratethatthenumberoflayersinthefirmhasalargeeffectonthelikelihoodofentrepreneurship.Conditionalonfirmsizeandage,employeesinfirmswithmorelayersarelessmobile.Thisistrueregardlessofthemover’sdestination,buttheeffectoflayersonbusinesscreationisgreaterthanitseffectonrelocationtootherincumbentfirms.Wedecomposebusinesscreationintoself‐employmentandentrepreneurship,definedbythelegalformofbusinessthatiscreated.Morelayersinafirmsuppressbothtypesofbusinesscreation,buttheeffectisespeciallypronouncedforentrepreneurship.

InSection4,wediscusstwoprominentarguments‐‐sortingbypreferenceandsortingbyability‐‐forwhybureaucracyshouldbecorrelatedwithentrepreneurship.Wefirstlookatevidenceforthepreferencesortinghypothesis.Individualswithastrongprefer‐enceforentrepreneurshipshouldbemorewillingtoestablishbusinessesthatyieldlowfinancialreturns,andtheyshouldalsopersistinbusinessforlonger.Wetesttheseim‐plications,andconclude thatemployeesof lesshierarchical firmsdobehave inwaysconsistentwithpreferencesorting.Wethenassesswhethertherearedifferencesinen‐trepreneurialabilityamongemployeesof firmswithdifferentnumbersof layers.WeframethisanalysisaroundGaricano(2000)andCaliendoandRossi‐Hansberg’s(2012)information‐processing model, which relates the number of layers in a firm to thebreadthofproblemsthatemployeesatdifferentranksareexpectedtobeabletosolve.

3Itshouldbenotedthatwithagivennumberofemployees,arelativelytallstructure(manyhi‐erarchicallevels)mustnecessarilyhaveanarroweraveragespanofcontrol,andsowewillex‐aminewhetherthereisanyaddedvalueofincludingbothmeasurescomparedwithusingjustone.

4

Althoughtheydonotstudytheirmodel’simplicationsforemployeeseparations,wede‐riveandtesttwoimplications.First,employeesinfirmsofagivensizebutwithfewerlayersshouldhaveahigherpropensitytoenterentrepreneurshipandself‐employmentandtobemoresuccessfulwhentheydo.Ourevidenceonbusinesscreationratesiscon‐sistentwiththeinformation‐processingmodel(butalsowithpreferencesorting);ourevidenceonearningsdonotsupportthemodel.Second,thepropensityforbusinesscre‐ationisgreaterwhenemployeesofagivenrankhavemorelayersbeneaththem.Wefindstrongsupportforthisprediction.

2.TheData

ThedataaredrawnfromtheStatisticsSweden’sLISAdatabasedrawingonseveraloffi‐cialregistrydatabasesofeverypersonlivinginSweden.Weobtaininformationforem‐ployed workers on occupation codes, firm‐worker links, worker’s labor income,worker’scapitalincome,andnumerousotherworkercharacteristics.Wealsoobtainin‐formationonfirmageforlimitedliabilityfirmsfromtheSwedishCompaniesRegistra‐tionsOffice.Wemakeuseofdatafortheperiod2001–2008forentrydataand2004‐2011forstart‐upperformancedata.4Wethenrestrictourattentiontoindividualsbe‐tweentheagesof20and60employedinfirmswithmorethanoneemployee(thisisourinitialdataset).

Becausewewill construct indicators of employee rank and organizational structurefrominformationonoccupationalcodesassignedbyStatisticsSweden,wefurtherre‐strictoursampletoincludeonlyfirmswithsufficientoccupationdataforitsemployees.InLISA,70percentofallpotentialindividual‐levelobservationscontainoccupationdatafortherelevantyearandfirm‐workerlinkastheoccupationdataisbasedonsurveysthatdonothavecompletecoverage.Becauseoccupationdataisnotcompleteforeveryyear,wedropfirm‐yearobservationswithfewerthan75percentofallemployeesinafirmhaving accurate occupationdata. This affects observationsmainly from smallerfirmsthatare less likely tobesampledbyStatisticsSweden.5 Inour finalsample,90percentofall individual‐levelobservationscontainoccupationdata,whichcovers62percentofallindividual‐yearobservationsintheinitialdataset.

Ineachyear,wetrackwhetheremployeesremainedwiththeircurrentfirm,switched

4Accurateoccupationdataisonlyavailablefrom2001andonwards.Workerlevelinformationis,however,availablefrom1990andonwards.Weuseinformationfrompriorto2001tocon‐structmeasuresofworkertenure,workerwagegrowth,andfirmageforunlimitedliabilityfirms.Forworkertenureandfirmage,weincludetruncationdummiesintheregressionstoaccountforthefactthatweonlyobservefirmsandworkeremploymenthistoriesbackto1990.5Inrobustnessanalysis(inanappendixavailablefromtheauthors)wetestwhetherdroppingfirm‐yearobservationswithfewerthan90,95and99percentofallemployeesinafirmhavingaccurate occupation data change our results. The key resultswe report in Section 3 are un‐changed.

5

toanotherincumbentfirm,orcreatedandbecameprimarilyoccupiedinrunningtheirownbusiness.6WefollowthepracticeusedbyStatisticsSwedentodefineentrepreneursandtheself‐employed.StatisticsSwedendefinesanindividualasbeingemployedinherownfirminagivenyearifhertotalincomefromherowncompany(laborandcapitalincome)isgreaterthan62.5percentofallotherlaborincome.7Thisisastrictercriterionthanjustreportingtobeanownerofabusinessorreportingsomeearningsfromabusi‐ness.Wedefineanindividualasenteringentrepreneurshipinanygivenyearifthefol‐lowingcriteriaaresimultaneouslyfulfilled:

1.Newlyoccupiedinownbusiness.AnindividualisclassifiedbyStatisticsSwe‐denasworkinginherowncompanyinthecurrentyear,buthadnotbeeninthepreviousyear.

2.Newplaceofwork.Theindividual’scurrentfirmandestablishmentidentifiersaredifferentfromthepreviousyear,and

3.Newfirm.Noindividualinoursamplehadworkedforthecurrentfirminthepreviousyear.

StatisticsSwedenalsoseparatessoleproprietorshipsfromlimitedliabilitycompanies.Becauseanindividualwhointendstocreateagrowingcompanylikelywillestablishalimitedliabilitycompany,wecharacterizeagentsstartinglimitedliabilitycompaniesas“entrepreneurs”,andagentsstartingsoleproprietorships“self‐employed”.8

Table1displaysnumbersonthesamplesizeof the finaldataset. Itcontains243,689firm‐yearobservationscovering61percentofvalueaddedand53percentofemploy‐mentintheSwedishprivatesector.Ofaround16.5millionemployee‐levelobservations,about9.5percenttransitiontoanewincumbentemployerineachyear,about0.28per‐cententerself‐employment,whilejust0.06percententerentrepreneurship.Thenum‐beroffirmsinthesamplevariesconsiderablyacrossyears,aresultofdifferentsamplingof occupation codes and industry sectors over time. Differences in sampling framesacrossyearswillbedealtwithbyindustryandyeardummiesinregressions.

6Allotheremployeeexits(suchasmovestounemploymentormovesabroad)areignored:wedroptheseobservationsatthetimeofexit.Theappendixprovidesrobustnessanalysiswhereweincludemovestounemploymentintheanalysis.7StatisticsSwedentreatsself‐employmentastheprimaryoccupationeventhoughitmaygener‐atelessincomethanotheractivities,becauseself‐employmenttypicallygenerateslowerhourlywages.8Theselabelsarearbitrary.Wecouldinsteadhaveusedtheterms“growth”and“life‐style”en‐trepreneurs.Themeansizeofalimitedliabilitycompanytwoyearsaftercreationis5.3employ‐eeswithastandarddeviationof10.5,whilethemeansizeofasoleproprietorshipis1.2employ‐eeswithastandarddeviationof0.75.HenreksonandSanandaji(2013)surveytheliteratureonentrepreneurshipandself‐employmentanddemonstratetheimportanceofseparatingbetweenentrepreneursandtheself‐employedusingdataonbillionaireentrepreneurs.

6

Table1

Samplesize

Year Firms Workers TootherfirmToself

employmentTo

entrepreneurship

2001 44,959 2,011,755 206,443 5,167 673

2002 36,289 2,112,875 191,439 5,084 542

2003 28,574 2,039,069 186,524 6,149 1,304

2004 23,637 2,003,621 176,524 6,151 1,248

2005 22,481 2,033,432 189,111 6,305 1,263

2006 27,822 2,071,002 221,266 5,609 1,516

2007 32,211 2,138,253 200,666 5,896 1,927

2008 27,716 2,075,520 179,036 5,463 1,154

Total 243,689 16,485,527 1,551,009 45,824 9,627

2.1Occupationclassifications,rankandlayers

Ourdataonthehierarchicalstructureoffirmsaredevelopedfromoccupationalclassi‐ficationsobtainedfromStatisticsSweden’sLISAdatabase.TheSwedishStandardClas‐sificationofOccupations1996(SSYK)isaSwedishversionoftheInternationalStandardClassificationofOccupations(ISCO‐88).Atthelevelofaggregationweareinterestedin,thetwoclassificationschemesareclosetoidentical.TheSSYKdataavailablefromtheLISAdatabasecomemainlyfromtheofficialwagestatisticssurvey(Lönestrukturstatis‐tiken)andfromasupplementarysurveyoffirmsnotincludedintheofficialwagesurvey.TheSSYKassignsworkerstooneoftenmainoccupationalcategories,andoneofalargenumberofsubcategories.Thisassignmentisaprecisedefinitionofanindividual'sjobduties.StatisticsSwedenalsoassignseachoftheseoccupationalcategoriestooneoffourskilllevels.9

Betweenthetwosurveys,atleast40,000firmsaresampledeveryyear.Thesamplingdesignisarollingpanel,andalleligiblefirmsaresurveyedatleastonceeveryfiveyears.Firmswithatleast500employees,however,arealwaysincluded.Notallfirmsarein‐cludedinthesesurveys.Thelargestexcludedcategoryisofself‐employedworkerswhodonotobtainanywagefromalimitedliabilitycompany,butthesurveysalsoexcludeownerswho receive payment from their companies only in the formof dividends.10

9StatisticsSwedennotesthat,althoughtheseskilllevelshavebeenmadeoperationalintermsoftheeducationalcategoriesoftheInternationalStandardClassificationofEducation,theydonotimplythattheskillsnecessarytoperformthetasksanddutiesofagivenjobcanbeacquiredonlythroughformaleducation.ExceptfordisplayingtheseskilllevelsinTable2,wemakenouseofthem.10Therehavebeensomeattemptssince2004tosurveypartnershipsandsoleproprietorships

7

Firmsinoursamplearethereforenotrepresentativeoftheentirepopulationoffirms.11However,despitetheseexclusions,therearegoodreasonstoexpectourfindingstoberepresentativeofthepopulationoflargerfirms.Forlargerfirms,theoccupationdataiseithercollectedforallfirmsaboveasizecutofforcollectedthroughrandomsamplingoffirmsconditionalonfirmsizeandindustryaffiliation.Thesamplingvariationswillbecontrolledforwithdummyvariablesinouranalysis.However,toconfirmourintuitionthatthefinalsampleweuseisrepresentative,wealsoreportintheappendixavailablefromtheauthorsoutput fromaregressionusingsampleweights.Ourresultsremainvirtuallyunchangedwhenaccountingforthesampleselectioninthismanner.Adraw‐backofthefinalsampleis,however,thatduetotheuseofrollingpanelswedonottyp‐icallyobservethehierarchicalstructureofagivenfirmformanyyearsmakingitdifficulttousefirmfixedeffectsinouranalysis.

FollowingtheempiricalmethoddevelopedbyCaliendoetal.(2015)forFrenchoccupa‐tionaldataandre‐examinedbyTåg(2013)onSwedishoccupationaldataformanufac‐turingindustries,weusetheSSYKcodestoassignaranktoeachemployeeinfirmswithmorethanoneemployee(seeTable2).Theideabehindtheclassificationistogroupasetofworkerswithsimilarlevelsofknowledgetothesamerank,representingapartic‐ularlayerinthehierarchy.WechosetoclassifyindividualsintofourranksonthebasisthatthisclassificationhasbeenshownbyCaliendoetal.(2015)andTåg(2013)toyieldlogicallyconsistenthierarchicalstructures.Caliendoetal.(2015)usethePCS‐ESEclas‐sificationfortheiranalysis,whichdifferssomewhatfromtheclassificationweusehere.However,asshowninTåg(2013),boththeclassificationweusehereandaclassificationbasedonmappingthePCS‐ESEcodestotheSwedishSSYKcodesyieldsimilarresultsintermsofgeneratinghierarchicalstructuresthatcorrespondtotheoreticalpredictions.

Wewillalsoshowinthenextfewtablesthatweobtainsimilarlyconsistenthierarchies.Workersindifferentoccupationsareclassifiedintoranksonthebasisoftheirwages,whichshouldproxy for theknowledgeofworkers.Thehighest rank,whichwe labelCEOs,consistsofdirectors,chiefexecutives,managersofsmallenterprises,andcertainotherseniorofficials.Thenexttworankscomprisetwolevelsofmanagement.Themoresenior,whichwelabelas“Seniorstaff”,containsproductionandoperationsmanagersandcertainotherspecialistmanagers.Thelesssenior,“Supervisors”,consistsofwork‐erswithoccupationsclassifiedintheSSYKasprofessionals,techniciansandassociateprofessionals. The fourth category, “Production workers”, comprises clerks, serviceworkers,plantandmachineoperators,andothernon‐supervisorypositions.Forfirmswithonlyonereportedrankofworkers,wedesignatethehighestpaidindividualtobe

butthecoverageisextremelylimited.11Theappendixdetailsthedifferencesbetweenthebasesampleandthefinalsampleweusefortheanalysis.Sampleselectionisapparentsinceallvariableslistedintheappendixarestatisti‐callydifferentinmeansatthe1%level.

8

arankaboveallotherworkers.12AsTable2shows,ourrankclassificationsbroadlyco‐incidewiththeskilllevelsattributedtothesepositionsbyStatisticsSweden,whichatfirstblushmakestheclassificationmethodseemreasonableingroupingasetofworkerswithsimilarlevelsofknowledgetothesamerank.13

Table2Occupationalclassifications,skilllevels,andrank

SYYKOccupationalClassification SkillLevel Rank1.Legislators,seniorofficials&managers NA 4.CEOs:SSYK121(Directorsandchiefex‐

ecutives), 131 (Managers of small enter‐prises),111(legislatorsandseniorgovern‐mentofficials),112(seniorofficialsofspe‐cial‐interestorganizations)3.Seniorstaff:SSYK122(Productionandoperationsmanagers),123(Otherspecial‐istmanagers)

2.Professionals 4 2. Supervisors: SSYK 200‐399 (Profes‐sionals, technicians and associate profes‐sionals)

3.Technicians&associateprofessionals 3

4.Clerks 2 1. Production workers: SSYK 400‐999(Clerks, service workers and shop salesworkers, skilled agricultural and fisheryworkers,craftandrelatedtradesworkers,plant and machine operators and assem‐blers,andelementaryoccupations).

5.Serviceworkers&shopsalesworkers 26.Skilledagricultural&fisheryworkers 27.Craft&relatedtradesworkers 28.Plant&machineoperators&assemblers 29.Elementaryoccupations 1

0.Armedforces NA Omitted

12Ahandfulofsmallfirmscontainemployeesallhavingtherank“CEO”.Wedroptheseaberrantfirmsfromthesample.13Usingfewerrankswouldnotbereasonable,inlightofthedistributionofwagesacrossranksdisplayedinTable3.Ifarankissupposedtoconsistofworkersofsimilarlevelsofknowledge,andwagesproxyforknowledge,wewouldbegroupingworkersofdifferentknowledgelevelstogether.Itwouldperhapsbepossibletoconsiderusingmorethanfourranks.Suchaclassifica‐tionwouldneed tobebasedonawiderdistributionofoccupationcategories than thosedis‐playedinTable3.OnepossibilityweconsideredwastousethenineSSYKcodesdisplayedinthelefthandcolumnofTable2asnineranks.However,whiletherearesomedifferencesinaveragewagesacrosstheninecategories,thehierarchicalstructuresthatresultfrequentlyfailtoinducestructuresthatresemblethosewewouldbeexpecting,withfewerandhigherpaidworkersinhigherranks.

9

Doourclassificationsinducepatternsconsistentwithournotionsofrankandhierar‐chies?Welookfirstatearningsbyrank.Theearningsdataweusethroughoutthepapermeasureeachindividual’sannuallaborincome.However,forcompactness,wewillgen‐erallyrefertoourearningsmeasureasthe“wage”.Table3displaysthewagedistribu‐tionbyrank,whichareoverwhelminglyconsistentwithourgoalofgroupingasetofworkerswithsimilar levelsofknowledgeto thesamerank.Workers inhigherrankstendtoearnmoreonaveragethanworkersinlowerranks,withthenotableexceptionthatSeniorstaffonaverageearnmorethantheCEOrankatallbutthehighestpercen‐tiles.However,thisexceptiondoesnotimplythatSeniorstaffearnmorethantheirim‐mediatemanagers.Insteadindividualswiththeoccupationcode131“Managersofsmallenterprises”,whobelongtotheCEOrank,typicallyearnlessthanSeniorstaffinlargeenterprises,thereforecreatingtheillusionthatSeniorstaffearnmorethantheirimme‐diatemanagers.Furtheranalysisintheappendixdetailsthesepoolingeffects,14whichwecontrol forbyincludingfirmsize inallregressions.Thetablealsoshowsthatthewithin‐rankwagedistributionislowestatthebottomrankandincreasesmonotonicallywithrank,aswouldbeexpectedfromthetheoryoffirmsasknowledgehierarchies,duetothegreaterabilityofmoreknowledgeableworkerstoleverageminorskilldifferencesforgreaterpay(GaricanoandRossi‐Hansberg,2004).

Table3

WagedistributionacrossbroadoccupationclassesbasedonSSYK

Rank MeanPercentiles Wage

Dispersion10th 25th 50th 75th 90th

4.CEOsanddirectors 523 217 285 368 565 940 1,114

3.Seniorstaff 505 293 359 444 575 761 678

2.Supervisors 308 159 228 285 359 469 453

1.Productionworkers 219 112 172 222 266 313 264

Mean 272 131 195 251 317 415 373Data are in units of 1,000 SEK in 2005 prices. Sample sizes are as follows. Rank 4:109,194.Rank3:366,927.Rank2:5,628,096.Rank1:5,665,137.

Table4plotsranktransitionsamongworkersremainingwiththesamefirm.Transitionratesdeclinewithdistancebetweenrankpairs,andmostemployeesdonotchangerank.Forexample,94.3percentofproductionworkersareatthesamerankthreeyearslater,while5.1percentareattherankofsupervisorandlessthanonepercenthaveattainedeitherofthetwohighestranks.Amongthosethatdoswitchrank,movingonerankis

14Inaddition,ifweaddcapitalincometowages,Rank3employeesearnonaveragemuchmorethanRank2employees,Seeappendixfordetails.

10

muchmorelikelythanmovingtwoforallranksexceptCEOs,whoexhibitthegreatestmobility.15Tables3and4togethersuggestthatouroccupationalclassificationssucceedincapturingaformofdistanceconsistentwithournotionofranksandhierarchies.16

Ourremainingdescriptiveanalysisfocusesontwomeasuresofthestructureoforgani‐zations:thenumberoflayersinthefirmandthespanofcontrol.Thenumberoflayersissimplythenumberofdistinctranksoccupiedinagivenfirm.Asweassignthehighest‐paidemployeetobeCEOwhenthedataindicateonlyasinglelayer,firmsmayhavetwo,threeorfourlayers.Wecountthenumberofemployeesateachrankandcomputethespanofcontrolastheratioofthenumberofemployeesatalowerranktothenumberatthenexthigherrank,andaveragethismeasureacrossalladjacentranksforafirm‐yearobservation.

Table4

Ratesofwithin‐companyranktransitions Toproduction

workerTo

supervisorTo

seniorstaffToCEO

Productionworker 94.3 5.12 0.40 0.19

Supervisors 3.96 93.3 2.24 0.47

SeniorStaff 3.25 19.3 72.1 5.39

CEOordirector 5.44 13.0 16.0 65.6

Transitionratesoverathree‐yearperiodifavailable;otherwisetwoyears,otherwiseoneyear.

Table5providesinformationaboutwithin‐firmdifferencesinlayerpopulationsandinearningsacrosslayersasafunctionofthenumberoflayersinthefirm.PanelAofthetabledocumentsthemeannumberofemployeesineachlayer,whilePanelBprovidestheprobabilitiesthatalowerlayercontainsmoreemployeesthanthelayerimmediately

15Table4indicatesasomewhatperplexing5.4percentagechancethatapersonatthehighestrankmovesdowntothelowestrankinafirmwithinthreeyears.However,reportingofoccupa‐tionisnotwithoutmeasurementerror.Inparticular,wehaveanalyzedalltransitionsinthebot‐tomlefthandcornerofTable4andfoundthatmanyoftheseappeartobemisclassificationsoftemporaryworkerssuchasconsultantsortradeworkersthatreportthroughtheirtemporaryemployertobeaCEO(astheyareintheirownconsulting/tradesfirm),andthatthislatergetscorrectedtotheirtrueoccupationattheirtemporaryemployer.Further,thisrowpercentageisasmall fractionofallmoves. Whenwecompute theglobalpercentagesofTable4:a)on thediagonal;b)abovethediagonal(=promotions);andc)belowthediagonal(=demotions),92.8%remainintheirrank,4.3%arepromoted,and2.9%aredemoted.InapaperusingrelateddatafromSweden,LazearandOyer(2004)findthattheincidenceofdemotionsis2.1%.Demotionsarethusquiteinfrequentoverallevenwhenincludingapparentmeasurementerror.16Transitionratesbetweenranksareverysimilarifweincludeemployeesthatswitchedfirms.

11

aboveitinthefirmasafunctionofthenumberoflayers.Inthelargemajorityofcases,thefirm’sstructurecorrespondstoournotionofhierarchiesastriangularstructures,wherehigher layersconsistofsmallnumbersofpeoplesupervising largergroupsofworkersinlowerlayers.Forexample,infirmswithonlytwolayers,thelowerofthetwohasmoreworkersin92percentofthefirm‐yearobservations,andthelowerlayercon‐tainsonaveragemorethanthreetimesasmanyemployees.Notealsothatfirmswithmorelayerstendtobemuchlarger:firmswithjusttwolayershaveonaveragetenem‐ployees,comparedwith55and427amongfirmswiththreeandfour.PanelsCandDcompareearningsacrosslayers.Theyshowthatinthemajorityoffirmsmembersofagivenlayerearnmoreonaveragethanworkersinthelayerimmediatelybelowtheminthefirm.

Table5

Sizeandwagebylayers

AMeanSizeofLayer

CMeanWageinLayer

No.ofLayers

Layer1

Layer2

Layer3

Layer4

Total Layer1

Layer2

Layer3

Layer4

2 7.6 2.2 9.8 210.9 307.3

3 30.2 22.2 3.0 55.4 229.1 312.1 470.8

4 244.7 206.2 16.5 5.3 427.2 235.1 327.3 497.9 887.3

Total 44.0 33.7 8.3 5.3 91.3 217.8 311.0 481.6 887.3

BFractionoffirmswithmoreemployeesinlower

layerthaninnexthigherlayer.

DFractionoffirmswithlowermeanwageinlowerlayerthaninnextlayerup.

No.ofLayers

1>2 2>3 3>4 1<2 2<3 3<4

2 0.92 0.88

3 0.69 0.91 0.86 0.86 4 0.60 0.92 0.92 0.95 0.93 0.82

ThelayernumbersdonotnecessarilycorrespondtotheranknumbersusedinTable2.Forexample,afirmwithoneCEOandseveralblue‐collarworkerswillhavetwolayersandconsistofworkersinranks3and0.Suchfirmswillappear in the same cells as firms consisting only of, for example, ranks 3 and 2. Sample sizes: 2‐layer firms:N=161,861;3‐layerfirms:N=49,043;4‐layerfirms:N=32,785.

Table6providessummarystatisticsonoursecondmeasureofbureaucracy,thespanofcontrol.Themeanfirmspanofcontrolis6.17withastandarddeviationof17.4.Spanofcontrolisgreaterinlargerfirmsandinfirmswithmorelayers.Thesesummarystatisticsindicatethatfirmspanofcontrolispositivelycorrelatedwithboththenumberoflayersandfirmsize.

12

Table6

Spanofcontrolbyfirmsizeandnumberoflayers

NumberofLayers

No.ofemployeesperimmediatesupervisor

Firmsize(employees)

No.ofemployeesperimmediatesupervisor

234

5.736.707.62

<5050–100100–500500–1,500>1,500

4.5413.1615.3118.4231.80

SpanofcontrolbylayerforfirmswithdifferentnumbersoflayersNumberofLayers Layer2 Layer3 Layer4

2 5.73

3 6.37 7.03

4 5.58 10.60 6.69

Total 5.84 8.46 6.69

TheresultsreportedinTables3through6areconsistentwithTåg(2013)andconfirmthefindingsinCaliendoetal.(2015)forthecaseofSweden.Theremainderofthepaperfocusesonnewfindingsabouttheassociationbetweenthehierarchicalstructureofthefirmandnewfirmformation.

3.HierarchiesandMobility

Inthissectionweexploretherelationshipbetweenworkermobility,especiallyintoen‐trepreneurshipandself‐employment,andthehierarchicalstructureofthefirm.Table7reportsthemainresultsofthissection.Weestimatemultinomiallogitregressionswithfourpossibleoutcomesineachyear:remainwiththecurrentemployer,enterentrepre‐neurship,enterself‐employment,andswitchtoanotheremployer.17Inallregressions,weclusterthestandarderrorsatthefirmlevel.Thekeyresulthereisthatanincreaseinthenumberoflayersisnegativelyassociatedwithmobilityofalltypes,althoughtheeffectismuchstrongerforentryintoentrepreneurshipandself‐employmentthanitisformovingtoanotheremployer.Subsequently,Table8revealsthatanincreaseinthespanofcontrol isassociatedwithanincreaseinmovementtootheremployersbutadeclineinthelikelihoodofbusinesscreation;themagnitudeoftheeffectsofchangesinthespanofcontrolis,however,verysmall.

17Intheappendixwealsotakeintoaccounttransitionstounemployment.

13

3.1Controlvariables

Beforelookingatthekeyresultsinmoredetail,wereviewtheresultsconcerningthecontrolsintheregressions.Firmageandfirmsizehavethecustomaryeffectsonentre‐preneurship.Althoughthequadratictermssuggestnon‐monotoniceffects,themagni‐tudesofthecoefficientsindicatenegativerelationshipsbetweensizeandage,andthelikelihoodofentrepreneurship.Asisthecaseinmostsamples,employeetenureisneg‐ativelyassociatedwithmobilityofallkinds,18whilethemoreeducated19andmalesaremoremobileregardlessofdestination.Interestingly,individualwageshaveimpactsthatdifferbydestination.Individualswithhigherwagesarelesslikelytoswitchtoanotherincumbentemployerandlesslikelytoenterself‐employment.However,ahigherwageisassociatedwithanincrease inthepropensitytobecomeanentrepreneur.Thecon‐trasts between these effects of individualwages arenot only statistically significant,theyareeconomicallymeaningful.Forexample,aonestandarddeviationincreaseinlogincomeisassociatedwithadeclineintheoddsofswitchingemployers(relativetostay‐ingwiththecurrentemployer)of15percent,withadeclineintheoddsofenteringself‐employmentof38percent,butwithanincreaseintheoddsofenteringentrepreneur‐shipof4.4percent.Employeeagesimilarlyhasdisparateeffectsonmobilitybydestina‐tion. Increasing age raises theprobability of entering entrepreneurship and self‐em‐ploymentuntilapeakhazardisattainedat38yearsofageforentrepreneurshipand49yearsforself‐employment,afterwhichthehazarddeclines.Incontrast,increasingagereduces theprobabilityofswitching incumbentemployers throughout theentireagerangecoveredinoursample.

Theseresultsforthecontrolvariablessuggestthatmobilityisdrivenbyacomplexin‐terplayofmultipleforces.Forexamples:thenegativeeffectoftenureonmobilityofallkindsisconsistentbothwithjobmatchingmodels(e.g.,Jovanovic,1979)andwithsur‐vivorbiascausedbyunobservedheterogeneityacrossindividualsinthepropensitytomove;theinitiallypositiveimpactofageonbusinesscreationmightreflecttheeffectofwealthconstraintsthatarerelaxedasanagentagesandsaves,ortheconsequencesofon‐the‐joblearningspecifictothedemandsofbusinesscreation;andthecontrastbe‐tweentheeffectsofindividualwagesonentrepreneurshipandself‐employmentrecallsthemismatchingmodelofÅstebro,ChenandThompson(2011),inwhichbusinesscre‐ationismorelikelyamongagentswithespeciallyhighandespeciallylowability.

18Althoughthepositivecoefficientsonthequadratictermsindicateanon‐monotoniceffectoftenure,theturningpointsareattenurelevelsgreaterthanmostofoursampleobservations.19Educationisonascalefrom1–6correspondingto:6.Postgraduateeducation;5.Post‐second‐aryeducation,twoyearsorlonger;4.Post‐secondaryeducation,lessthantwoyears;3.Uppersecondaryeducation;2.Primaryandlowersecondaryeducation;9or10years;and1.Primaryandlowersecondaryeducation,lessthan9years.

14

Table7WorkerTransitions.MultinomialLogitEstimation

Entrepreneurship Self‐employment Otherfirm (1) (2) (3) Firmlayers:3 ‐0.729*** ‐0.541*** ‐0.151*** (0.0454) (0.0275) (0.0461) Firmlayers:4 ‐1.071*** ‐0.714*** ‐0.289*** (0.0493) (0.0334) (0.0476) CEOsandDirectors 1.420*** 0.363*** 0.0202 (0.0727) (0.0486) (0.0504) Seniorstaff 0.988*** 0.351*** 0.172*** (0.0550) (0.0347) (0.0355) Supervisors 0.608*** 0.238*** ‐0.00580 (0.0387) (0.0239) (0.0515) Firmsize ‐9.29e‐05*** ‐7.35e‐05*** ‐2.75e‐05 (1.20e‐05) (8.54e‐06) (1.87e‐05) Firmsizesquared 1.78e‐09*** 1.36e‐09*** 4.95e‐10 (2.35e‐10) (1.72e‐10) (3.40e‐10) Firmage ‐0.00253 ‐0.00372** ‐0.00260 (0.00260) (0.00179) (0.00353) Firmagesquared ‐5.56e‐06 5.71e‐06 1.86e‐06 (2.52e‐05) (1.66e‐05) (3.12e‐05) Age(years) 0.113*** 0.110*** ‐0.0446*** (0.0106) (0.00482) (0.00312) Agesquared ‐0.00148*** ‐0.00112*** 0.000316*** (0.000126) (5.87e‐05) (3.85e‐05) Female ‐0.862*** ‐0.762*** ‐0.124*** (0.0311) (0.0158) (0.0166) Education 0.0214 0.0704*** 0.0759*** (0.0151) (0.00844) (0.00931) Log(laborincome) 0.504*** ‐0.735*** ‐0.158*** (0.0372) (0.0128) (0.0172) Tenure(years) ‐0.129*** ‐0.189*** ‐0.185*** (0.00901) (0.00557) (0.0159) Tenuresquared 0.00303*** 0.00655*** 0.00726*** (0.000613) (0.000336) (0.000721) 3yrwagegrowth ‐0.384*** 0.0562*** ‐0.0415*** (0.0278) (0.00933) (0.0136)

N=16,485,527.Standarderrorsinparentheses.***p<0.01,**p<0.05,*p<0.1.Regressionsincludeindustry,region,andyearfixedeffects,anddummiesindicatingwhetherfirmageandemployeetenurearetruncated. 

15

3.2Hierarchiesandmobility

Weturnnowtoconsiderationofthemaineffectsofrankandlayersonmobility.Con‐siderfirsttheeffectoflayersinTable7;two‐layerfirmsconstitutetheomittedcategory.Even thoughweconditionon firmsizeandage,andemployeerank,more layersarestronglyassociatedwithlessmobilityofallkinds(controllingforfirmsizeusingdummyvariablesforvarioussizeclassesdoesnotchangetheseresults.)Businesscreationrates,especiallyinentrepreneurship,arehighlysensitivetothenumberoflayers.Forexam‐ple,movingfromtwotothreelayersreducesthelikelihoodofatransitiontoentrepre‐neurshipby52percent,andthelikelihoodofatransitiontoself‐employmentby42per‐cent(seeFigure1).Theeffectoflayersonmovementtootherincumbentfirms,whilestatisticallysignificant, issmaller inmagnitudethantheireffectonbusinesscreationrates:eachadditionallayerreducestheoddsofjobswitchingrelativetoremainingwiththefirmbyabout18percent.

Figure1.Oddsratiosofmobility,relativetotwo‐layer firms,bydestinationandnumberoflayers,with95%confidencein‐tervals.FromestimatesinTable7.

Considernexttheroleofrank.Employeesinsupervisorypositionsaremuchmorelikelythanproductionworkers(theomittedcategory)toenterentrepreneurship,andthepro‐pensitytodosoriseswitheachincreaseinrank.CEOs,directorsandseniorstaffarethreetofourtimesmorelikelythanproductionworkerstofoundalimitedliabilitycom‐pany,whilesupervisorsaretwiceaslikelytodoso(Figure2plotstheoddsratios).Sen‐iorstaffandworkerswithsupervisoryrankarealsomorelikelytoenterself‐employ‐ment.Theeffectsaresmallerthanforentrepreneurship,however.Finally,CEOsandDi‐rectors, and Supervisors, areno less likely to switch employers than are productionworkers,whileSeniorStaffareonlyalittlemorelikelytodoso.Thelargeeffectofrankonentrepreneurshipisespeciallynoteworthy,andwillberelevantinSection4forour

16

examinationofcompetingexplanationsfortheeffectoflayersonentrepreneurship.

Figure2.Odds ratiosofmobility relative toproductionworkers,bydestinationandrank,with95%confidencein‐tervals.FromestimatesinTable7.

InmodelAofTable8weaddasaregressor theaveragespanofcontrol in the firm.Conditionalonfirmsize,thereisbyconstructionanegative(albeitnonlinear)relation‐shipbetweentheaveragespanofcontrolandthenumberoflayers.However,asTable8shows,theadditionofthisvariableinducesnochangesintheestimatedcoefficientsonlayers.Moreover,theeffectsofchangesinthespanofcontrolareverysmall:forex‐ample,aonestandarddeviationincreaseinthespanofcontrolisassociatedwithonlyaonepercentdeclineintheoddsofentrepreneurship.Theresultsclearlyshowthatthenumberoflayersisamuchbetterpredictorofmobilitythanisthespanofcontrol.

Because self‐employment and entrepreneurship are likely strong substitutes amongthoseconsideringseparationfromtheiremployer,itisusefultoverifytherobustnessofourresultsbyrestrictingthechoiceset.Wedosobyexaminingtheoddsofself‐employ‐mentandentrepreneurshiprelativetojobswitchingafterlimitingthesampletomovers.ModelBofTable8reportsestimatesofmultinomiallogitestimationonthisrestrictedsample; the baseline category ismoving to an incumbent firm.20 The results clearly

20Ifweweretore‐estimatemodelAofTable8usingmovingtoanotherfirmasthebaseline,itwouldofcoursebepossibletoformallytesttheIIAassumptionusingtheHausman‐McFadden(1984)test.IftheresultsinmodelAarerobust,weshouldfindthatthedifferencesbetweenanycoefficientonentrepreneurshiporself‐employmentandthecorrespondingcoefficientonmove‐menttoanotheremployerinmodelAaresimilartothecorrespondingcoefficientonentrepre‐neurshiporself‐employmentinmodelB.Theresultsmeetthisexpectation.

0

1

2

3

4

5

0

1

2

3

4

5

OddsRatio

Rank

Entrepreneurship

Self‐employmentJob switching

17

demonstrateaneffectofhierarchiesonbusinesscreation:amongmovers,andcondi‐tionalonfirmsize,firmage,spanofcontrol,andrank,thelikelihoodofbusinesscreationdeclinesmarkedlyasthenumberoflayersincreases.Asalreadynoted,theeffectoflay‐ers is greater for entrepreneurship than for self‐employment, and in the latter casemuchoftheeffectisduetothecontrastinself‐employmententryratesbetweenfirmswithtwolayersandthosewithmorethantwo.

Table8

WorkerTransitions.MultinomialLogitEstimation

ModelA

AllworkersModelB

MoversonlyEntrepreneur‐

shipSelf‐employ‐

mentOtherfirm

Entrepre‐neurship

Self‐employ‐ment

(1) (2) (3) (4) (5)Firmlayers:3 ‐0.726*** ‐0.539*** ‐0.134*** ‐0.624*** ‐0.419*** (0.0455) (0.0272) (0.0450) (0.0516) (0.0361)Firmlayers:4 ‐1.073*** ‐0.724*** ‐0.243*** ‐0.854*** ‐0.474*** (0.0494) (0.0336) (0.0478) (0.0528) (0.0421)Spanofcontrol ‐0.000358 ‐0.000960*** 0.00112*** ‐0.00197** ‐0.00354*** (0.000414) (0.000291) (0.000416) (0.000980) (0.000958)

ModelA:N=16,485,527.ModelB:N=1,606,460.Standarderrorsinparentheses.***p<0.01,**p<0.05,*p<0.1.Additionalregressors,notreported,arethesameasinTable7.

3.3Causalinference

Whiletheregressionssofarshowconvincingcorrelationsbetweenthenumberofhier‐archicallevelsandtheentrepreneurialproclivitiesofemployees,theregressionsarede‐scriptiveandresistcausalinterpretation.Intheremainderofthissectionwewillad‐dresswhetheritispossibletoinferacausaleffectofthehierarchyontheprobabilitythatemployeesleavetostartanewbusiness.Thisanalysisisnecessarilylimitedbythewaythedataonhierarchieswascollectedandbyourabilitytocollectdataonaplausibleexogenousshocktofirms’hierarchies.Nevertheless,wewillmakesomeheadwaythatwillshowinstrumentalvariableregressionsconsistentwiththeresultspresentedsofar.

TocreateaninstrumentforthenumberofhierarchicallevelsinafirmwedrawonFrie‐drich (2015) who studies how international trade shocks affect wage inequality bychangestothehierarchyofthefirm.Friedrich(2015)usesthesamemeasureofhierar‐chiesasCaliendoetal.(2015)andborrowsthemethodologyofHummelsetal.(2014)toconstructafirm‐specifictradeshockthataffectsthehierarchicalstructureofthefirm.InlinewiththispriorworkweclaimthataninternationaltradeshockwillaffectthehierarchicalstructureofthefirmsinthesamewaymodeledandtestedbyCaliendoetal.(2015)butwillnot,atleastnotintheshort‐run,affecttheprobabilitythatanem‐

18

ployeeleavesthefirmtostartanewbusiness.Thisisbecausefirst,engagingininterna‐tionaltradeisarareactivity.Forexample,ofthe5.5millionfirmsoperatingintheUnitedStatesin2000, just4percentwereexporters.Second,exportingisdoneprimarilybylargefirms.Amongexportingfirms,thetop10percentaccountedfor96percentoftotalU.S.exports(Bernardetal.,2007.)Inparticular,exportingfirmshave229percentmoreemployeesthannon‐exporters(Bernardetal.,2007.)Smallandnewfirmstypicallydonotimmediatelyexporttheirgoods,andthosethatdoareneverthelessnotthetypeoffirmsthattypicallyreacttoexogenoustradeshocks,butratherexportdirectlybecausetheproduct‐marketforthatgoodhappenstobeforeign.Wethusclaimthatforeigntradeshocksareplausiblyexogenoustotheextensivemarginofentrepreneurship.

WefollowFriedrich(2015)andHummelsetal(2014),whichdefinesanalgorithmforcomputingexogenoustradeshockstofirmsusingglobalUNComtradedata.Wedonothaveaccesstofirm‐leveltradedataasFriedrich(2015)andHummelsetal(2014)did,soinsteadwegenerateexogenoustradeshockstofourdigitindustries,whichwehaveavailableinourdata.21Theaggregationofproducttradedatatoindustrieswillintroducenoiseandattenuateourcoefficientestimates.However,themediannumberoffirmsper4‐digitindustryis17inoursample,sothelevelofaggregationbyindustrymaybelowenoughtoidentifyareasonabletradeshockforagivenfirm.Wedefine

∑ ∑ ,

whereWIDisWorldImportDemand,withj=industry,t=time,k=product,andc=country.WecalculatetheweightingvariablesjckasexportstocountryxfromSwedishindustryyofproductk,weightedbyallexportsinSwedishindustryy.

Animportantissuewithaggregatingtradeshocksisthechoiceofweights.Tradesharesfordifferentproduct‐destinationpairswillvaryovertimeinresponsetocurrentmarketconditions.Inordertoaddressthisendogeneityissue,wefollowFriedrich(2015)andHummelsetal.(2014)anduseindustryj’stradesharesinyearsbeforethebeginningofthesampleforsjck.Notethatintroducingnewproductsorsellingoldproductstonewdestinationcountrieswillhavenoeffectonthetradeshockinstrumentforagivenin‐

21LinkingtradedataforproductcodesHS96toSwedish4‐digitSNI2002industrycodesiscom‐plicatedbythefactthatthelinksarenotalwaysunique.First,0.2%ofthelinksbetweenHS96productcodesandCPC1productcategoriesareone‐to‐manylinks.Second,24%ofthelinksbe‐tweenCPC1productgroupsandISIRev3industriesareone‐to‐manylinks,andafurther7%areone‐to‐manylinksbetweenISIRev3industriestoISIRev3.1industries.Finally,asmuchas43%areone‐to‐manylinksbetweenISICRev3.1andNACERev1.1(whichequalsSNI2002availableinourdata).Werandomlyallocateproductstoindustriesandindustriestoindustrieswhenthereareone‐to‐manylinks.Wehaverunrobustnesscheckswherewerepeattherandomizationfordifferentseedsandourresultsremainthesame.Ingeneraltherandomallocationwilldragdowncoefficientestimatestowardszerobyintroducingwhitenoiseinestimations.

19

dustry,andthusalleviatesconcernsthattheinstrumentisnotexogenoustoentrepre‐neurship.Wecomputesjckforthepre‐sampleyears1996to2000andtaketheaverageweightovertheseyears.

WecannotrunourpreferredmaximumlikelihoodmultinomialchoicemodelsasIVre‐gressionsarenotdefinedundersuchconditions.InsteadwerunlinearprobabilityOLSregressionsandthenIV2SLSregressionstoshowthatflatterfirmsspawnmoreentre‐preneurs.ForthepurposeoftheIVanalysiswecreateaLHSvariable=1formoverstoentrepreneurshiporself‐employment,elsezero,andasemi‐continuousRHSvariablereflectingthenumberoffirmlayers.WeincludeallcontrolvariablespresentedinTable8,ModelAbutdonotreportthem.ThemainresultsfromtheanalysisarepresentedinTable9.TheOLS regression in column1 reports anegative correlationbetween thenumberoflayersinafirmandthelinearprobabilityestimateofmoving.However,thecorrelationdisappearsintheIVregressionastheendogenousvariableisnotsufficientlyidentifiedinfirststageoftheIVmodel.Thereisevidentlyalotofnoiseintheestimates,whichisunderstandablegivenfourfeaturesofourdata(twowhicharedifferentthaninCaliendoetal.,2015):

1. Ourfirm‐leveldataonhierarchiesisonlysampledeveryfifthyearformostfirmsandsincewehaveeightyearsofdata,theaveragenumberofobserva‐tionsperfirmisonly1.9,makingidentificationbasedonchangesinhierarchiesovertimeforagivenfirmdifficult.Thecross‐sectionalidentificationdoesnotsufferfromthesamedifficultyaswehaveplentyoffirmsinthedataset.

2. Wedonothavefirm‐leveldataonproducttrade,butindustrydata.Themap‐pingprocedurefromproducttradetoindustrytoimpactonfirmsasoutlinedabovesufferfromtheintroductionofnoise.

3. TheIVprocedurebynecessityoflinearizationgroupsentryintoentrepreneur‐shipoftwokindstogetherwhereweinMNLanalysisdetectedsignificantdif‐ferencesbetweenthem,andalsogroupsmoverstootherfirmswithstayers,alsointroducingpotentialaggregationbias.

4. Wehavemissingdataonthetradeshocksforafairlylargenumberofobserva‐tionsbecausethesefirmsareinfourdigitindustriesthatdidnotengageinin‐ternationaltradeduringoursampleperiod.Whilewehavecross‐sectionaldataonhierarchiesfor128,262firms,weonlyhavetradeshockdatafor18,266firms.

Wecanmakesomefurtherheadwaybychangingthelevelofanalysis.Inparticular,itisplausibletolookatchangesatthefirmlevelsincethetradeshockoccursatthefirm‐levelnotattheemployeelevel.Theassociatedchangeinthenumberoflayersisalsoatthe firm level. Responses to these changes are of course by individuals. However, alargerfirmhasmoreemployeesthanasmallerfirm,andwecurrentlyrecordthecontri‐butiontotheaverageestimatedeffectofthechangeinthehierarchyatagivenfirmbythenumberofemployeesinthatfirm,puttingmuchmoreweightintheaverageestimateforchangesinlargerthansmallerfirms.Wedotheaggregationtofirm‐levelanalysisintwoways.

20

Wefirstruntheregressionsatthefirmleveltakingasthedependentvariablethefrac‐tionofemployeesthatdepartforentrepreneurshiporself‐employment,clusteringthestandarderrorsatthefirmleveltocontrolforrepeatedobservationsinafirmovertime.Toobtainothercontrols,wesimilarlyformmeansforthefirmofourindividuallevelvariables,whileleavingoutanindividual’spositioninthehierarchyandthefirmagetruncationdummy.ResultsarereportedinColumns3and4.NoticethattheIVregres‐sionproducesacoefficientestimatethatisapproximatelyeighttimeslargerinmagni‐tudethanintheOLSregression.

Analternativewayofdoingthefirm‐levelanalysisistoweighteachindividualobserva‐tionbytheinverseofthenumberofemployeesineachfirm.ResultsfromthisapproacharereportedinColumns5and6ofTable9.Thefirst‐stageidentificationoftheendoge‐nousvariablesisnowsomewhatstronger,asisthe2ndstageeffect.Asatestoftheva‐gariesoflinearizingandcollapsingtheoutcomevariableforthepurposesofIVanalysis,wereruntheregressionreportedincolumn6butonlyformovers,comparingtheprob‐abilityofbecomingeitheranentrepreneurorself‐employed(collapsed)withmovingtoanotherestablished firm.Theequivalentanalysis in themultinomial frameworkwaspresented inTable8,modelB.Thisremovesoneof theproblemswherewecollapseboththosestayingandthosemovingtoanotherfirmintotheintercept.Resultsarere‐portedinColumn7.Resultsareconsistentinthesign,butwithamuchlargercoefficientfortheeffectofhierarchiesonentrepreneurship.Thisindicatesthattheremainingma‐nipulationswhichweapplytodatainordertofittheIVframeworkarestilllikelytobeassociatedwithbias.

Acautiousconclusionmightbethatalthoughsufferingfromseveraldatalimitations,theIVresultpointstoalargerpointestimateofthecausaleffectthantheOLScorrelationalresults,althoughwithaconsiderableamountofnoiseintroducedandwithbiaslikelyintroducedduetothenecessityofcomplyingwithIVregressionconditionsthatmaynotbeappropriateforthedata.

4.ExplainingtheHierarchyEffect

Intheprevioussectionwefoundthatthenumberoflayerswithinafirmisnegativelycorrelatedwithmobilityofallkinds.Morelayersappeartomarkedlysuppressratesofbusinesscreationbyemployees,whilethemagnitudeoftheimpactonmovementstootheremployersissomewhatsmaller.Inthissectionwediscusstwopotentialexplana‐tionsforthecorrelationwefind.

21

Table9

OLSand2SLSwithIV

OLS IV2SLS OLS IV2SLS OLS IV2SLS IV2SLS

1 2 3 4 5 6 7

Numberoflayers ‐.0018*** 0.002 ‐.0032*** ‐.024* ‐.0030*** ‐.020** ‐.111*

t‐value ‐25.33 0.50 ‐18.36 ‐1.80 ‐21.24 ‐1.98 ‐1.80

1ststageequation

Worldimportdemand 0.000009 .000015** .000018*** 0.00002***

F‐value 1.33 5.47 6.86 8.04

Numberofobs.:individual‐year

firms

16,485,527

128,062

2,404,659

18,266

n.a

243,689

n.a

18,266

16,485,527

128,062

2,404,659

18,266

239,311

10,249

***p<0.01,**p<0.05,*p<0.1.Regressionsarelinearprobabilitymodelsofwhetheranemployeebecomesanentrepreneurorself‐employed(=1),comparedtostayingemployedormovingtoanotherestablishedfirm(=0).IVresultsarebasedonjointsimultaneousestimationofstage1and2equations.Regressionsincludeaconstantterm,andcontrolsforrank,labourincome,pastwagegrowth,age,agesquared,gender,education,tenure,tenuresquared,firmage,firmsize,spanofcontrol,industry,region,andyeardummies,anddummiesindicatingwhetherfirmageandemployeetenurearetruncated.Forcolumns3and4weformfirm‐levelaveragesfromemployeedataanddroprankandthedummyindicatingwhetheremployeetenurearetruncated.Forcolumns5and6weweighteachindividual‐yearobservationwiththeinverseofthenumberofemployeesinthefirm.Forcolumn7werunregressionasin6,butcomparemoversbecominganentrepreneurorself‐employed(=1),tothosemovingtoanestablishedfirm(=0).Thoseremaininginthefirmaredropped.

22

4.1Preferencesorting

EversincetheindustrialrevolutionandthesubsequentworkbyFrederickTaylor,HenryFord,MaxWeberandHenryFayol,modernbureaucracieshavebeencreatedto improveefficiencythroughthedivisionoflabor,strictreportingrelationships,andtheinstitutional‐izationofhierarchiestosupportefficienttaskexecutionanddecisionmaking.However,bu‐reaucracieswithpreciselydefinedtasks,strictmanagerialcontrol,andconcomitantlytallhierarchiesoftenreducethepossibilitiesforemployeestoperformvariedworkortoeasilychangebetweendifferentjobswithinthefirm.Suchfeaturesoforganizationsmaythereforeleadpeoplewithanaversiontojobswithmonotonoustasksandstrictrulestoselectawayfromorganizationscontainingsuchjobs.Instead,smallerfirmswithflatterhierarchiesandmoreflexibleproceduresandtasksmayofferpeoplegreaterjobsatisfaction.Indeed,em‐piricalresearchhasshownthata)individualsaremoresatisfiedatworkwithmoreflexibleand varied job tasks (Walker and Guest, 1952; Hackman and Oldham, 1976; Rousseau,1977); b) smaller firmsofferboth greaterworkvariety and job satisfaction (Carter andKeon,1986;BenzandFrey,2008a;Elfenbeinetal.,2010);c)jobsatisfactionisgreaterinlessbureaucraticfirms(Arches,1991;Finlayetal.,1995).

Whilebureaucraticorganizationsmightinducechangesintheiremployees(forexample,theymightcauseemployeestobecomelessinnovativeortheymaymoreeasilyseedtherootsofdisagreementwithfellowworkersormanagers)wefocusourattentionhereonselection.It isnotdifficulttoimaginethat individualswhodislikebureaucracyaremorelikelytochoosetoworkinsmaller,flatter,andlessbureaucraticfirms,andalsomorelikelytochoosetobecomeentrepreneurs.

In favorof theargumentthatthere isapreference forentrepreneurshipstemmingfromdissatisfactionwithwagework,itiswelldocumentedthatacrossawiderangeofcountries,self‐employedaremoresatisfiedwiththeirworkthanwageworkers(Taylor,1996;Blanch‐flower andOswald, 1998; Blanchflower, 2000; Hundley, 2001a; Prottas and Thompson,2006;BenzandFrey,2008a,2008b;Kawaguchi,2008;Fuchs‐Schundeln,2009).22Moreo‐ver, theself‐employedreport theyaremoresatisfiedwiththeir jobsbecause theirworkprovides more autonomy, flexibility, and skill utilization (Hundley, 2001; Prottas andThompson,2006;BenzandFrey,2008b).Theseunderlyingmechanismshavebeenstableoverthepastthirtyyearsandarenotduesimplytopersonalitydifferences.However,thejobsatisfactionadvantagetoself‐employmentisrelativelysmallornonexistentcompared

22UsingtheunificationofEastandWestGermanyasanaturalexperiment,BenzandFrey(2008a)furthershowthatthisresultisnotduetoreversecausation(i.e.,itisnotthecasethatmorehappypeopleenterentrepreneurship).

23

tomanagersandmembersoftheestablishedprofessions,occupationswhereworkershaverelativelyhighautonomyandskillutilization(Hundley,2001),andtheadvantageislargerwhen compared against wage earners working in larger organizations (Benz and Frey,2008a),suggestingthatwhatpeoplereallydislikeaboutwageworkisbureaucracy.

Finally,thereisanextensiveliteraturedocumentingthatsomeindividualsdeclare,oftenatanearlystageintheircareer,apreferencetocreateandoperatetheirownbusiness,andthatsuchdeclarationspredictentryintoentrepreneurship(Hamilton,2000;Halaby,2003;BenzandFrey,2008a;ÅstebroandThompson,2011).Suchpreferencesmaybeshapedal‐readywhenyoung.Forexample,Halaby (2003)andSørensen(2007)showthat theoff‐springofself‐employedparents,whohaveonaveragemuchgreaterentrepreneurialaspi‐rationsthantheoffspringofwageearners,aremorelikelytoworkinsmallerandpresum‐ablylessbureaucraticorganizations.Furthermore,anotherstrandofliteraturedocumentsthatengaginginwageworkwithalotoftaskvarietyisstronglyassociatedwithahigherprobabilityofbecominganentrepreneur(Lazear,2005;Elfenbeinetal.,2010;ÅstebroandThompson,2011).Takingallthisliteraturetogether,thereseemsthereforetobeanassoci‐ationbetweenthepreferencesofpeoplewhofoundbusinessesandthepropensitytoworkinlesshierarchicalorganizations.

Preferencesortingpredictsthatindividualsworkinginsmallfirms,youngerfirmsand,inparticular, firmswith flatterhierarchieswouldbemore likely tobecomeentrepreneurssincepeoplethatdislikebureaucracywouldbemorelikelytosortintoboththesetypesofemployment. Thehypothesis also generates several predictions that are potentially dis‐criminating.First,itpredictsthatthosewithahigherrankinanorganizationwillbelesslikelytoleaveforentrepreneurshipastheyaremorelikelytoenjoywagework.Second,ithasimplicationsforhowwellpeopledoinentrepreneurshipafterleavinganestablishedfirm.Agentscreateandthencontinuetooperatebusinessesifthetotalcompensationex‐pectedfromthebusinessexceedstheforegonewage.However,totalcompensationfromentrepreneurshipandself‐employmentincludesanon‐pecuniarycomponentamongthosewithapreferenceforentrepreneurship,andthiswillinducethemtocreatebusinesseseveniftheyarelikelytoyieldonlymodestmonetaryreturns.Andtocontinueoperatingthosebusinesses longer thanwould someonewithoutnon‐pecuniarybenefits.Becausepeoplewithsuchpreferencesarepresumablyoverrepresented insmallerand less‐bureaucraticfirms,wemaythereforeanticipateeitherorbothof(i)apositiverelationshipbetweenthenumberof layersandsubsequentmonetaryperformance inentrepreneurship,and(ii)anegativerelationshipbetweenthenumberoflayersandthelikelihoodofbusinesssurvival.Similarly,wemayanticipateeitherorbothof(iii)apositiverelationshipbetweenfirmsizeandsubsequentmonetaryperformanceinentrepreneurship,and(iv)anegativerelation‐shipbetweenthefirmsizeandthelikelihoodofbusinesssurvival.

24

4.2.Abilitysorting

Inawell‐knownstudy,Lazear(2005)proposesthatentrepreneursareJacksofalltradeswhosediverseskillsenablethemtocarryoutthevariedtasksinvolvedinthecreationofasuccessfulbusiness.Heconjecturesthatindividualscaninvestintheaccumulationofdi‐verseskillsthrougheducationoron‐the‐jobtraining.Individualswhointendtopursueanentrepreneurialcareerchoosetolearnavarietyofskills,eitherbypursuingavariedcurric‐ulumincollege,byacceptingavarietyofpositionsinworkorbyworkinginfirmsinwhichavarietyofactivitiesneedtobeundertakenbyeachemployee.LazearprovidessupportingevidencefromtheworkhistoriesanduniversitytranscriptsofStanfordMBAalumni,whileothershavefoundsupportingevidenceindiverseandmorerepresentativedatasets(e.g.,Wagner, 2003, 2006; Silva, 2007; Åstebro, Chen and Thompson, 2011; and Chen andThompson,2016).

To linkLazear’s theory to thehierarchicalstructureof firms,weconsideravariationonCaliendoandRossi‐Hansberg’s(2012)modelofhierarchy, inwhichoutputissecuredbysolvingproblems,andworkerknowledgeandtimearekeyinputsintotheproductionpro‐cess.Workersontheshopfloor(theircontextismanufacturing)producebysolvingprob‐lemsthatpresentthemselves. If theyhavetheknowledge, theysolvetheproblemthem‐selves,whichyieldssomeoutput.Iftheydonothavetheknowledgetheypasstheproblemuptotheirsupervisor,whomayeithersolvetheproblemorpassituptothenextlayerofmanagement.Employeesare constrainednotonlyby theirknowledge,but alsoby time.Thus,asupervisorinchargeoftoomanyworkers,orofworkerswithtoolittleknowledge,willnotbeabletoaddressalltheproblemsthatcomeuptheline.

Supposethatproblemscanbeindexedontheline,withthemostcommonproblemsbeingassignedalowerindex.Figure3illustratesafirmwiththreelayers,thearrowsindicatingtherangeofproblemseachtypeofworkerisabletosolve.23Ineachunitoftime,aproduc‐tionworkerreceivesadraw,,ofaproblem.If 1[0, ],z theproductionworkersolvestheproblemhimselfandproducesafixedoutput.However,if 1 ,z theproductionworkerknowsonlythathecannotsolvetheproblemandturnstohisimmediatesupervisorforhelp,whosolvesitif 1 2[ , ]z z andpassesituptotheCEOforasolutionif 2 .z Helpingpro‐

23InCaliendoandRossi‐Hansberg’smodelthereisnooverlapintheknowledgesetsofeachrank,butconsistencywithLazear’stheorydemandsthatworkersinhigherlayersareatleastfamiliarwiththesimplerproblemsfacedbyworkersinlowerlayers.However,Garicano(2000,SectionIV)devel‐opsamodelsimilartoCaliendoandRossi‐Hansberg’sandshowsthatthecharacteristicsoftheopti‐malorganizationdesignremainsimilarwhentheknowledgesetsofworkersathigherlayersincludetheknowledgesetsatlowerlayers.

25

ductionworkerstakestime,h,soineachunitoftimeamanagercanonlysolve1/hprob‐lems.Letf()denotethefrequencywithwhichproblemarises,andletF()bethecorre‐sponding distribution. With n1 production workers, the expected number of problemsbrought to the attention of the supervisor in each period is 1 1(1 ( )),n F z of which

1 2(1 ( ))n F z willhavetobepasseduptotheCEO.

Figure3

Itisnoweasytoseethefirm’spotentialresponsestoanincreaseinthenumberofproduc‐tionworkersinduced,say,byanexogenousincreaseindemand.IfnothingchangesinFigure3,middlemanagersarelikelytobeoverwhelmedwithproblemstheyhavenotimetoad‐dress. Soa firstpossibility is that the firm increases thenumberofmiddlemanagers tomatchthe increase in thenumberofproductionworkers.However, thiscanonlyofferapartialsolution,becausethesingleCEOwillalsobecomeoverwhelmedwiththedemandsonhistime.Inresponse,thefirmmustincreasez2,sothatasmallerfractionofproblemsarepassedtothetopofthechain.Thisrequireshiringmiddlemanagerswithagreaterspanofknowledge.Intheirmodel,CaliendoandRossi‐Hansberg(2012)showthatwhenafirmin‐creasessizewithoutaddingalayer,itincreasesbothz1andz2,hiringmorequalifiedwork‐ersatahigherwageateverylevelbelowtheCEO.Alternatively,thefirmcouldaddalayer.CaliendoandRossi‐Hansberg(2012)showthat,whenafirmexpandslayers,itincreasesthenumberofworkersateachpre‐existinglayer.Theadditionallayersenableworkerstopassmoreproblemsuptheline.Theythereforeneedlessknowledgeandtheaveragewagede‐clines.Caliendo,etal.(2015)exhaustivelyexaminethesepredictionsofthemodelusinga

26

panelofobservationsonthelargemajorityofFrenchmanufacturingfirms,andfindnoin‐stance inwhich the evidence contradicts thepredictions.Tåg (2013) replicates parts oftheirstudyandconcludesthatthepredictionsofCaliendoandRossi‐Hansberg(2012)alsoholdforSwedishmanufacturingfirms.

Ourrenderingoftheinformation‐processingstory,inwhichworkersathigherranksknowhowtosolveproblemsthatthoseatlowerrankscansolve,predictsthatconditionalonfirmsizefirmswithfewerlayersemployworkerswiththeabilitytosolveabroaderrangeofproblems.Similarly,conditionalon thenumberof layers, larger firmsaremore likely tohave employeeswith broad skill sets. Following Lazear (2005), suchworkers aremorelikelytoenterentrepreneurshipandself‐employment,andtheyarelikelytoperformbetterwhentheydo.Abilitysortingthereforepredictsanegativeeffectofmorelayers,andapos‐itiveeffectoffirmsize,ontherateofentrepreneurship,onbusinessearningsamongentre‐preneurs,andonbusinesssurvival.Thetwoexplanations‐‐sortingbypreferenceandsort‐ingbyability‐‐alsohavediscriminatingpredictionsfortheeffectofrankonentryintoen‐trepreneurship:whilepreferenceselectionindicatesthatemployeesathigherrankarelesslikely to found a business because they find their jobmore satisfying, information pro‐cessingsuggeststhathigherrankinducesentrepreneurshipbecausesuchemployeeshaveabroaderrangeofskills.

Table10summarizesthepredictionsofthetwomodels,alongsidetheresultsthatarere‐portedbelow.H1simplynotesthatbothmodelspredictthekeyfindingdocumentedinSec‐tion3.H2andH3,whichrelatefirmsizeandindividualranktoentrepreneurshiprates,arediscriminatingpredictions,asareH4(relatinglayerstobusinessearnings)andH7(relatingfirmsizetobusinesssurvival).

Table10

PredictedandObservedCorrelations

Preferencesorting

Informationprocessing

Results

H1.Numberoflayersv.entrepreneurshipH2.Firmsizev.entrepreneurshipH3.Rankv.entrepreneurshipH4.Numberoflayersv.businessearningsH5.Numberoflayersv.businesssurvivalH6.Firmsizev.businessearningsH7.Firmsizev.businesssurvival

27

4.3Evidence

TheresultsreportedinTable7providesunambiguousevidenceonH2andH3.First,largerfirmsareassociatedwithlowerratesofentryintobusinesscreation,consistentwiththepreferencesortingmodel.Second,individualsathigherranksaremuchmorelikelytobe‐comeentrepreneursthanarethoseatlowerranks,providingsupportforabilitysorting.

TotestH4throughH7,weregressentrepreneurialandself‐employmentearningsandsur‐vivalratestwoyearsafterbusinesscreationonthenumberoflayersinthebusinessowner’sprevious employer.We run regressions including controls forwages, and this inclusionmeritssomecommentary.InCaliendoandRossi‐Hansberg(2015),aworker’swageisasuf‐ficientstatisticforhisskill,sothatlayersandfirmsizeshouldnotmatteronceonecontrolsforearnings.Inpractice,however,wagesareacrudeproxyforskill–becauseofthediffi‐cultiesfirmshaveinmeasuringanindividual’sskilllevelandbecauseoflabormarketsearchfrictionsthatinducewagedispersionamongotherwiseidenticalindividuals.Weanticipate,then,thatlayersandfirmsizeshouldcontinuetomatterintheinformationprocessingstory,evenwiththeinclusionofthewagecontrol.Moreover,asthewagealsomeasurestheop‐portunitycostofentrepreneurshipandiscorrelatedwithfirmsizeandlayers,excludingitisproblematic.

Wereporttheresultsoftheseregressions,separatelyforentrepreneursandtheself‐em‐ployed, inTable11.24Our findingsaremixed.First,we findnorelationshipbetweenthenumberoflayersinthepreviousemployerandanagent'searningsfromentrepreneurshiporself‐employment.However, firmsize ispositivelyrelated toearnings, consistentwithabilitysorting.Second,wefindanegativeassociationbetweensurvivalandlayers,andbe‐tweensurvivalandfirmsize;theformerresultisconsistentwithbothsortingmechanisms,butthelatterisinconsistentwiththeabilitysorting.

Ofcourse,preferenceandabilitysortingarenotmutuallyexclusiveprocesses,soitisquitepossiblethattheevidenceconfoundscontributionsfrombothsortingmechanisms.Wecan,however,conductasharpertestoftheinformationprocessingmodel.Sofar,wehavecon‐trolledinourmultinomiallogitregressionsforrank,thenumberoflayersandfirmsize.Ofparticularnotehereisthepositiveeffectwefoundofhigherrankonthelikelihoodofbusi‐ness creation, andespeciallyof entrepreneurship.However, themeaningof aparticularrankdependsonthehierarchicalstructureofthefirm.Theinformationprocessingstoryimplies that themore layers there are in a firm the greater the difference between the

24Resultsreported intheappendixshowthat theseresultshold ifweexamineperformance fouryearsafterbusinesscreation.Spanofcontrolisnotincludedasthereisnoclearpredictionemanatingfromthisconstruct.However,includingitdoesnotchangeourresults.

28

breadthofknowledgeofemployeesinthehighestandlowestranks.Asaresult,thelikeli‐hoodof business creation of any rank relative to the lowest rank in the firm should begreaterthemoreranksthatliebetweenthem.

Table11Performanceofentrepreneurs,twoyearsafterbusinesscreation

OLSRegressions

Log(Totalincome) Survival Entrepreneurs Self‐employed Entrepreneurs Self‐employed (1) (2) (3) (4)Firmlayers:3 0.0102 ‐0.0521 ‐0.0236 ‐0.0429*** (0.0218) (0.0580) (0.0146) (0.00857)Firmlayers:4 0.00412 0.0197 ‐0.0394*** ‐0.0533*** (0.0218) (0.0537) (0.0143) (0.00828)CEOsandDirectors 0.147*** 0.834*** ‐0.0943*** ‐0.120*** (0.0445) (0.142) (0.0250) (0.0203)Managers 0.126*** 0.126 ‐0.0267 ‐0.0932*** (0.0292) (0.106) (0.0188) (0.0144)Supervisors 0.0973*** 0.113*** ‐0.00372 ‐0.0396*** (0.0202) (0.0409) (0.0131) (0.00614)FirmSize 2.26e‐06 1.36e‐05* ‐3.28e‐06 ‐3.50e‐06*** (4.30e‐06) (7.33e‐06) (2.41e‐06) (1.05e‐06)FirmSize^2 ‐0 ‐2.17e‐10 7.72e‐11* 6.79e‐11*** (8.36e‐11) (1.53e‐10) (0) (0)FirmAge 0.000200 0.00502** 0.000442 ‐0.000487 (0.00106) (0.00220) (0.000719) (0.000347)FirmAge^2 ‐3.37e‐06 ‐4.14e‐05** 9.19e‐07 6.62e‐06** (1.06e‐05) (2.07e‐05) (6.99e‐06) (3.32e‐06)Log(wage) 0.324*** 0.408*** 0.0500*** ‐0.00296 (0.0203) (0.0264) (0.0104) (0.00328)3yrwagegrowth ‐0.0617*** ‐0.175*** ‐0.0350*** ‐0.00937*** (0.0154) (0.0234) (0.00880) (0.00304)Observations 7,094 12,683 9,627 45,824Rsquared 0.206 0.086 0.046 0.021

Standarderrorsinparentheses.***p<0.01,**p<0.05,*p<0.1.Regressionsincludeaconstantterm,andcontrolsforage,agesquared,gender,education,tenure,tenuresquared,industry,region,andyear.Totalincomeisthesumofcapitalandlaborincomeaccruingtotheindividualtwoyearsafterbusinesscreation.Successisequaltooneifabusinesscreatedinyeartisactive,inthesensethatithasmorethanoneemployee(includingthefounder),twoyearsafterbusinesscreation.

29

EvidenceonthispredictionisprovidedinTable12,whichestimatestheeffectofrankonthe likelihoodofbusinesscreationseparately for firmswithdifferentnumbersof layers.Consistentwiththeinformationprocessingstory,wefindfirstthatforanygivennumberoflayersinafirmtheprobabilityofenteringentrepreneurshipisgreaterthehighertherela‐tiveposition.Moreover,thecoefficientsforentrepreneurshiponthetoplayersarelargerthemorelayersthereareinthefirm.Thisistrueforemployeesinthetopranksaswemovefromtwo‐layertothree‐layerfirms,andthentofour‐layerfirms;anditisalsotrueforem‐ployeesinthesecond‐highestrankaswemovefromtwotothreelayersinthefirm.Thesepatternsforentryintoentrepreneurshipsuggest,consistentwithCaliendoandRossi‐Hans‐berg(2012),thatthespanofknowledgeofmanagersisrelatedtothenumberoflayersbe‐neaththem,ratherthanthetitleoftheirrank.Thesepatternsarenotevidentforself‐em‐ployment,however,suggestingthat theinformationprocessingstoryisrelevantonlyforentryintoentrepreneurship.

4.Conclusions

Inthispaper,weaskedwhethersmallfirmsaremorefrequentincubatorsofentrepreneursbecausetheytendtobelesshierarchical.Thiscouldhelpimproveourunderstandingofwhobecomesanentrepreneur,howfirmscanbetterretainworkers,andwhatmightdrivecross‐firmandcross‐countrydifferencesinratesofentrepreneurship.Wefoundthathierarchyisindeedlessprevalentinsmallfirmsandisassociatedwithmorefrequenttransitionsofem‐ployeesintoself‐employmentandentrepreneurship.

Wethenexaminedtwopotentialmechanismsfortheimpactofhierarchyonbusinesscrea‐tionrates.First,individualswithapreferenceforentrepreneurshipmaychoosetoseekem‐ploymentinfirmswithfewerlayers(preferencesorting).Second,employeesinfirmswithfewerlayersmayhaveabroaderrangeofskills,andthismakesthemmorefitforentrepre‐neurship(abilitysorting).Wefoundevidenceforbothpreferencesortingandabilitysort‐ing.However,ourstrongestresult,atestshowingthattheprobabilityofbecominganen‐trepreneurclearlyriseswiththerankinanorganizationwasconsistentwithabilitysortingandinconsistentwithpreferencesorting.Nevertheless,thereremainsasubstantialunex‐plainedcomponentoftheeffectoflayersonbusinesscreationratesandwedonotclaimtohaveuncoveredacausalrelationship.Anappealingavenueforfuturetheoreticalandem‐piricalworkisstudyingmoreindetailthemechanismsbehindthishierarchyeffectandin‐vestigatingtowhatextentitiscausal.

OuranalysisisbasedontheSwedishmatchedemployer‐employeedataset.Conventionalwisdomhasitthat,alongwithotherScandinaviancountries,Sweden’slabormarketis

30

Table12WorkerTransitions,MultinomialLogitestimations

FOURLAYERS THREELAYERS TWOLAYERS

EntrepreneurshipSelf‐employ‐

ment Otherfirm EntrepreneurshipSelf‐employ‐

ment Otherfirm EntrepreneurshipSelf‐employ‐

ment Otherfirm

Top 1.434*** 0.265*** 0.0749 0.901*** 0.431*** 0.137*** 0.671*** 0.197*** ‐0.0391 (0.123) (0.0684) (0.0675) (0.0962) (0.0561) (0.0417) (0.0582) (0.0323) (0.0355)Second 1.059*** 0.301*** 0.127** 0.458*** 0.262*** 0.105*** (0.0757) (0.0431) (0.0495) (0.0750) (0.0361) (0.0370) Third 0.697*** 0.235*** ‐0.111 (0.0580) (0.0317) (0.0740) Spanofcontrol ‐0.000994 ‐8.84e‐05 0.00162 9.77e‐05 ‐0.00116** 0.000890 0.000870 0.000346 0.00143*** (0.000764) (0.000558) (0.00121) (0.000296) (0.000463) (0.000657) (0.000743) (0.000484) (0.000463)FirmSize ‐7.51e‐05*** ‐6.75e‐05*** ‐2.83e‐05 ‐0.000292*** ‐0.000198*** ‐2.44e‐05 ‐0.00134* ‐0.00165*** 0.00116** (1.26e‐05) (8.99e‐06) (1.82e‐05) (5.51e‐05) (4.16e‐05) (7.63e‐05) (0.000758) (0.000470) (0.000488)FirmSize^2 1.42e‐09*** 1.24e‐09*** 4.91e‐10 1.28e‐08*** 9.26e‐09*** ‐2.71e‐09 2.26e‐07* 2.58e‐07*** ‐6.08e‐07** (2.47e‐10) (1.81e‐10) (3.31e‐10) (3.47e‐09) (2.85e‐09) (4.70e‐09) (1.21e‐07) (7.73e‐08) (2.70e‐07)FirmAge 0.00398 0.000723 ‐0.00192 ‐0.00355 ‐0.00292 ‐0.00968*** ‐0.0203*** ‐0.0148*** ‐0.00345* (0.00385) (0.00259) (0.00503) (0.00397) (0.00225) (0.00321) (0.00335) (0.00184) (0.00207)FirmAge^2 ‐4.95e‐05 ‐2.09e‐05 ‐1.10e‐06 3.10e‐06 ‐1.65e‐07 4.56e‐05 0.000128*** 8.66e‐05*** ‐2.07e‐06 (3.56e‐05) (2.23e‐05) (4.16e‐05) (3.97e‐05) (2.21e‐05) (3.71e‐05) (3.40e‐05) (2.19e‐05) (4.13e‐05)Log(wage) 0.475*** ‐0.714*** ‐0.156*** 0.683*** ‐0.794*** ‐0.142*** 0.552*** ‐0.838*** ‐0.124*** (0.0536) (0.0149) (0.0227) (0.0699) (0.0204) (0.0231) (0.0687) (0.0173) (0.0173)3yr.wagegrowth

‐0.397*** 0.0577*** ‐0.0393** ‐0.358*** 0.0583*** ‐0.0362*** ‐0.414*** 0.0131 ‐0.0922***(0.0378) (0.0125) (0.0200) (0.0601) (0.0204) (0.0131) (0.0546) (0.0168) (0.00603)

Observations 12,882,472 2,342,313 1,260,742

Standarderrorsinparentheses.***p<0.01,**p<0.05,*p<0.1.Regressionsincludecontrolsforage,agesquared,female,education,tenure,tenuresquared,industry,yearandregion,anddummiesindicatingtruncationoffirmageandtenurevariables.

31

atypical.Thisconventionalwisdomisnowdated,25andmostemploymentdatarevealthatSwedeniscomparabletootherOECDcountries.WhilelabormobilityremainslowerthantheOECDaverage, itswagestructureisnowmuchlikeotherhigh‐incomecoun‐tries.Infact,LazearandShaw(2008),showthatacrossseveralmeasuressuchaswagedispersionwithinfirms,thevarianceofwagegrowthrateswithinfirms,andevenover‐allwagedispersion,Swedenisnotremarkablydifferentfromsevenothercountriestheystudy,includingtheU.S.A.TheenvironmentfornewfirmformationinSwedenisfurthernotmarkedlydifferentfromothercountriessuchastheU.S.A.,BrazilorDenmark(An‐dersson andKlepper, 2013).We thus see no reasonwhy results found in this paperwouldnotbereplicatedelsewhere.

ReferencesAnderson,PatriciaM.,andBruceD.Mayer(1994):“Theextentandconsequencesofjobturn‐

over.”BrookingsPapersonEconomicActivity:Microeconomics,177–236.

Andersson,Martin, andStevenKlepper (2013): ”CharacteristicsandperformanceofnewfirmsandspinoffsinSweden.”IndustrialandCorporateChange,22(1):245–80.

Arches, Joan(1991): “Social structure,burnout, and jobsatisfaction.”SocialWork,36(3):202‐206.

Åstebro,Thomas,JingChen,andPeterThompson(2011):“Starsandmisfits:Self‐employ‐mentandlabormarketfrictions.”ManagementScience,57(11):1999–2017.

Åstebro,Thomas,andPeterThompson(2011): “Entrepreneurs: Jacksofall tradesorho‐bos?”ResearchPolicy,40(5):637–649.

Benz,MatthiasandBrunoS.Frey(2008a):“Beingindependentisagreatthing:Subjectiveevaluationsofself‐employmentandhierarchy.”Economica,75(298):362–383.

Benz,MatthiasandBrunoS.Frey(2008b):“Thevalueofdoingwhatyoulike:Evidencefromthe self‐employed in 23 countries.” Journal of Economic Behavior&Organization,68(3–4):445–455.

Bernard, Andrew B., J. Bradford Jensen, Stephen J. Redding, and Peter K. Schott (2007):“Firmsininternationaltrade.”JournalofEconomicPerspectives,21(3):105–130.

Blanchflower, David (2000): “Self‐employment in OECD countries.” Labour Economics,7(5):471–505.

25Forthemajorityof the20thcentury,annualaveragewage increasesweresetby“collectivebargaining”betweenthethreemajorlaborunionsandthethreemajoremployers.However,be‐tween1970and1990centralizedcontroloverwageswasprogressivelydismantled,anditwasabandonedentirely in1997.Furtherderegulationintheearly1990sgaverisetoadual labormarketconsistingoftemporaryandpermanently‐employedworkers.Theseandotherchangesinduced an increase in the Gini coefficient for income from 0.199 in 1981 to 0.332 in 2007(http://www.scb.se/Pages/TableAndChart____163550.aspx).The2007figureplacesSwedenattheOECDaverage(TheEconomist,2013).

32

Blanchflower,DavidG.,andAndrewJ.Oswald(1998):“Whatmakesanentrepreneur?”Jour‐nalofLaborEconomics,16(1):26–60.

Caliendo,Lorenzo,FerdinandoMonte,andEstebanRossi‐Hansberg(2015):“TheanatomyofFrenchproductionhierarchies.”JournalofPoliticalEconomy,123(4):809‐85.

Caliendo,Lorenzo,andEstebanRossi‐Hansberg(2012):“Theimpactoftradeonorganiza‐tionandproductivity.”QuarterlyJournalofEconomics,127(3):1393–1467.

Carter,NancyM.,andThomasL.Keon(1986):“Theriseandfallofthedivisionoflabour,thePast25Years.”OrganizationStudies,7(1):54‐57.

Chen,Jing(2012):“Thesmallfirmeffectandthequalityofentrepreneurs.”Workingpaper,CopenhagenBusinessSchool.

Chen,Li‐Wei,andPeterThompson(2016):“SkillBalanceandEntrepreneurship.EvidencefromOnlineCareerHistories.”EntrepreneurshipTheoryandPractice,40(2):289‐305.

Dobrev,StanislavD.andWilliamP.Barnett(2005):“Organizationalrolesandtransitiontoentrepreneurship.”AcademyofManagementJournal,48(3):433–449.

Elfenbein,DanielW.,BartonH.Hamilton,andToddR.Zenger(2010):“Thesmallfirmeffectandtheentrepreneurialspawningofscientistsandengineers.”ManagementScience,56(4):659–681.

Finlay,William,JackK.Martin,PaulM.RomanandTerryC.Blum(1995):“Organizationalstructureandjobsatisfaction:Dobureaucraticorganizationsproducemoresatisfiedemployees?”Administration&Society,27(3):427‐450.

Friedrich,Benjamin(2015):“Tradeshocks,firmhierarchiesandwageinequality.”Man‐uscript,YaleUniversity.

Fuchs‐Schundeln,Nicola(2009):“Onpreferencesforbeingself‐employed.”JournalofEco‐nomicBehavior&Organization,71(2):162–171.

Garicano,Luis(2000):“Hierarchiesandtheorganizationofknowledgeinproduction.”Jour‐nalofPoliticalEconomy,108(5):874‐904.

Garicano, Luis, and Esteban Rossi‐Hansberg (2004): “Inequality and the organization ofknowledge.“AmericanEconomicReview,94(2):97‐202.

Gompers,Paul,JoshLerner,andDavidScharfstein(2005):“Entrepreneurialspawning:Pub‐liccorporationsandthegenesisofnewventures,1986to1999.”JournalofFinance,60(2):577–614.

Griffin,RickyW.,andGregoryMoorehead(2014):OrganizationalBehavior:ManagingPeopleandOrganizations,Mason,OH:CengageLearning.

Hackman,J.Richard,andGregR.Oldham(1976):“Motivationthroughthedesignofwork:Testofatheory.”OrganizationalBehaviorandHumanPerformance,16(2):250‐79.

Halaby,CharlesN.(2003):“Wherejobvaluescomefrom.Familyandschoolingbackground,cognitiveability,andgender.”AmericanSociologicalReview,68:251–278.

Haltiwanger,John,RonS.Jarmin,andJavierMiranda(2013):“Whocreatesjobs?Smallvs.largevs.young.”ReviewofEconomicsandStatistics,95(2):347–361.

Hamilton,Barton(2000):“Doesentrepreneurshippay?Anempiricalanalysisofthereturnstoself‐employment.”JournalofPoliticalEconomy,108(3):604–631.

Hausman,Jerry,andDanielMcFadden(1984):“Specificationtestsforthemultinomiallogit

33

model."Econometrica,52(5):1219–1240.

Henrekson,Magnus andTino Sanandaji (2014): "Billionaire entrepreneurs:A systematicanalysis."PNAS,ProceedingsoftheNationalAcademyofSciences.111(5):1760‐65.

Hummels,David,RasmusJørgensen,JakobMunch,andChongXiang(2014):“Thewageef‐fectsofoffshoring:EvidencefromDanishmatchedworker‐firmdata.”AmericanEco‐nomicReview,104(6):1597–1629.

Hundley, Greg (2001): “Why and when are the self‐employed more satisfied with theirwork?”IndustrialRelations,40(2):293–316.

Jovanovic,Boyan(1979):“Jobmatchingandthetheoryofturnover.”JournalofPoliticalEcon‐omy,87(5,Part1):972‐990.

Kacperczyk,AleksandraJ.(2012):“Opportunitystructuresinestablishedfirms:Entrepre‐neurship versus intrapreneurship in mutual funds.” Administrative Science Quar‐terly,57(3):484‐521.

Kawaguchi,D.(2008):“Self‐employmentrents:Evidencefromjobsatisfactionscores.”Hi‐totsubashiJournalofEconomics,49(1):35–45.

Lazear,EdwardP.(2005):”Entrepreneurship.”JournalofLaborEconomics,23(4):649‐680.

Lazear,EdwardP.andPaulOyer(2004):“Internalandexternallabormarkets:apersonneleconomicsapproach,”LabourEconomics,11(5):527‐554.

Lazear,EdwardP.,andKathrynL.Shaw,(2008)”“Introduction:Wagestructure,raisesandmobility: International comparisons of the structure of wages within and acrossfirms.”InE.LazearandK.Shaweds.,TheStructureofWages:AnInternationalCom‐parison,UniversityofChicagoPress.

Prottas,DavidJ.,andCynthiaA.Thompson(2006):“Stress,satisfaction,andthework‐familyinterface:Acomparisonofself‐employedbusinessowners,independents,andorgan‐izationalemployees.”JournalofOccupationalHealthPsychology,11(4):366–378.

Rousseau,DeniseM.(1977):“Technologicaldifferencesinjobcharacteristics,employeesat‐isfaction,andmotivation:Asynthesisofjobdesignresearchandsociotechnicalsys‐temstheory.”OrganizationalBehaviorandHumanPerformance,19(1):18‐42.

Sørensen,Jesper(2007):“Bureaucracyandentrepreneurship:Workplaceeffectsonentre‐preneurialentry.”AdministrativeScienceQuarterly,52:387–412.

Tåg, Joacim (2013): “Production hierarchies in Sweden.”EconomicsLetters,121(2):210‐213.

Taylor,MarkP.(1996):“Earnings,independenceorunemployment:Whybecomeself‐em‐ployed?”OxfordBulletinofEconomicsandStatistics,58(2):253‐266.

TheEconomist(2013):“Theinsandtheouts.Immigrationandgrowinginequalityaremak‐ingtheNordicslesshomogeneous.”TheEconomist,February2nd,2013.

Silva,Olmo(2007):"TheJack‐of‐all‐tradesentrepreneur: Innatetalentoracquiredskill?"EconomicsLetters,97(2):118–123.

Wagner,Joachim(2003):"TestingLazear'sJack‐of‐all‐tradesviewofentrepreneurshipwithGermanmicrodata."AppliedEconomicsLetters,10(11):687–689.

Wagner,Joachim(2004):“Areyoungandsmallfirmshothousesfornascententrepreneurs?”AppliedEconomicsQuarterly,50:379–391.

34

Wagner,Joachim(2006):"Arenascententrepreneurs'Jacks‐of‐all‐trades?'AtestofLazear'stheory of entrepreneurship with German data." Applied Economics, 38(20):2415–2419.

Walker,C.R.andR.Guest(1952):TheManontheAssemblyLine.Cambridge:MA:HarvardUniversityPress.

0

Appendixto

HierarchiesandEntrepreneurship

Notintendedforpublication

TableofContents

A.1  Summary Statistics of Sample .................................................................... 1 

A.2  Firm‐level Correlation Matrix ..................................................................... 3 

A.3  Worker‐level Correlation Matrix ................................................................ 3 

A.4  Layer‐by‐size Tabulation ............................................................................. 3 

A.5  90 % Accurate Occupation Data for Workers .............................................. 4 

A.6  95 % Accurate Occupation Data for Workers .............................................. 5 

A.7  99 % Accurate Occupation Data for Workers .............................................. 6 

A.8  Earnings Distributions Across Ranks for a) Narrower Classification of CEOs; and b) Total Income .............................................................................................. 7 

A.9  Transitions to Unemployment Included in Multinomial Logit ..................... 8 

A.10  Performance of Entrepreneurs Four Years after Business Creation ......... 9 

A.11  Sample Weighted Regression ................................................................ 10 

1

A.1 SummaryStatisticsofSample

TableA1SummaryStatisticsonSample

PanelA:FirmLevelComparison

BaseSample FinalSample

Mean Standarddeviation

Mean Standarddeviation

FirmSize 23.23 337.0 89.11 775.0

FirmAge 13.36 12.69 17.39 16.42

NumberofLayers ‐ ‐ 2.47 0.72

SpanofControl ‐ ‐ 6.18 17.46

Industry(%)

Manufacturing 12.13 16.79

WholesaleandRetail 23.21 23.71

RealEstate/Renting/Bus.Act 18.94 17.02

Education 2.33 3.81

HealthandSocialWork 3.45 4.34

Other 39.93 34.32

Location(%)

Stockholm 24.46 23.92

ÖstraMellansverige 14.70 14.66

SmålandmedÖarna 9.16 9.70

Sydsverige 13.90 14.05

Västsverige 19.97 20.03

NorraMellansverige 8.72 8.72

MellerstaNorrland 4.10 3.91

ÖvreNorrland 4.98 5.01

Firm‐yearobservations 1,368,388 243,689

Continuedonnextpage

2

Continuedfrompreviouspage

PanelB:IndividualLevelComparison

Basesample FinalSample

Mean Standard

deviation Mean Standard

deviation

Age 40.75 11.18 42.29 10.73

Tenure 4.991 5.009 6.020 5.200

Female 0.494 0.500 0.544 0.498

Education 3.533 1.129 3.650 1.152

Wage 260.0 180.7 272.4 166.0

WageGrowth 0.321 0.720 0.276 0.628

Transitions

Stay 86.33 90.26

Entrepreneurship 0.11 0.06

Self‐employment 0.57 0.28

OtherIncumbent 12.99 9.41

Worker‐yearobservations 26,399,697 16,485,527Notes.Thistabledisplayssummarystatisticsonthebasesampleofallfirmsandworkersinfirmswithmorethanoneemployeeandthefinalsampleweuseintheanalysis.Alldifferencesdisplayedinthistablearestatisticallysignificantatthe1%levelonaccountofthesamplesize.Firms inoursamplearenotarepresentativesampleof theentirepopulationof firms.Thereasonisthatoccupationdataisavailableprimarilyforlargerfirmsandhasgreatercoverageinsomeindustriesthanothers.Ourfindingsshould,however,berepresentativeoftheentirepopulationof largerfirms.Theoccupationdataiseithercollectedforall firmsaboveasizecutofforcollectedthroughrandomsamplingof firmsconditionalonfirmsizeandindustryaffiliation.Becauseourregressionscontrolforbothfirmsize,firmage,year,andindustry,weshouldobtaingoodrepresentationofthepopulationoflargerfirms.

3

A.2 Firm‐levelCorrelationMatrix

TableA2

Firm‐levelCorrelationMatrix.

Layers Span Size Age

Layers 1

Span 0.0393 1

Size 0.1959 0.1092 1

Age 0.2710 0.0512 0.0795 1

Notes.N=243,689.

A.3 Worker‐levelCorrelationMatrix

TableA3Worker‐levelcorrelationmatrix

Age Tenure Female Education Wage WageGr.Age 1 Tenure 0.4409 1 Female 0.4409 0.0705 1 Education ‐0.0623 ‐0.0515 0.1084 1 Wage 0.1659 0.0528 ‐0.2904 0.2123 1 WageGr. ‐0.3373 ‐0.3208 0.0047 0.0533 0.0050 1Notes.N=16,485,527.

A.4 Layer‐by‐sizeTabulation

TableA4NumberofFirm‐yearObservations

SizeCategories

<50 50‐100 100‐500 500‐1500 >1500 Total

2Layers 157,403 3,345 104 55 18 161.861

3Layers 38,126 5,855 4,312 581 169 49.043

4Layers 12,736 6,463 8,219 3,096 2,271 32.785

Total 208,265 15,663 13,571 3,732 2,458 243.689Notes.Thistablereportsthenumberoffirm‐yearobservationswithineachlayerandsizebin.

4

A.5 90%AccurateOccupationDataforWorkers

TableA5WorkerTransitions.MultinomialLogitEstimation

Entrepreneurship Self‐employment Otherfirm (1) (2) (3) Firmlayers:3 ‐0.920*** ‐0.730*** ‐0.295*** (0.0636) (0.0399) (0.0696) Firmlayers:4 ‐1.268*** ‐0.936*** ‐0.369*** (0.0639) (0.0448) (0.0767) Spanofcontrol 1.32e‐05 ‐0.000577 0.00185*** (0.000435) (0.000353) (0.000662) CEOsandDirectors 1.324*** 0.206*** 0.0469 (0.0855) (0.0598) (0.0798) Seniorstaff 0.981*** 0.223*** 0.198*** (0.0737) (0.0450) (0.0633) Supervisors 0.561*** 0.189*** ‐0.0449 (0.0501) (0.0283) (0.0927)

Notes.Thistablereportsregressionsthatrestrictattentiontofirmswith90%accurateoccupationdata.N=11,482,968.Standarderrorsinparentheses.***p<0.01,**p<0.05,*p<0.1.Regressionsincludeindustry,region,andyearfixedeffects;anddummiesindicat‐ingwhetherdataonfirmageandemployeetenurearetruncated.Thestandarderrorsareclusteredatthefirmlevel.Additionalregressors,notreported,arethesameasinTable6.Thestandarderrorsareclusteredatthefirmlevel.

5

A.6 95%AccurateOccupationDataforWorkers

TableA6WorkerTransitions.MultinomialLogitEstimation

Entrepreneurship Self‐employment Otherfirm (1) (2) (3) Firmlayers:2 ‐1.025*** ‐0.768*** ‐0.247*** (0.0827) (0.0556) (0.0869) Firmlayers:3 ‐1.383*** ‐1.035*** ‐0.378*** (0.0778) (0.0618) (0.0763) Spanofcontrol ‐5.67e‐05 ‐0.00120* 0.00246** (0.000686) (0.000619) (0.00103) CEOsandDirectors 1.174*** 0.200*** ‐0.0246 (0.102) (0.0718) (0.0500) Seniorstaff 0.823*** 0.216*** 0.319*** (0.105) (0.0700) (0.0866) Supervisors 0.463*** 0.185*** 0.200*** (0.0701) (0.0362) (0.0624)

Notes.Thistablereportsregressionsthatrestrictattentiontofirmswith95%accurateoccupationdata.N=4,324,587.Standarderrorsinparentheses.***p<0.01,**p<0.05,*p<0.1.Regressionsincludeindustry,region,andyearfixedeffects;anddummiesindicat‐ingwhetherdataonfirmageandemployeetenurearetruncated.Thestandarderrorsareclusteredatthefirmlevel.Additionalregressors,notreported,arethesameasinTable6.Thestandarderrorsareclusteredatthefirmlevel.

6

A.7 99%AccurateOccupationDataforWorkers

TableA7WorkerTransitions.MultinomialLogitEstimation

Entrepreneurship Self‐employment Otherfirm (1) (2) (3) Firmlayers:2 ‐0.546*** ‐0.393*** 0.0509 (0.0909) (0.0576) (0.0683) Firmlayers:3 ‐1.128*** ‐0.541*** ‐0.190 (0.137) (0.0869) (0.136) Spanofcontrol ‐0.0404*** ‐0.00959** 0.00618*** (0.00715) (0.00417) (0.000916) CEOsandDirectors 0.980*** 0.218*** ‐0.229*** (0.115) (0.0800) (0.0391) Seniorstaff 0.872*** 0.416*** 0.207*** (0.166) (0.121) (0.0566) Supervisors 0.395*** 0.192*** 0.0631* (0.0842) (0.0461) (0.0356)

Notes.Thistablereportsregressionsthatrestrictattentiontofirmswith99%accurateoccupationdata.N=640,627.Standarderrors inparentheses.***p<0.01,**p<0.05,*p<0.1.Regressionsincludeindustry,region,andyearfixedeffects;anddummiesindicat‐ingwhetherdataonfirmageandemployeetenurearetruncated.Thestandarderrorsareclusteredatthefirmlevel.Additionalregressors,notreported,arethesameasinTable6.Thestandarderrorsareclusteredatthefirmlevel.

7

A.8 EarningsDistributionsAcrossRanksfora)NarrowerClassificationofCEOs;andb)TotalIncome

TableA8a

WagedistributionforNarrowerClassificationofCEOsandDirectors

Rank MeanPercentiles Wage

Dispersion10th 25th 50th 75th 90th

4.CEOsanddirectors 1,011 322 390 536 763 1,132 1,755

3.Seniorstaff 505 293 359 444 575 761 678

4.Supervisors 308 159 228 285 359 469 453

1.Productionworkers 219 112 172 222 266 313 264

Mean 272 131 195 251 317 415 373Notes.ThistabledisplaysthewagedistributionacrossoccupationclassesbasedonSSYKbutwith“Rank3.CEOsanddirectors”correspondingonlytoworkerswithSSYKcode121(Directorsandchiefexecutives).Dataareinunitsof1,000SEKin2005prices.Samplesizesareasfollows.Rank4:45,868.Rank3:607,753.Rank2:7,302,636.Rank1:8,376,591.

TableA8b

Totalincome(wage+capitalincome)distribution

Rank MeanPercentiles Wage

Dispersion10th 25th 50th 75th 90th

4.CEOsanddirectors 585 204 271 357 572 985 1,265

3.Seniorstaff 509 276 342 430 568 774 737

2.Supervisors 304 148 217 276 352 469 467

1.Productionworkers 213 106 165 214 258 307 267

Mean 269 123 186 242 309 412 389Notes.ThistabledisplaysthedistributionoftotalearningsacrossbroadoccupationclassesbasedonSSYK.

8

A.9 TransitionstoUnemploymentIncludedinMultinomialLogit

TableA9

WorkerTransitions.MultinomialLogitEstimation

Entrepreneur‐ship

Self‐employ‐ment

Otherfirm Unemployment

(1) (2) (3) (4)Firmlayers:2 ‐0.725*** ‐0.535*** ‐0.134*** ‐0.0689*** (0.0455) (0.0272) (0.0450) (0.0248)Firmlayers:3 ‐1.073*** ‐0.718*** ‐0.244*** ‐0.127*** (0.0494) (0.0337) (0.0478) (0.0310)Spanofcontrol ‐0.000358 ‐0.000951*** 0.00112*** ‐8.49e‐05 (0.000412) (0.000288) (0.000415) (0.000318)CEOsandDirectors 1.410*** 0.368*** 0.0356 ‐0.478*** (0.0726) (0.0486) (0.0491) (0.0454)Seniorstaff 0.978*** 0.366*** 0.177*** ‐0.759*** (0.0550) (0.0348) (0.0353) (0.0406)Supervisors 0.605*** 0.249*** ‐0.0179 ‐0.399*** (0.0387) (0.0238) (0.0527) (0.0280)

Notes.Thistableincludedmovestounemploymentasanoutcome.Thereare202,136workersthantransitiontounemployment,sosamplesizeincreasestoN=16,687,663.Standarderrorsinparentheses.***p<0.01,**p<0.05,*p<0.1.Regressionsincludeindus‐try,region,andyearfixedeffects;anddummiesindicatingwhetherdataonfirmageandemployeetenurearetruncated.Thestandarderrorsareclusteredatthefirmlevel.Addi‐tionalregressors,notreported,arethesameasinTable6.Thestandarderrorsareclus‐teredatthefirmlevel.

9

A.10 PerformanceofEntrepreneursFourYearsafterBusinessCreation

TableA10Performanceofentrepreneurs,fouryearsafterbusinesscreation

OLSRegressions

Log(Totalincome) Survival Entrepreneurs Self‐employed Entrepreneurs Self‐employed (1) (2) (3) (4)Firmlayers:2 ‐0.00835 0.0945 ‐0.0201 ‐0.0269*** (0.0299) (0.0778) (0.0203) (0.00973)Firmlayers:3 0.0116 0.0733 ‐0.0228 ‐0.0502*** (0.0285) (0.0722) (0.0194) (0.00908)CEOsandDirectors 0.121** 0.718*** ‐0.0936*** ‐0.154*** (0.0528) (0.219) (0.0287) (0.0210)Managers 0.165*** 0.197 ‐0.0544** ‐0.0972*** (0.0385) (0.139) (0.0263) (0.0159)Supervisors 0.118*** 0.156*** ‐0.0291 ‐0.0290*** (0.0274) (0.0592) (0.0178) (0.00713)FirmSize ‐4.31e‐06 7.60e‐06 ‐6.08e‐06** ‐3.26e‐06*** (4.91e‐06) (8.31e‐06) (2.82e‐06) (1.10e‐06)FirmSize^2 2.05e‐10* ‐2.15e‐10 1.18e‐10** 5.97e‐11*** (1.15e‐10) (1.94e‐10) (5.70e‐11) (0)FirmAge 0.000182 0.00243 0.00143 ‐0.000282 (0.00142) (0.00302) (0.000988) (0.000391)FirmAge^2 4.83e‐06 ‐1.82e‐05 ‐1.15e‐05 5.74e‐06 (1.47e‐05) (2.94e‐05) (1.01e‐05) (3.96e‐06)Log(wage) 0.290*** 0.370*** 0.0405*** ‐0.0138*** (0.0293) (0.0329) (0.0132) (0.00377)3yrwagegrowth ‐0.0752*** ‐0.114*** ‐0.0173 ‐0.00168 (0.0213) (0.0313) (0.0113) (0.00346)Observations 4,057 7,204 6,546 34,465Rsquared 0.174 0.062 0.040 0.017

Notes.Thistablepresentsresultsofregressionsexplainingtheperformanceofentrepreneursfor four years after business creation. Standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.1.Regressions includea constant term,andcontrols forage,agesquared,gender,education,tenure,tenuresquared,industry,region,andyear.Totalincomeisthesumofcapitalandlaborin‐comeaccruingtotheindividualfouryearsafterbusinesscreation.Successisequaltooneifabusi‐nesscreatedisactive,inthesensethatithasmorethanoneemployee(includingthefounder),fouryearsafterbusinesscreation.Thestandarderrorsareclusteredatthefirmlevel.

10

A.11 SampleWeightedRegression TableA11

WorkerTransitions.MultinomialLogitEstimation

Entrepreneurship Self‐employment Otherfirm (1) (2) (3) Firmlayers:2 ‐0.677*** ‐0.553*** ‐0.129*** (0.0489) (0.0284) (0.0325) Firmlayers:3 ‐1.027*** ‐0.721*** ‐0.263*** (0.0502) (0.0311) (0.0357) CEOsandDirectors 1.645*** 0.350*** ‐0.0728** (0.0772) (0.0577) (0.0350) Seniorstaff 1.157*** 0.251*** 0.118*** (0.0713) (0.0377) (0.0253) Supervisors 0.670*** 0.198*** ‐0.0289 (0.0492) (0.0274) (0.0368) Firmsize ‐0.000107*** ‐8.35e‐05*** ‐2.77e‐05 (1.25e‐05) (8.85e‐06) (1.83e‐05) Firmsizesquared 2.05e‐09*** 1.55e‐09*** 5.00e‐10 (2.47e‐10) (1.77e‐10) (3.36e‐10) Firmage ‐0.00303 ‐0.00449*** ‐0.00247 (0.00239) (0.00153) (0.00240) Firmagesquared 2.54e‐07 1.03e‐05 1.11e‐06 (2.21e‐05) (1.41e‐05) (2.17e‐05) Age(years) 0.136*** 0.125*** ‐0.0497*** (0.0127) (0.00550) (0.00272) Agesquared ‐0.00175*** ‐0.00132*** 0.000365*** (0.000147) (6.70e‐05) (3.37e‐05) Female ‐0.961*** ‐0.741*** ‐0.0927*** (0.0463) (0.0193) (0.0135) Education 0.0503*** 0.0655*** 0.0803*** (0.0167) (0.00949) (0.00720) Log(laborincome) 0.204** ‐0.623*** ‐0.127*** (0.0849) (0.0122) (0.0100) Tenure(years) ‐0.122*** ‐0.206*** ‐0.187*** (0.00963) (0.00593) (0.0112) Tenuresquared 0.00289*** 0.00755*** 0.00764*** (0.000667) (0.000385) (0.000548) 3yrwagegrowth ‐0.0498 ‐0.0484 ‐0.382*** (0.0620) (0.0354) (0.0743)

Notes.ThistablereplicatesTable7butusessampleweightstoaccountforthedifferencesbetweenthefinalsampleandthebasesampledisplayedinTableA1.Thesampleweightsarebasedonalogisticregressionexplainingbeinginthefinalsamplewithallthevaria‐blesdisplayedinTableA1.N=16,485,527.Standarderrorsinparentheses.***p<0.01,**p<0.05,*p<0.1.Regressionsincludeindustry,region,andyearfixedeffects,anddummiesindicatingwhetherfirmageandemployeetenurearetruncated.