ContentslistsavailableatScienceDirect
Journal of Economic Behavior & Organization
jo u r n al h om ep age :w w w . e l s e v i e r . c o m / l o c a t e / j e b o
The gender gap in entrepreneurship – The role of peer effects 夽
Simen Markussen, Knut Røed
∗TheRagnarFrischCentreforEconomicResearch,Oslo,Norway
a r t i c l e i n f o
Articlehistory:
Received23March2016
Receivedinrevisedform9December2016 Accepted15December2016
Availableonline16January2017
JELclassification:
L26 M13 J16
Keywords:
Earlycareerentrepreneurship Peereffects
Gendergap
Instrumentalvariables
a b s t r a c t
Invirtuallyallindustrializedcountries,womenareunderrepresentedinentrepreneurship, andthegendergapexhibitsaremarkablepersistence.Weexamineoneparticularsource ofpersistence,namelytheprevalenceofgenderedpeerinfluences.Westudyhowearly careerentrepreneurshipisaffectedbyexistingentrepreneurshipamongneighbors,family members,andrecentschoolmates.Basedonaninstrumentalvariablesstrategy,weiden- tifystrongandheavilygenderedpeereffects.Whilemenaremoreinfluencedbyother men,womenaremoreinfluencedbyotherwomen.Weestimatethatdifferencesbetween maleandfemalepeergroupsexplainapproximatelyhalfofthegendergapinearlycareer entrepreneurship.
©2017ElsevierB.V.Allrightsreserved.
1. Introduction
Anentrepreneurisapersonwhoseekstocreatehis/herownworkplace,andpotentiallyalsogeneratesworkplacesfor others.Inallindustrializedeconomies,thereareconsiderablyfewerfemalethanmaleentrepreneurs(Kelleyetal.,2012).The existingliteratureoffersnogenerallyacceptedexplanationforthisgendergap.Itsuniversalitypointstowardfundamental genderdifferencesrelatedto,e.g.,riskaversion(JianakoplosandBernasek,1998;Byrnesetal.,1999;CrosonandogGneezy, 2009;Borghansetal.,2009),inattitudestowardcompetition(NiederleandVesterlund,2007;Gneezyetal.,2003;Bönteand Piegeler,2013),orinthesubjectiveperceptionofowncapabilities(LangowitzandMinniti,2007).However,althoughthere issomeevidenceindicatingageneticcomponentinthesedifferences(Sapienzaetal.,2009),itappearsprobablethatthey tosomeextentareculturallyinherited,andthereforewilldiminishovertimeastraditionalgenderrolesaremoderated.
Yet,itisnotgenerallythecasethatthegendergapinentrepreneurshipisparticularlysmallinlabormarketsconsideredto havecomefarintermsofgenderequality.Thecountrythatwestudyinthepresentpaper–Norway–illustratesthispoint.
Intermsoflaborforceparticipation,Norwayisoneofthemostgender-equalsocietiesintheworld:48%oftheactivelabor marketparticipantsarefemale.Intermsofentrepreneurship,itisoneofthemostgender-unequalsocieties:Only25%of
夽 Thispaperispartoftheproject“Entrepreneurship,Gender,andSocialCapital”financedbytheNorwegianResearchCouncil(researchgrantno.201336).
ThankstoRolfGolombekandseminarparticipantsatIFAU,Uppsala,forcommentsanddiscussion.Thanksalsototwoanonymousrefereesandanassociate editorforconstructivecriticismandanumberofusefulsuggestions.
∗ Correspondingauthor.
E-mailaddress:[email protected](K.Røed).
http://dx.doi.org/10.1016/j.jebo.2016.12.013 0167-2681/©2017ElsevierB.V.Allrightsreserved.
theentrepreneursarefemale(Berglannetal.,2011).Andonlyasmallfractionofthisgendergapcanbeaccountedforby observedindividualcharacteristics,suchaseducationandindustry(Berglannetal.,2013).
Onepossibleexplanationforthelargeandpersistentgendergapsinentrepreneurshipactivitiesisthatthehistorically inheritedmaledominanceinthisareaispreservedthroughgenderedpeerinfluences.Peerinfluencesmayoperateintwo forms.First,peersmayactasrolemodels;see,e.g.,Gibson(2004),VanAukenetal.(2006),andBarNiretal.(2011).People mightfindentrepreneurshipamoreattractiveandrealisticoccupationalchoicewhentheyseeotherentrepreneursintheir socialnetwork.Second,peersmayprovidelearningopportunitiesandaccesstoimportantnetworks.Inordertostartanew business,onemayneedinformants,customers,and(maybe)investors.Entrepreneurialpeerscandelivertheseintangible
“services”.Inaddition,theycanmotivateandeducateforentrepreneurshipduringtheformativeadolescenceyears(Guiso etal.,2015).EmpiricalevidenceprovidedbyVerheuletal.(2012),basedonsurveydatafrom29differentcountries,indicates thatthegendergapinentrepreneurshipismoreaboutthecognitivestageof“wantingit”thanaboutthebehavioralstageof
“doingit”.Ithaslongbeenrecognizedthatrolemodelsplayanimportantroleatthe“wantingstage”,andperhapsparticularly soforwomen;see,e.g.,HisrichandBrush(1984).Basedonastudyofasampleof393undergraduatestudentsfromthe U.S.,BarNiretal.(2011)presentevidencethatrolemodelshaveaparticularlylargeeffectonentrepreneurialself-efficacy forwomen,whichagainaffectsentrepreneurialintentionspositively.
Recentempiricalevidencealsosuggeststhatpeopletendtolookforsame-sexedrolemodels;see,e.g.,Ruefetal.(2003) andBosmaetal.(2012).Throughinterviewswith292entrepreneursintheNetherlands,Bosmaetal.(2012)establishes thatthemajorityhavearolemodelinthepre-and/orpost-start-upphase,andthatsame-sexmodelsarestrongly(and statisticallysignificantly)overrepresentedcomparedtowhatrandomassignmentwouldimply.Theyalsofindthatalmost noneoftheentrepreneursconsideradistantandfamousentrepreneurastheirrolemodel.Hence,theoverrepresentation ofmenamongexistingentrepreneursimpliesthatmenalsohavemoresame-sexedentrepreneurialrolemodelstoinspire newentrepreneurshipattempts.
Thereisalreadyanexistingliteratureindicatingthatspatialvariationsinentrepreneurshipareextremelypersistentover time.FritschandWyrwich(2014),forexample,showthatself-employmentratesobservedinGermanregionsasfarbackas in1925arerobustpredictorsfortheregionalpatternsofentrepreneurshiptoday.AndrecentstudiesfromSweden(Giannetti andSimonov,2009;AnderssonandLarsson,2016)andDenmark(NandaandSørensen,2008)indicateaprominentrolefor peerinfluencesinexplainingspatialpersistenceinentrepreneurship:Thehigheristheentrepreneurshipactivityamong neighborsorcolleagues,thehigheristheprobabilitythatyetanotherpersonembarksonentrepreneurship,ceterisparibus.
Minniti(2005)arguesthatentrepreneurshipcreatesa“culture”ofitselfthatinfluencesindividualbehaviorinitsfavor.
Existingevidencealsoindicatesthatpersonalnetworksmayhavealargerinfluenceonentrepreneurshipbehaviorinsmall communitiesthaninlargeones;seeBauernschusteretal.(2010).Thismaypotentiallyexplainwhygender-differences appeartobeparticularlypersistentinasparselypopulatedandhighlydecentralizedcountrylikeNorway.
Theresearchquestionweaddressinthepresentpaperiswhether–andtowhichextent–genderedpeerinfluences alsocanexplainthepersistenceofthegendergapinentrepreneurship.Toidentifyandestimatepeereffectsisknownto beachallengingmethodologicalproblem;see,e.g.,Angrist(2013)forcriticaldiscussion.Anumberofconfoundingfactors mayexist,suchasendogenousgeographicalmigrationandunobservedlocalvariationsinindustry-composition.Moreover, whenconsideringhowagroup’saggregatebehaviorinfluencesthebehaviorofitsindividualmembers,thereiswhatManski (1993)labelledareflectionproblem:Itisdifficulttodisentanglethegroup’seffectsonitsindividualmembersfromthefact thatthegroup’sbehaviorisamechanicalreflectionofitsmembers’behavior.
OuranalysisisbasedonadministrativeregisterdatafromNorwaywithpopulation-wideannualinformation about individuallabormarketstatesfrom2002through2012.We examinepeerinfluencesonearlycareerentrepreneurship withinnetworksconfinedtoneighborhoods,families,andschoolmates.Ouranalysispopulationconsistsoflabormarket entrants,whichwefollowforupto10yearsafterentry.Weinvestigatehowtheiroccupationalchoices–intermsofregular employmentorentrepreneurship–areaffectedbythecorrespondingchoicesalreadymadebytheirolderpeers.Inthis partoftheexercise,thepeers’behaviorsarestrictlypre-determined,andcan,withappropriatecontrols,beinterpretedas exogenous.Inaddition,weexaminehowtheirownoccupationalchoicesareaffectedbythoseoftheirschoolmatesorfellow students(hereafterreferredtoasschoolmates).Thesechoicesaretosomeextentmadesimultaneously,andpeerinfluences canrunbothways.Wedealwiththisandtheassociatedreflectionproblembyusingthepre-determinedentrepreneurship activityamongtheschoolmates’parentsasinstruments.Inthisexercise,weexploitarecentfindingreportedbyLindquist etal.(2015)thattheintergenerationaltransmissionofentrepreneurshippropensityisheavilygendered:Mothersinfluence daughterswhereasfathersinfluencesons.Thisclearlyalsorepresentsachannelformakingthegendergappersistent.
Ourfindingsconsistentlyconfirmtheimportanceofpeereffectsatalllevels.Theincidenceofanearlycareerentrepreneur- shipendeavorisinfluencedbyexistingentrepreneurshipratesinthelocalcommunity,thefamily,andinthegroupofrecent schoolmates.Closefamilyhasalargerinfluencethanmoredistantfamily.Closeneighborshavealargerinfluencethanmore distantneighbors.Andimportantly:same-sexpeersgenerallyhavelargerinfluencethanopposite-sexpeers.Thelatter impliesthatmenhavemuchmoreentrepreneursintheirmostinfluentialpeergroupsthanwomenhave.Weestimatethat thisexplainsapproximately50%ofthegendergapinearlycareerentrepreneurship.Thestatisticaluncertaintyisconsider- able,however,anda90%confidenceintervalonthefractionofthegendergapthatisaccountedforbydifferencesbetween maleandfemalepeergroupsrangesfrom21to81%.
Fig.1.Entrepreneurshipratesbygender2002–2012.
Note:Theconditionalentrepreneurshiprateinpanel(a)isdefinedasthenumberofpersonsaged18–66engagedinanyformofentrepreneurialactivity (incorporatedorunincorporated)dividedbythetotalnumberofeconomicallyactivepersons(employeesandentrepreneurs)inthesameagegroup.The unconditionalrateinpanel(b)isdefinedasthesamenumberofpersonswithentrepreneurialactivitydividedbythetotalpopulationaged18–66.
2. Backgroundanddata
Thefoundationforouranalysisis(encrypted)administrativeregisterdatafromNorway,combiningemployer-employee registerswithinformationonearningsandbusinessincomeandfirmownership.Foreachyear2002–2012,weusethesedata toidentifyregularemploymentandentrepreneurshipactivities.Ourentrepreneurshipdefinitionisconsiderablywiderthan theself-employmentconceptoftenencounteredintheeconomicsliterature,asitalsoincludespersonswhoareemployed inlimitedliabilityfirmsinwhichtheyhavealargeownershipshare(morethan30%),eitherdirectlyorindirectlythrough otherfirms;seeBerglannetal.(2011)fordetails.WeextendtheBerglannetal.(2011)definitionsomewhat,however,by alsoincludingpersonswhohaveregularemploymentastheirmainsourceofpersonalincome,yetstilloperateanactive businessasself-employed(regardlessitssizeandprofitability).1Themotivationbehindthisextensionisthatwewishto capturenascententrepreneurshipandentrepreneurshipendeavorswithoutimplicitlyconditioningoneconomicsuccess.
Fig.1showsformenandwomen,respectively,annualentrepreneurshipratesinNorwayfrom2002to2012.Inpanel (a),theratesareconditionaloneconomicself-sufficiency,definedashavingannualearningsfromemploymentand/or entrepreneurshipexceedingasubsistencelevelofNOK180,000(approximately$21,200).2Inpanel(b),theyareuncon- ditional.Bothconditionalandunconditionalentrepreneurshipratesareapproximatelythreetimesashighformenasfor women.Therehas,however,beenaconvergenceduringtheperiodwelookat,withslightlyincreasingfemaleentrepreneur- shipratesandslightlydecreasingmaleentrepreneurshiprates.
Thereareconsiderablegeographicaldifferencesintheentrepreneurshiprate,andthesedifferencesarehighlypersistent overtime.ThisisillustratedinFig.2,whereweplotgender-specificentrepreneurshipratesbytravel-to-workarea(TWA) in2002againstthecorrespondingratesin2012.3ThecirclesizesinFig.2areproportionaltothenumberofinhabitantsin eachTWA.ItisclearthatthegeographicaldistributionofentrepreneurshipinNorwaywasvirtuallyunchangedoverthis 10yearperiod,andthatthepositiveshiftsinfemaleentrepreneurshiprateshavebeenofsimilarmagnitudeinallpartsof thecountry.
1 Wedefinean“activebusiness”asabusinesswithatleastsomerecordedeconomicactivityduringtheyearinthesensethatassociatedearningsare strictlynon-zero.
2 Thisthresholdcorrespondstoapproximatelyonethirdofaveragefull-year-full-timeearningsinNorway.Monetaryamountsreportedinthispaper areinflatedto2016-value,andNOKisconvertedto$basedontheexchangerateapplyinginMarch2016($1=NOK8.5).
3 Weuseapartitionwith46suchregionsinNorway,withapproximately110,000inhabitantsonaverage;seeBhuller(2009).
Fig.2.Entrepreneurshipratesbytravel-to-workarea(TWA).2002and2012.
Note:CirclesizesareproportionaltoTWAsize(averagenumberofinhabitantsoverthetwoinvolvedyears).Thelinesarethe45◦lines.
Ourdatacontainrichinformationonfamilylinkages,education(includingschoolidentity,type/level,andgraduation year),placesofresidence(atthelevelofsmallneighborhoods),nationality,anddemographiccharacteristics.Thesedimen- sionofthedatawillbeusedtoestablishtheindividuallyassignedpeergroupsthatpotentiallyplayaroleinencouraging ordiscouragingentrepreneurshipendeavors.Providedthatsomeemploymentorentrepreneurshipactivityisrecorded,the dataalsocontaininformationaboutthechosenindustry.
3. Empiricalapproach
Thestartingpointofourempiricalanalysisisthegroupofpersonswhocompletedtheireducationin2001–2007.We interpretaneducationalcareerascompletedinagivensemesterifapersonwasregisteredasapupil/studentthatsemester, butnotinanyofthefollowingsixsemesters.4Werefertotheyearofcompletionasthegraduationyear,irrespectiveof whetheragradewasobtainedornot.Wecollectinformationaboutsubsequentlabormarketstatesandconstructannual entrepreneurshipindicatorsforeachyearafterthegraduationyearanduntil2012.Hence,fortheseindividualswehave panelsof5–11consecutiveoutcomeobservations(dependingongraduationyear),eachindicatingentrepreneurialactivity.
Thefocusonearlycareerentrepreneurshipclearlyentailsthelimitationthatwemissoutontheentrepreneurshipactivities occurringlaterinthelifecycle.Berglannetal.(2011)showthattheentrepreneurshiprateamongNorwegianstendstogrow withage,andthatentrepreneurshipactivitiesareroughlytwiceascommoninthemid40’scomparedtothemid20’s.
However,bystudyinglabormarketentrants,weensurethatwemodeloccupationalchoicesfromtheverystartofthelabor marketcareer,atwhichpointtheyarenotgovernedbythepersistenceofpreviouslychosenstates,whereastheirolder peers’entrepreneurshipbehaviorscansafelybeconsideredexogenous.Thiswayweensurethatwhilethemembersofour analysispopulationmayhavebeenaffectedbyongoingentrepreneurshipactivitiesintheirlocalcommunities,theyhavenot yetbeenabletoinfluencetheseactivitiesthemselves.Wethushaveahierarchicalmodel,wherebythe“old”mayaffectthe behaviorofthe“young”,butnotviceversa,andwecircumventthereflectionproblemdiscussedbyManski(1993).Ourfocus onearlycareerentrepreneurshipalsoimpliesthatwecanratherdirectlyexaminehowgenderpatternsinentrepreneurship are(orarenot)transferredacrossgenerations.Wehaveasimultaneityprobleminrelationtoonepotentiallyimportantpeer group,though,namelythatconsistingofschoolmates.Asweexplaininmoredetailbelow,wedealwiththisbyapplyingan instrumentalvariablesstrategy.
4Moreprecisely,werequirethataneducationlastingatleastsixmonthsendedandthatnoeducationlastingmorethanthreemonthswasrecordedthe nextthreeyears.
Table1
Descriptivestatisticsanalysissample.
Men Women
Numberofgraduates 133,714 119,585
Ageatgraduation 22.1 22.7
Educationallevel(%)
Primaryeducationoruncompletedsecondaryeducation 32.3 25.1
Secondaryeducation 44.8 35.2
College/University 22.9 39.7
Anyeconomicactivity(employmentorentrepreneurship)duringfirstfiveyears(%) 83.1 82.1
Anyentrepreneurshipactivityduringfirstfiveyears(%) 10.5 5.7
Fig.3.Unconditionalratesofregularemploymentandentrepreneurshipbyyearssincegraduation(2001–2002graduationcohorts).
Thewaywehaveconstructedthedataensuresthatallgraduationcohortscanbefollowedforatleastfiveyears.Inorder toexaminetheimpactofvariouspeergroups’influenceonownentrepreneurshipbehavior,wedefineasourmainoutcome variableanindicatorforatleastsomeentrepreneurshipactivitywithinthefirstfiveyearsaftertheyearofgraduation.We returntoalternativeoutcomeslateron,bothintheformofnarrowerentrepreneurshipdefinitions,e.g.,requiringthata completelynewfirmisestablishedorthatentrepreneurshipisthemaineconomicactivity,andintheformofayear-by-year analysiswhereweexploiteachgraduationcohortaslongasweareabletoobserveit.
Table1showssomedescriptivestatisticsforouranalysispopulation.Wefollowaround253,000schoolgraduatesfor fiveyearsormore.Duringthefirstfiveyears,10.5%ofthemen,and5.7%ofthewomenhasbeenengagedinsomeformof entrepreneurship.Hence,thegendergapatthisstageofthelabormarketcareeris4.8percentagepoints.
Fig.3presentsunconditionalemploymentandentrepreneurshippropensitiesbyyearssincegraduationforthe2001–2002 graduationcohorts.Wefocusonthesetwocohortsinthisparticulargraphforthereasonthattheycanbefollowedfora full10-yearperiod.Lookingatpanel(b)itisevidentthatentrepreneurshiprates,aswellasthegendergap,increaserather monotonicallywithyearssincegraduation.5
InFig.4,weplotthefractionswithatleastoneincidenceofearlycareerentrepreneurshipduringthefirstfiveyearsafter graduationagainsttheexistingsame-sexentrepreneurshiprates(inthegraduationyear)intheresidentialtravel-to-work
5 Notethattheconspicuouslyhighfemaleemploymentrateinthefirsttwoyearsaftergraduation(panel(a)),aswellasthesubsequentdrop,may beexplainedbythecombinationofagenerousparentalleaveschemeinNorwayproviding(almost)fullwagereplacementforayear,but(intheperiod coveredhere)onlyconditionalonatleastsixmonthsofregularemployment.
Fig.4.Gender-specificgraduation-yearentrepreneurshiprateintravel-to-work-area(TWA)andfractionwithearlycareerentrepreneurshipduringfirst fiveyearsaftergraduation.
area(TWA).Again,aremarkablepatternofpersistenceemerges.Earlycareerentrepreneurshipishigherthehigherthelocal rateofsame-sexentrepreneurshipistostartwith.
Thedescriptivepatternspresentedsofardoofcoursenotnecessarilyreflectpeereffects.Theymayalsoreflectother sourcesofgeographicalvariationsineducational/occupationalchoicesand/orindustrycomposition.Toisolateandestimate thepeereffects,wesetupastatisticalmodeldesignedtoeliminatepotentiallyconfoundingfactors.
LetEntibeouroutcomevariableforindividuali,whichinthemainpartofouranalysisisequalto1ifsomeentrepreneur- shipactivityhasbeenrecorded(eitherasamainactivityorasoneofmultipleactivities)withinfiveyearsafterschool completion,and0otherwise.Wethensetuplinearprobabilitymodelofthefollowingformseparatelyformenandwomen:
Enti=
k
mkemki+fkefki+controls+εi. (1)
Theright-hand-sidevariablesofinterestaretheindicatorsforentrepreneurshipbehaviorinthepeergroupsrelevantfor personi,denotedemkiandefki,wherethesubscriptkindicatesthetypeofpeergroupandthesubscripts(m,f)distinguish malesfromfemales.Weusepeergroupsofthreedifferenttypes:Neighbors,family,andschoolmates.Thegroupsareinall casesdefinedsuchthattheyexcludethereferenceperson.Thegroupsandtheirassociatedindicatorsaredefinedasfollows:
Neighbors:Wedistinguishbetweencloseanddistantneighbors,withbothgroupsidentifiedonthebasisofresiden- tialaddressesintheyearofgraduation.By“closeneighbors”,wemeanpersonslivinginthesame“basicstatisticalunit”
(“grunnkrets”)asdefinedbyStatisticsNorway.Thesearedesignedtoresemblegenuineneighborhoodswhereresidentsare likelytointeract.6Thereare13,700basicstatisticalunitsinNorway,eachpopulatedbyaround350individualsonaverage.
By“distantneighbors”,wemeanpersonslivinginadjacentneighborhoodsbelongingtothesame“statisticaltract”(“delom- råder”).ThesearealsodrawnupbyStatisticsNorway,andaredesignedtoencompassneighborhoodsthatsharecommon service/shoppingcenterfacilities.Atypicalstatisticaltractcomprisesaround8–9neighborhoodsand3100inhabitants.As indicatorsforthetwoneighborgroups’entrepreneurshipbehaviorweusetheoverallfractionofentrepreneursinthepopu- lationagedbetween30and61intheyearofowngraduation(excludingownfamilymembers).Bysettingaloweragelimit of30,weavoidoverlapbetweenneighborsandschoolmates,andbysettinganupperlimitof61years,weavoidinterference fromearlyretirement(whichformostworkersinNorwaymaystartatage62).
Familymembers:Forfamilymembers,wealsodistinguishbetweencloseandmoredistantrelatives.By“closerelatives”, wemeanparentsandsiblings.By“distantrelatives”,wemeanuncles,aunts,and(first)cousins.Asindicatorforthefamily
6Foramorethoroughdescriptionoftheneighborhoodconceptandothergeographicalentitiesusedinthispaper,seeStatisticsNorway(1999).
Table2
Descriptivestatisticsforpeergroups.
ICloseneighbors IIDistantneighbors IIIClosefamily IVDistantfamily VSchool-mates Men:
Malepeergroups
Averagesize 150 1,070 1.23 3.33 48
Averageentrepreneurshipindicator 0.16 0.16 0.13 0.10 0.09
Femalepeergroups
Averagesize 150 1062 1.29 3.21 24
Averageentrepreneurshipindicator 0.04 0.04 0.03 0.03 0.03
Women:
Malepeergroups
Averagesize 154 1103 1.22 3.32 26
Averageentrepreneurshipindicator 0.16 0.16 0.13 0.10 0.09
Femalepeergroups
Averagesize 153 1,091 1.29 3.21 58
Averageentrepreneurshipindicator 0.04 0.04 0.03 0.03 0.05
members’entrepreneurshipbehavior,weagainusethefractionsinvolvedinentrepreneurshipatthetimeofown(person i’s)graduation.
Schoolmates: We identify schoolmates as the persons below age 30 who graduated from the same school/college/universitywithexactlythesameeducation(basedonasix-digiteducationcode)inthesamesemester.By requiringthattheageisbelow30yearsweavoidoverlapwiththegroupofneighbors,andalsoreducethelikelihoodthat entrepreneurshiphasalreadyoccurredatthetimeofgraduation.Asindicatorsforentrepreneurshipbehavior,weusethe fractionofschoolmatesthathasengagedinsomeformofentrepreneurshipwithinfiveyearsaftergraduation.
Eq.(1)embodiesatleasttwopotentialidentificationchallenges.Thefirstisthatofconfoundingfactors:Theremayexist localoreducation-specificvariationsinentrepreneurshippropensitythathavenothingtodowithpeereffects.Wedeal withthischallengebyincludingextensivesetsofcontrolvariables.Thecontrolvariablesin(1)incorporatealargenumber offixedeffects.Inabaselinemodel,theyinclude:
•Age-at-graduationfixedeffects(age=18,19,...,29),
•Schoolfixedeffects(1166differenteducationalinstitutions),
•Educationfixedeffectsforthelastobservededucationaltrack(219differentcategoriesbasedonathree-digitinternational standardclassificationofeducation(ISCED)),
•Travel-to-workareabygraduation-semesterfixedeffects(460differentcombinations),
•Forimmigrants:Region-of-origin-countryfixedeffects(5differentregions).
Inarobustnessanalysisbelow,wealsoincludeindustrydummyvariablesforthesubsetofobservationswhereemploy- mentorentrepreneurshiphasbeenrecordedsuchthatwehavetherequiredinformationaboutindustry.
Thesecondchallengeisthatofreversecausality:Whilethepeervariablesforneighborsandfamilymembersarestrictly predeterminedwithrespecttotheoutcomes,thisisnotthecaseforschoolmates.Thesevariablesareendogenous,inthat theymayhavebeenaffectedby–aswellasaffected–thedependentvariablein(1).Todealwiththeresultantsimultaneity problem,weuseaninstrumentalvariablestrategy.Asinstrumentsforthecontemporaneousentrepreneurshipactivities inthegroupsofschoolmatesweusethefractionsoftheirmothersandfathersthatwereengagedinentrepreneurshipat thetimeofgraduation.Sincemothersandfathersmayaffectsonsanddaughtersdifferently,thisgivesusfourinstruments forthetwo endogenouspeergroupvariables(i.e.,entrepreneurshipratesamongfatherstosons,fatherstodaughters, motherstosons,andmotherstodaughters).Theidentifyingassumptionisthat theconditionalcorrelationbetweena person’sownentrepreneurshipactivitiesandthatoftheparentstohis/herschoolmatesisgovernedbythelatter’simpact onentrepreneurshipamongtheschoolmatesonly.
Another point to note is that while it is natural to interpret causal relationships between the different groups’
entrepreneurshippropensitiesandtheoutcomevariableassomehowrelatedtopeereffects,theeffectsidentifiedforclose familyarealsolikelytoreflectthetransmissionofgeneticfactors(NicolaouandShane,2010;Lindquistetal.,2015).Inaddi- tion,itispossiblethatinheritanceoffamilybusinesses,accesstocheapcapital,andparent-offspringsimilaritiesinchoice ofindustryplayarole,althoughLindquistetal.(2015)findlittleempiricalsupportfortheseexplanationbasedonSwedish data.
DescriptivestatisticsforthevariouspeergroupsarepresentedinTable2.Thegroupsizesofcoursedifferenormously, withtheaveragenumbersvaryingfromonly1–3forthetwofamily-groups,25–50fortheschoolmategroups,around150 forthecloseneighborgroupsandmorethan1000forthegroupsofdistantneighbors.Itisalsonotablethatentrepreneurship ratesinthemalepeergroupsaremuchhigherthaninthefemalepeergroups.
Giventhateachgraduate’sentrepreneurshipbehaviorcanbeaffectedbyasmuchas10differentpeergroups,identi- ficationofeachpeergroup’sisolatedinfluencemaybeproblematicifthepeergroups’entrepreneurshipratesarehighly correlated.Asitturnsout,thereisaconsiderable(generallypositive)correlationbetweenthepeergroups’entrepreneurship
rates,butthecorrelationisfarfromperfect,andonlytwooutofthe45differentcorrelationcoefficientsthatcanbecalculated areabove0.5;seeTableA1intheAppendixAfordetails.
Inordertospecifyourinstrumentalvariables(2SLS)model,let(emsi,efsi)denotetheentrepreneurshipratesforperson i’smale(m)andfemale(f)schoolmates(s),respectively.Thefirststepequationsthentakethefollowingform:
egsi=g(fs)e(fs)i +(fd)g e(fd)i +g(ms)e(ms)i +(md)g e(md)i +controls+i,g=m,f, (2) where(e(fs)i ,e(fd)i ,e(ms)i ,e(md)i )aretheobservedentrepreneurshipratesobservedfortherespectivegroups’parents,where thesuperscriptsindicatefatherstosons(fs),fatherstodaughters(fd),motherstosons(ms),andmotherstodaughters(md).
Hence,ourfirststepequationisdesignedtoexploitthefindingsreportedbyLindquistetal.(2015)thatfathersandmothers influencetheirsonsanddaughtersdifferently.Weexpect,ofcourse,thatthemothersandfatherstosonsaremostimportant forthemalepeergroup,whereasmothersandfathersofdaughtersaremostimportantforthefemalegroup.However,we includeallfourinstrumentsinboththefirst-stepequations,asschoolmatesmayalsohavebeenaffectedbyeachother.
Let(ˆemsi,ˆefsi)bethepredictionsfromordinaryleastsquare(OLS)estimationsofEq.(2).Oursecondstepequationthen becomes
Enti=
k=/s
mkemki+fkefki+mseˆmsi+fsˆefsi+controls+i. (3)
Astheentrepreneurshipoutcomesofinterestarebinary,itmaybearguedthatthelinearregressionframeworkisinap- propriate,andthatweshoulduselogitorprobitmodelsinstead.However,giventhatweneedtocontrolforaverylarge numberofpotentiallyconfoundingfactorsinthisanalysis–andthatweconsideritessentialtodothiswithoutimposing unjustifiedfunctionalformrestriction–wehavealmost1900dichotomouscovariatesinthecontrolvariablevector.Inthe linearmodel,wecandealwiththisbymeansofanalgorithmdesignedforprojectingoutdummy-encodedcategoricalvari- ables;seeGaure(2013).Thisisnotfeasiblewithinalogit/probitframework.However,basedonasimplifiedversionofthe model,weshowintheAppendixA(TableA2)thattheaverageestimatedmarginaleffectsfromatwo-steplogitmodel(witha linearfirststep)areveryclosetothecoefficientestimatesbasedonthe2SLSmodel.7AsalsoshownintheAppendix(Fig.A1), thepatternsofpredictedentrepreneurshipprobabilitiesbasedonthetwomodelsarealsorelativelysimilar,althoughthe linearmodeldoesgiverisetoanumberof(slightly)negativeprobabilityestimates.
4. Mainresults
OurmainestimationresultsarepresentedinTable3.Forcomparison,wepresentboththeOLSresultsandthesecond stage2SLSresults.Thefirststage2SLSresultsarepresentedinTable4.Asexpected,theestimatesfromtheOLS(columns IandIII)and2SLS(columnsIIandIV)modelsinTable3arealmostidenticalforallthepeerinfluences,exceptforthetwo endogenousschoolmatepeerentrepreneurshiprateswherethe2SLSestimatesaresomewhatlargerthantheOLSestimates.
Apossibleinterpretationoftheselatterdifferencesisthattheentrepreneurshipactivitiesofschoolmatesaremeasuredwith someerrorinrelationtotheirpotentialinfluenceonthefocalindividuals’entrepreneurshipdecisions,eitherbecausethey occurafterthefocalindividualhasmadehis/herownoccupationalchoiceorbecausetheyoccurbeforeafirmisformally established.SuchmechanismswilltendtobiastheOLSestimatestowardzero.Inourdiscussionoftheresults,wefocus entirelyonthe2SLSestimates.
Beforeweturntotheresultsofsubstantiveinterest,webrieflydiscussthevalidityandpowerofourinstrumentsfor entrepreneurshipamongmaleandfemaleschoolmates.Sincewehavefourinstrumentsfortwoendogenousvariables,our modelisoveridentified,hence,wecantestforinvalidinstruments.WereporttheSargantestsforoveridentifyingrestrictions atthebottomofTable3(Sargan,1958).Theseteststatisticsshowlittleevidencethatourinstrumentsarecorrelatedwith theerrorterminthesecondstageequation.Theteststatisticformen(ColumnII)isborderlinesignificant,butthismay reflectthatthetruepeereffectsareheterogeneousandthusthattheestimateofapresumedhomogenouseffectwillvary somewhatdependingonthemarginusedforidentification.InTable4,wereportthefirst-stageestimatesforschoolmates’
entrepreneurshipactivities,togetherwithteststatisticsforweakinstruments.Itisclearthattheinstrumentsbasedon thepredeterminedentrepreneurshipbehaviorofthepeers’parentsdohaveaconsiderableinfluenceontheschoolmates’
entrepreneurialactivities,andthatfathersarerelativelymoreimportantforsonsthanfordaughters.Unsurprisingly,both themaleandfemalepeergroupsareprimarilyaffectedbytheirownparents.Thereisoneexceptionfromthispattern, though,namelythatthemalepeerstofemalegraduatesareinfluencedbythefathersoftheirfemaleschoolmates.Thismay reflectthatmentakingmorefemale-dominatededucationsaremorelikelytobeinfluencedbytheirfemaleschoolmates.We presenttwodifferentF-statisticsforthepoweroftheinstruments.ThepartialF-statisticgivestheconventionaltestforthe jointimpactoftheexcludedinstrumentsseparatelyforeachoftheendogenousvariables.Theysuggestthattheinstruments arerelativelystrong,withapossibleexceptionfortheinstrumentsforfemaleschoolmatesinthemaleregression(whichhas anF-statisticslightlybelow10).However,withmultipleendogenousvariables,thepartialF-statisticsareunabletodetect
7Inthesimplifiedmodelwehavedroppedtheschoolfixedeffectsandreplacedthetravel-to-workareabygraduation-semesterfixedeffectswith separatedummyvariablesfortravel-to-workareaandgraduationsemester.
Table3
Estimatedpeereffects(standarderrorsinparentheses).OLSandSecondstage2SLS.Dependentvariable=Ownentrepreneurshipwithinfiveyearsafter graduation.
Men Women
I II III IV
OLS 2SLS OLS 2SLS
Entrep.ratecloseneighbors
Male 0.153*** 0.151*** 0.041*** 0.040***
(0.014) (0.014) (0.009) (0.009)
Female 0.020 0.021 0.023 0.021
(0.028) (0.029) (0.022) (0.022)
Entrep.ratedist.neighbors
Male 0.093*** 0.093*** 0.015 0.016
(0.022) (0.023) (0.015) (0.015)
Female 0.059 0.053 0.113** 0.110**
(0.068) (0.069) (0.050) (0.050)
Entrep.rateclosefamily
Male 0.059*** 0.059*** 0.020*** 0.020***
(0.003) (0.003) (0.002) (0.002)
Female 0.042*** 0.041*** 0.042*** 0.042***
(0.006) (0.006) (0.005) (0.005)
Entrep.ratedist.family
Male 0.032*** 0.032*** 0.007 0.006
(0.005) (0.005) (0.004) (0.004)
Female 0.002 0.002 0.016** 0.015**
(0.008) (0.008) (0.007) (0.007)
Entrep.rateschoolmates
Male 0.164*** 0.268*** 0.059*** 0.079*
(0.012) (0.073) (0.008) (0.043)
Female 0.119*** 0.277* 0.286*** 0.485***
(0.016) (0.166) (0.019) (0.132)
Meanoutcome 0.105 0.105 0.057 0.057
Overidentifyingrestrictionstest(SarganChi-square(2)) 4.689*
[p=0.096]
1.710 [p=0.425]
Excludedinstrumentsforentrep.rateschoolmates Male
Fpartial 62.62 23.58
Fconditional 84.41 26.22
Female
Fpartial 8.695 21.01
Fconditional 11.86 27.83
Numberofobservations(N) 133,714 133,714 119,585 119,585
Note:Allregressionsincludeindicatorvariablesforage-at-graduation(12categories),graduationschool(1166categories),educationlevel/type(219 categories),travel-to-workareabygraduationsemester(460categories),andorigin-regionforfirst-andsecondgenerationimmigrants(5categories).Stan- darderrorsarecomputedwithatwo-wayclusteronneighborhood(closeneighbors)andschoolmate/co-studentpeergroup.*(**)(***)indicatestatistical significanceatthe10(5)(1)%levels.
casesinwhichinterdependenciesimplythatitisdifficulttoidentifywhichoftheendogenousvariablestheyoperatethrough.
WethereforealsoprovideF-statisticsproposedbySandersonandWindmeijer(2016),whichareconditionalontheother endogenousvariable.Thesestatisticsturnouttobewellaboveconventionalthresholdlevelsforweakinstruments(Stock andYogo,2005).Hence,ourinstrumentsappeartonicelydisentanglethepeerinfluencesofmaleandfemaleschoolmates.
Takenatfacevalue,thesecondstagecoefficientscanbeinterpretedastheestimatedchangeinearlycareerentrepreneur- shiparisingfromachangeintherespectivepeergroups’entrepreneurshipratefrom0to1.Note,however,thattheactual variationinthedata–andthusthemarginusedforidentification–variesenormouslyacrossthedifferentpeergroups.For thesmallestpeergroups(closefamily),thevariationinthedataactuallygoesfrom0to1,whereasforthelargergroups (distantneighbors)ittypicallygoesfromaround0.10to0.30forthemalepeergroupsandfromaround0.03to0.10forthe femalegroups.
Thesecondstageresultssuggestthatmen’sentrepreneurshipbehaviorissignificantlyaffectedbyallthemalepeer groups.Femalepeergroupshaveconsiderablylessinfluenceonmen,withstatisticallysignificanteffects onlyforclose family(mothersandsisters)andschoolmates.Women’sentrepreneurshipbehavioristoalargerextentaffectedbyboth maleandfemalepeergroups.Yet,forallpeergroupsexceptcloseneighbors,ownsexpeersaremuchmoreimportantthan thoseofoppositesexalsoforwomen.
Table4
Firststage2SLS.Dependentvariable=Averageentrepreneurshiprateinpeergroupwithinfiveyearsaftergraduation.
Men Women
I II III IV
Maleschoolmates Femaleschoolmates Maleschoolmates Femaleschoolmates Entreprep.rateparents
Fathersofmaleschoolmates 0.083*** 0.002 0.095*** −0.001
(0.006) (0.004) (0.012) (0.003)
Mothersofmaleschoolmates 0.057*** 0.001 0.051** 0.009
(0.011) (0.006) (0.021) (0.006)
Fathersoffemaleschoolmates −0.005 0.027*** 0.015** 0.034***
(0.003) (0.006) (0.007) (0.005)
Mothersoffemaleschoolmates 0.003 0.017* 0.025 0.0574***
(0.007) (0.008) (0.016) (0.012)
Numberofobservations(N) 133,714 133,714 119,585 119,585
Note:Allregressionsincludeindicatorvariablesforage-at-graduation(12categories),graduationschool(1166categories),educationlevel/type(219 categories),travel-to-workareabygraduationsemester(460categories),andorigin-regionforfirst-andsecondgenerationimmigrants(5categories).Stan- darderrorsarecomputedwithatwo-wayclusteronneighborhood(closeneighbors)andschoolmate/co-studentpeergroup.*(**)(***)indicatestatistical significanceatthe10(5)(1)%levels.
Table5
Estimatedcontributionstothegendergapinearlycareerentrepreneurshipbythedifferencesbetweenmaleandfemalepeergroups.
Inpercentagepoints [90%confidenceinterval]
In%ofoverallgendergap [90%confidenceinterval]
Allpeergroups 2.54[0.99,3.98] 51.7%[20.5,81.1]
Neighbors 1.45[0.50,2.39] 29.4%[10.2,49.2]
Family 0.33[0.23,0.43] 6.8%[4.7,8.8]
Schoolmates 0.77[−0.41,1.84] 15.6%[−8.5,37.2]
Note:Theresultsinthetablearebasedonanonparametricbootstrapwith1000re-samplings(withreplacement)andre-estimations.Thereportednumbers arethemean,the5thpercentile,andthe95thpercentileinthedistributionsoftherespectivestatisticsgeneratedbythesetrials.
Forbothmenandwomen,thereisatendencythatcloseneighborsaremoreimportantthandistantneighbors,and thatclosefamilymembersaremoreimportantthandistantfamilymembers.Theformeroftheseobservationsbecome muchmoreevidentwhenwetakeintoaccountthatthereare(onaverage)seventimesasmanydistantasthereareclose neighbors.8
Viewedasawhole,ourestimatessuggestthatpeereffectsareofconsiderableimportanceforearlycareerentrepreneur- ship.Wewillnowusetheestimated2SLSmodeltoassesshowmuchofthegendergapthatcanbeattributedtodifferences inpeerinfluences.Wedothisbycomputingthehypotheticalentrepreneurshipbehaviorundertheassumptionthatmale andfemalepeergroupswerecharacterizedbyexactlythesame(average)entrepreneurshiprates(equaltotheaverageofthe observedmaleandfemalepeergroupaverages).Formen,wethenfindthattheincidenceofentrepreneurshipwouldhave been9.2%insteadoftheobserved10.5%.Forwomen,itwouldhavebeen6.9%insteadoftheobserved5.7.Hence,thegender gapinouroutcomevariablewouldhavebeen2.3insteadof4.8percentagepoints.Peergroupcompositionisthusestimated toexplain2.5percentagepoints(52%)oftheobservedgendergapinearlycareerentrepreneurshipbehavior.Following thislogic,wecanexaminethecontributiontothegendergapprovidedbyeachofthepeergrouptypes:neighbors,family, andschoolmates.Wethenfindthatthepeergroupcompositionamongneighborsaremostimportant(explains29%ofthe gendergap),followedbyschoolmates(16%)andfamily(7%).
Thesenumbersareestimatedwithlargestatisticaluncertainty,however.Toobtainconfidenceintervalsontheexplana- torypowerofpeergroupcomposition,wehaveperformedanon-parametric(clustered)bootstrapexercise,basedon1000 re-samplings(withreplacement)andre-estimations.TheresultsarepresentedinTable5.Theyshowthata90%confidence intervalfortheoverallimpactonthegendergapfrompeergroupcompositionrunsfromaround1to4percentagepoints (21–81%ofthegap).Thestatisticaluncertaintyisparticularlylargefortheroleofschoolmatepeergroups.
5. Alternativemodelsandoutcomes
Inthissection,wefirstassesstherobustnessofourfindingswithrespecttotheselectionofcontrolvariablesandthe compositionofpeergroups.Wethenlookatalternativeoutcomemeasures,withrespecttothedefinitionandtimingof entrepreneurship.
8SeeMarkussenandRøed(2015)foradiscussionofhowpeereffectsarisingfromgroupsofdifferentsizesshouldbecomparedandinterpreted.
Onepotentiallyimportantidentificationchallengecomesfromlocalvariationsinindustrycomposition,i.e.,thattypically entrepreneurialindustriesaremoreprevalentinsomelocalareasthaninothers.Failuretoaccountforthismayimplythat thespatialvariationinentrepreneurialindustriesisfalselyinterpretedasneighborhoodpeereffects.Tosomeextent,the useofeducationdummyvariablesalsoindirectlycontrolsforindustrycomposition,asmanyoftheeducationsspecialize forparticularindustries.Moreover,theuseofTWA-by-yeardummyvariablescontrolsnon-parametricallyforvariationsin industrycompositionatthecommutingzonelevels.Toneverthelesscheckthisfurther,weaddintothemodelcontrolsforthe industryactuallychosenbyeachindividual.Forthispurpose,weuseatwo-digitcodebasedontheStatisticalClassification ofEconomicActivitiesintheEuropeanCommunity(NACE)with88differentcategories.9Inthisexercise,wealsoneedto conditiononatleastoneindustryaffiliationhavingbeenobserved.Inpractice,thismeansthatwecanonlyusegraduatesfor whicheitheranemploymentrelationshiporanentrepreneurshipactivityhasbeenobservedduringthefirstfiveyearsafter graduation.Thisreducesthesampleby17–18%;conf.Table1.TheresultsarepresentedinTable6,columnsIandIV.They areverysimilartothoseobtainedinthebaselineanalysis.Ifanything,theestimatedneighborhoodpeereffectsbecomea bitlargerthaninthebaselinemodel.10
Havingintroducedindustrydummyvariables,wecanalsoassesswhetherpeersinthesameindustryastheindividual itselfhavethesameinfluenceaspeersinotherindustries.Intuitively,wemayexpectpeersinthesameindustrytohave largerinfluence,astheypotentiallyplayatwinroleofbeingbothrolemodelsandcoaches.Hence,ifweremovepersons operatingasemployeesorentrepreneursinthesameindustryfromtherespectivepeergroups,wemayexpecttheestimated peereffectstodecline.Asitturnsout,removingsame-industrypeersamongneighborsandschoolmateshaslittleinfluence ontheestimatedpeereffects,whereastheremovalofsame-industryfamilymembersdoesimplyconsiderablysmallerpeer effects;seeTable6,columnsIIandV.Hence,itappearsthatfamilymembersdoaffectoccupationalchoicesboththrough directlearning/coachingandthroughsocialacceptance/inspiration,whereasmoredistantpeergroupsprimarilyaffectthem asrolemodelsthroughsocialacceptance/inspiration.
Finally,thewaywehavedesignedtheanalysisdataimpliesthatsomeoftheparentsofschoolmatesalsoliveinthesame neighborhoodasthefocalindividual,potentiallychallengingtheexclusionrestriction.Tocheckwhetherthishasinfluenced ourestimates,wereportincolumnsIIIandVIresultsobtainedwhenwehavedroppedfromthedatasetallschoolmates whoseparentsliveinthesameneighborhoodasthefocalindividual.Ascanbeseenfromtheseresults,thishardlychanges ourcoefficientestimatesatall.
Aspointedoutabove,wehavesofarusedafairlycomprehensiveentrepreneurshipdefinitioninthispaper,includingall attemptsatrunningabusiness,eitherasself-employedorasemployedowner,andregardlessofwhetherthebusinessis newornot.Wenowturntothreealternative–andnarrower–definitionsofentrepreneurship.Thefirstconsideraperson asentrepreneuronlyinsofarasentrepreneurshipatsomepointintime(duringthefiveyearoutcomeperiod)isaperson’s maineconomicactivity–inthesensethatitrepresentsthemostimportantsourceofincome.Thesecondconsideraperson asentrepreneurifhe/sheatsomepointintimeisself-employed,andthethirddefinesanentrepreneurasapersonwho createsacompletelynewbusiness(inthesensethatthefirmdidnotexistatalltwoyearspriortoourrecordingofan entrepreneurshipactivity).
Allthesedefinitionsimplylowerestimatedpeereffects;seeTable7.However,sincetheyarenarrowerthantheoneused inourmainanalysis,theyneedtobeinterpretedrelativetoloweraverageoutcomes.Thisisparticularlythecaseforthe entrepreneurshipdefinitionsbasedonthemaineconomicactivityonlyandonthecreationofanewfirm.However,taking thisintoaccount,thepatternofestimatedpeereffectsisverysimilarasthatfoundinthemainanalysis.Menaregenerally moreaffectedbyothermen(withtheexceptionofdistantneighbors’influenceonentrepreneurshipasthemainactivity;
seeColumnI),whereaswomenaremoreaffectedbyotherwomen.
Uptonow,ouranalysishasfocusedentirelyonthesummaryoutcomeof“someentrepreneurshipwithinfiveyears aftergraduation”, which is observed for allthe graduates in ourdataset.We now useourbaseline model toexam- inepeereffectsonannualentrepreneurshipoutcomes,rangingfromentrepreneurshipinthefirstyearaftergraduation anduptothe10thyearaftergraduation.Thisanalysisisbasedonour(wide)baselinedefinitionofentrepreneurship.
Eachoutcome is defined suchthat it takes thevalue one if entrepreneurship occurredin theyear in question, and zerootherwise.Giventhelargenumberofestimatesinvolved inthis exercise,we presenttheresultsgraphically;see Fig.5(men)and6(women).Asthenumberofobservationsdeclineswhenweexceedfiveyearsaftergraduation(since weloseonegraduationcohortfor eachyearwe extendtheoutcomeperiod),thestatisticaluncertainty alsobecomes larger.
Giventheconsiderablestatisticaluncertaintyassociatedwithmanyofthe200uniquecoefficientestimatesreported inFigs. 5 and6, wefocus ontheoverall patternof estimatedeffects here,rather thanonspecific coefficients.There areinparticular twopoints tonote.The firstisthat theoverallstructure of estimatedpeereffects– interms ofthe
9 TheacronymoriginatesfromtheFrenchtranslation:NomenclaturestatistiquedesActivitéséconomiquesdanslaCommunautéEuropéenne.
10 Anotherpotentialidentificationchallengecomesfromnon-randomsortingintoeducationalprograms,assomeeducationaltracksareclearlymore entrepreneur-orientedthatothers.Inourbaselinemodel,wehavecontrolledforeducationbymeansof219dummyvariables,representingboththelevel, direction,andtypeofeducation,usingathreedigitcodebasedontheinternationalstandardclassificationofeducation(ISCED).Inapreviousworking paperversionofthispaper(MarkussenandRøed,2016),wehavetakenthisastepfurtherbyemployingafive-digitcodewith669categories.Thisisdone atthecostoflosingpotentiallyvaluablevariationinpeergroupcomposition,asmany(70)ofthefive-digiteducationcodesareassociatedwithunique peergroups.Theresultsturnouttobeverysimilar,however.
Fig.5.Estimatedpeereffectsformenbyyearssincegraduation(with95%confidenceintervals).
Note:Foreachyear,theoutcomeisequaltooneifsomeentrepreneurshipactivityoccurredthatyear,otherwisezero.SeealsothenotetoTable3fora descriptionofcontrolvariablesandstandarderrorcalculations.
Fig.6.Estimatedpeereffectsforwomenbyyearssincegraduation(with95%confidenceintervals).
Note:Foreachyear,theoutcomeisequaltooneifsomeentrepreneurshipactivityoccurredthatyear,otherwisezero.SeealsothenotetoTable3fora descriptionofcontrolvariablesandstandarderrorcalculations.
Table6
Robustness.Secondstage2SLSresults(standarderrorsinparentheses).Dependentvariable=Ownentrep.withinfiveyearsaftergraduation.
Men Women
I II III IV V VI
Withindustry controls
Excl.peersinsame industry
Excl.schoolm.with parentsinsame neighbor.
Withindustry controls
Excl.peersinsame industry
Exclschoolm.
withparentsin sameneighbor.
Entrep.rateclose neighbors
Male 0.168*** 0.160*** 0.152*** 0.047*** 0.043*** 0.040***
(0.015) (0.015) (0.014) (0.011) (0.011) (0.009)
Female 0.039 0.047 0.021 0.029 0.005 0.021
(0.032) (0.029) (0.029) (0.026) (0.021) (0.022)
Entrep.ratedist.
neighbors
Male 0.103*** 0.101*** 0.093*** 0.020 0.038** 0.016
(0.026) (0.025) (0.023) (0.017) (0.017) (0.015)
Female 0.072 0.037 0.056 0.137** 0.018 0.110**
(0.077) (0.074) (0.069) (0.056) (0.048) (0.050)
Entrep.rateclose family
Male 0.061*** 0.039*** 0.059*** 0.021*** 0.015*** 0.020***
(0.004) (0.003) (0.003) (0.003) (0.002) (0.002)
Female 0.043*** 0.038*** 0.041*** 0.044*** 0.032*** 0.042***
(0.007) (0.006) (0.006) (0.006) (0.006) (0.005)
Entrep.ratedist.
family
Male 0.032*** 0.023*** 0.032*** 0.006 0.010** 0.002
(0.005) (0.004) (0.005) (0.004) (0.003) (0.006)
Female 0.000 0.013* 0.002 0.015* 0.008 0.005
(0.009) (0.007) (0.008) (0.008) (0.006) (0.015)
Entrep.rate schoolmates
Male 0.238** 0.256*** 0.250*** 0.082* 0.094* 0.077*
(0.086) (0.091) (0.074) (0.049) (0.053) (0.042)
Female 0.264 0.325* 0.272 0.427*** 0.341** 0.500***
(0.186) (0.185) (0.167) (0.155) (0.158) (0.136)
Meanoutcome 0.120 0.120 0.105 0.065 0.065 0.056
Overidentifying restrictionstest (SarganChi-square (2))
2.610 [p=0.2712]
2.334 [p=0.3113]
4.061 [0.1313]
1.456 [p=0.4829]
1.791 [p=0.4084]
2.030 [p=0.3624]
Excluded instrumentsmale schoolmates
Fpartial 51.96 48.07 61.27 19.57 16.57 23.53
Fconditional 70.07 64.68 82.23 23.59 21.56 26.34
Excluded instrumentsfemale schoolmates
Fpartial 7.89 8.32 8.74 16.85 15.62 20.20
Fconditional 10.80 11.35 12.00 22.86 21.37 26.88
Numberof observations(N)
111,062 110,961 133,674 98,169 98,062 119,572
Note:Allregressionsincludeindicatorvariablesforage-at-graduation(12categories),graduationschool(1166categories),educationtype/level(219 categories),travel-to-workareabygraduation(460categories),andorigin-regionforfirst-andsecondgenerationimmigrants(5categories).Themodelsin columnsI,II,IV,andValsocontain88industrydummyvariables.Thenumberofobservationsisreducedforthesemodels,aswecanonlyuseobservationsfor whichanindustryhasbeenrevealedthroughemploymentand/orentrepreneurship.Standarderrorsarecomputedwithatwo-wayclusteronneighborhood (closeneighbors)andschoolmate/co-studentpeergroup.*(**)(***)indicatestatisticalsignificanceatthe10(5)(1)%levels.
relative influencesofgenderandcloseness–issimilarregardlessofwhen werecordentrepreneurshipoutcomes.The secondisatendencyfortheestimatedpeereffectstogrowaswemovefurtherawayfromthegraduationyear.Thisisas expected,asthemagnitudeoftheaverageoutcomealsoincreasesconsiderablywithtimesincegraduation;conferFig.3, panel(b).