Gender Discrimination in the Australian Graduate Labour...

167AUSTRALIAN JOURNAL OF LABOUR ECONOMICS

Volume 15 • Number 3 • 2012 • pp 167 - 199

Gender Discrimination in the Australian Graduate Labour Market

Ian W. Li, CurtinUniversityandTheUniversityofWesternAustraliaandPaul W. Miller,CurtinUniversity

Abstract This paper examines gender discrimination in the Australian graduate labour market, using data from the Graduate Destination Surveys 1999-2009. A framework of analysis provided by the over-education/required education/under-education literature is applied. A smaller gender wage gap is found for university graduates than that reported for all workers in earlier studies. It is shown, however, that the gender wage gap widens with the age at graduation. This pattern is argued to reflect the influence of the mismeasurement of actual labour market experience in the conventional education and experience earnings equation on the standardised gender pay gap. Female graduates are less likely to be overeducated, compared to male graduates. Over-education, while associated with substantial penalties, is not a substantial contributor to the gender wage gap.

JELclassification:J240;J310;J700

1. Introduction In ameta-analysis of 263 international empirical studies spanning from the 1960sto the late1990s,Weichselbaumer andWinter-Ebmer (2005) reportedahalvingofthegenderwagegap,from65percentto30percent.Thisdeclinewasattributedbytheseauthorstotheequalisationofhumancapitalendowments.ThetypicalBlinder-Oaxacawage residual, or the genderwage gap usually attributed to labourmarketdiscrimination,wasreportedtohavebeenunchangedovertime.

ThegenderwagegapinAustraliahasalsofallenconsiderablysincethe1960s(see,forinstance,Borland,1999;Gregory,1999).Incontrasttothesituationoverseas,however,thechangeinAustraliaappearsinlargeparttobeduetoadecreaseinthesizeoftheBlinder-Oaxacawageresidual,followingtheEqualPaydecisionsof1969and 1972, and the SexDiscriminationAct of 1984 (Miller, 1994).Nevertheless, astandardisedgenderpaygapofupto15percentremains(Borland,1999).Addressforcorrespondence:IanW.Li,SchoolofPopulationHealth,TheUniversityofWesternAustralia,M431,35StirlingHwy,Crawley,WA6009.Email:[email protected]:Theauthorsaregrateful toGraduateCareersAustralia fordataprovision,andtoBoydHunterandtwoanonymousrefereesforhelpfulcomments.LiacknowledgesfinancialassistanceintheformofanAustralianPostgraduateAward,andMilleracknowledgesfinancialassistancefromtheAustralianResearchCouncil.Theviewsexpressedinthispaperarethoseoftheauthors,andshouldnotbeattributedtoGraduateCareersAustralia.©CentreforLabourMarketResearch,2012

168AUSTRALIAN JOURNAL OF LABOUR ECONOMICSVOLUME 15 • NUMBER 3 • 2012

Recentstudiesofthisgenderpaygaphavedocumentedanumberofstrikingpatterns.First,ithasbeenestablishedthatthegenderpaygapisquitemodestamong15-19year olds, and increaseswith age (ABS, 2004).Theminor genderwagegapamong15-19yearoldscouldbeduetominimumwageeffects.Themaleminimumwagewasextendedtofemalesin1974.Underthisexplanation,itwouldbeexpectedthat therewouldbea reasonably sharp jump in thegenderpaygapbetweenyouthandolderworkers,asonemovesfromawage-settingregimewhereminimumratesofpayaremoreprevalenttoasituationwherewagesaremorelikelytobeabovetheminimum.Analternativeexplanation,thatdrawsonamoregradualwideningofthefemalewagedisadvantagewithage,isthatthemeasureoflabourmarketexperienceincluded in the conventional education and experience earnings equation becomesincreasinglyerrorproneamongolderworkers.

Asecondpatternevidentinthestudiesisthatthegenderwagegapislargeramongthebettereducatedthanitisamongtheless-welleducated(OECD,2011).Thegreaterwagedisadvantageexperiencedbytertiaryeducatedfemalescomparedtotheircounterpartswithoutpost-secondaryqualificationsispresumablyareflectionoftheglassceilingeffectreportedbyKee(2006).Thismayalsobelinkedtoinstitutionalfactors.Thegraduate labourmarket is a relativelyhigh-wagemarket,with averagestartingsalarieswellabove theaverage fornon-graduatesofasimilarage.Preston(2001,p.199)arguesthat‘Femalesbenefitfrominstitutionalregulation(e.g.minimumwagelaws)whenitcomestowagedetermination’.Itwouldthereforebeexpectedthatasinstitutionalregulationwouldhavelittleimpactonthegraduatelabourmarketthegraduategenderwagegapcouldbesignificant.

Fromthisperspective,astudyofgenderwagediscriminationinthegraduatelabourmarket hasmuch appeal. Itwill permit assessment of the relative strengthsof the countervailing ‘young age’ and ‘high level of education’ influences on thegenderwagegap.Thisisoneaimofthecurrentpaper.Moreover, theanalyseswillbeundertakenusingperspectivesfromtheover-education/requirededucation/under-education literature. This seems particularly apt, given the changes in the highereducationparticipationratesofmalesandfemalesinrecentdecades.Uptotheearly1970s, the participation rate ofmales atmost levels of education exceeded that offemales (LeandMiller, 2002).Since then,however, the relative standinghasbeenreversed.Forexample,areportbyeducationconsultantsStrategyandResearchPolicyinEducation(2012),drawingondatafromtheAustralianBureauofStatistics,findsthat in 2011 the higher education attainment rate was around 40 and 30 per centforfemaleandmaleAustraliansaged25to34yearsold,respectively.Thisgreaterrepresentationoffemalesamonggraduatesraisesthequestionofwhetheritisreflectedinfemalesbeingmorelikelytobeovereducated.Inotherwords,giventhatfemalesaremorelikelytoengageinhighereducation,itmightbethatlargerproportionsofthemareinjobswhicharenotsuitedtotheirlevelofeducationalattainment.Relatedtothis,itisofinteresttoseehowindividualsfareinthegraduatelabourmarketiftheyarenotmatchedproperlytotheirjobs,whetheranypenaltytosucheducation-occupationmismatchdiffersformalesandfemales,andwhethertheover-educationphenomenoncontributes to the gender pay gap in the graduate labourmarket.These seeminglyimportantissuesareaddressedbelow.

169IAN W. LI AND PAUL W. MILLER


Therestofthepaperisorganisedinthefollowingmanner.Section2reviewsthe literature onover-education,with an emphasis on studies in this literature thatexamine gender differences. The gender differences in educational mismatch arediscussed in the context of the ‘job search’ hypothesis (Frank, 1978). Section 3describesthedataandpresentsthemethodology.Section4presentsanddiscussestheresultsoftheestimations.Finally,aconclusionisofferedinsection5.

2. Literature Review Over-, Required and Under-education (ORU) There is awell-developed literature on the incidence and earnings effects ofOver-,Required, and Under-education (ORU). This literature has its origins in Freeman(1976),thoughtheempiricalframeworkthatisnowcommonlyusedisduetoDuncanandHoffman(1981).Thekeyconceptconsideredisthatwhilemostindividualswillbe‘matched’totheirjobsonthebasisoftheireducation,somewillnotbe.Individualswith education in excessof that requiredby the jobare considered ‘overeducated’.Conversely, individualswho have less education than that required for the job aretakentobe‘undereducated’.Thelabourmarketoutcomesofmatchedandmismatchedindividualscoulddiffer.

Various conceptual frameworks and measurement issues in this literatureare discussed in Hartog (2000) and Leuven and Oosterbeek (2011). A discussionof theempiricalframeworkswhichhavebeenusedbystudiesexaminingORUcanbe found in McGuinness (2006). Each of these studies offers a summary of theempiricalfindingsintheoverseasliterature,whileasurveyoftheAustralianliteratureispresented inMiller (2007).Fivemainfindingshaveemergedfromtheempiricalstudies.Thefirstfindingrelatestothedefinitionofeducationalmismatch.Generally,three approaches have been used to define educational mismatch: i) job analysis;ii)worker self-assessment; and iii) realisedmatches.The job analysis and realisedmatchesapproachesrelyonobjectivemeasuresofover-education,whiletheworkerself-assessment uses theworker’s subjective assessment of educational (mis)match.Theobjectiveapproaches,whichwillbeusedintheanalysisbelow,areexplainedinmoredetailinsection3.Anumberofstudieshaveshownthatwhereastheestimatedwageeffectsarerobustregardlessoftheapproachadopted,themeasuredincidenceof over-education depends on the definition ofmismatch employed (Hartog, 2000;ChiswickandMiller,2010).

Thesecondfinding relates to the incidenceofeducationalmismatch in thelabourmarket.Hartog’s (2000) review found that, on average, a substantial 40percent ofworkers are not correctlymatched to their jobs. Third,while the earningsreturns to being overeducated are positive, they are less than the earnings returnson required education. Returns to surplus years of education typically range fromone-half to two-thirds of the returns to years of required education. Alternativelystated,thereisanearningspenaltyassociatedwithbeingovereducated.Thisearningspenalty could be due to these workers having unobserved characteristics that areviewednegativelybyemployers.Thus,education-occupationmismatchwillbefoundwheremultipleindicatorsareusedinthematchingprocess(e.g.,qualifications,labour


marketexperience,aptitude,motivation)andtheresearcheronlymeasuresasub-setofthese.Inthissituationitwouldbeexpectedthatworkersclassifiedasovereducatedwould have lower endowments of the unobserved indicators: the higher level ofqualificationthanisusualfortheworker’soccupationcanbeviewedascompensatingfor the deficiencies in these other attributes.The earnings penalty to overeducatedworkerscouldalsobeduetotheseworkershavingafocusonjobsearchratherthaninvestmentonthejob(deOliveiraet al.,2000).Afurtherreasonisthatthereisaskillsrecognitionproblem,whereby employers havedifficulty assessing theworthof theskillslearnedinsomedegreeordiplomaprograms.Thereasoninghereisakintotheargumentintheimmigrantskilltransferabilityliterature(ChiswickandMiller,2008).

Fourth,workerswhoareundereducatedearnmorethantheirpeerswiththesamelevelofeducationbutwhoareworkinginacorrectlymatched(andhencelowerlevel)job.Thismightbeattributedtothepresenceofunobservablefactors,suchasinnateabilityorthepropensitytoworkhard,suchthattheseundereducatedindividualsareabletoenterjobswhichtheywouldotherwisebeunqualifiedfor.1

Fifth, it has been reported that there are differences between males andfemalesinboththeincidenceofeducation-occupationmismatch,andthewageeffectsofsuchmismatch.

The Job Search Hypothesis Studies of gender differences in the incidence and effects onwages of education-occupation mismatch have been guided by several theoretical arguments. One oftheseisthejobsearchhypothesis.Thisproposesthatfemalesaresecondaryincomeearners in the household (Frank, 1978) and so are more constrained in their jobsearch.Thismorelimitedjobsearchisexpectedtoresultinfemalesbeingmorelikelyto bemismatched, and also to incur a largerwage penalty thanmales frombeingovereducated. Buchel and Battu (2003) report results that are consistent with thisdescriptionofthelabourmarket,inthatmarriedwomenwerefoundtobemorelikelytobeovereducated,relativetomenorunmarriedwomen.Thiswasparticularlysoinsmall,localisedlabourmarkets.

However, most studies reject the job search hypothesis. Vahey (2000),for example, restricted his sample to unmarried females (who are unlikely to be asecondary incomeearner)whowerebased in themetropolitanarea,andcomparedtheORUearningseffectsinthissampletothoseforthefemalesamplewithouttheserestrictions.Vahey(2000) found thatmostofhisORUdichotomousvariableswerestatistically insignificant, and the two that were significantly different from zeroexhibitedimpactswhichwerenotreconcilablewiththejobsearchtheory.Similarly,McGoldrick and Robst (1996) also reported that there was no indication thatovereducatedfemalesarepenalisedmorethanmalesduetogeographicalconstraintsintheirjobsearch.1Immigrantworkerswithlowlevelsofeducationareexpected,accordingtoChiswick’s(1978)model, to be intensely self-selected on the basis of positive unobservable characteristics. Thisprovides a basis for interpreting the finding byChiswick andMiller (2008), of undereducatedforeign-bornworkershavingsuperiorearningscomparedtoundereducatednative-bornworkers,asbeingduetothegreaterendowmentsofforeign-bornworkersofunobservedfactorsthatleadtohigherearnings.



A second theoretical perspective is offered by Robst (2007), based onconsiderationofboththesupply-anddemand-sidefactorsthatcouldleadtoeducationalmismatch. Supply-side factors include career-oriented ones, such as accepting apromotion,oramenity-andconstraint-relatedreasons,suchasacceptingajobnearerto home.Thedemand-side factors typically refer to the inability tofind a job thatmatchesthequalificationpossessed.Robst(2007)foundthatmalesweremorelikelytobeovereducatedduetocareer-relatedreasons,whilefemalesweremorelikelytobemismatchedduetofamily-relatedreasons,althoughheacknowledgesthepossibilityofreportingbiasduetosocialnorms.

ORU Differences by Gender in the Graduate Labour Market Anumberofstudieshaveexaminedgenderdifferences ineducationalmismatch inthegraduatelabourmarket.Thereisdisagreementintheresearchfindingsofthesestudies,andevenwithinparticularstudies.Thus,Kler(2005)examinedtheAustraliangraduate labourmarket.He reported that, using the realisedmatches approach, 38percentoffemalegraduateswereovereducated,comparedwith46percentofmalegraduates.Whenthejobanalysisapproachwasused,however,therewasnodifferencebetween males and females in the incidence of over-education.2 McGuinness andBennett (2007)studiedgraduateover-education inNorthernIrelandandfound thatfemalesweremorelikelytobeovereducatedthanmenintheirfirstjob,althoughthegapwasnarrowedsixyearslater.

Battu et al. (2000) used three different measures of over-education intheir study of the UK graduate labourmarket. The two subjectivemeasures usedrevealedthatmalesweremorelikelytobeovereducated,whiletherealisedmatchesapproachyieldedtheoppositefinding.Theabsenceofaclearpatternregardingtherelative importance of over-education formales and females seems to reflect boththemeasurementissuenotedearlieranddifferencesacrossthelabourmarketsofthevariouscountries.

The penalties to over-education among graduates have also been found todiffer by gender. Again there is considerable irregularity in the research findingsacrossstudies.McGuinnessandBennett(2007),forexample,reportedOLSestimatedcoefficientsof-11.3and-22.8percentforovereducatedmalesandfemales,respectively.ThispatternisalsoevidentinthestudybyBattuet al.(2000),thoughthedifferencebygenderworksintheoppositedirectioninanearlierstudybyDoltonandVignoles(2000).Kler’s(2005)studyoftheAustraliangraduatelabourmarketreportedgreaterreturnstosurpluseducationforfemalescomparedtomales.Thesestudiesdonotfocusonthegenderpaygap,oronthecontributionthatORUstatuscanmaketothisinthegraduatelabourmarket.Theseissuesareaddressedbelow.

2Acomparisonoftheseresultsforgraduateswithfindingsfromstudiesofallworkerssuggeststhere are outcomes specific to particular groups. Thus, anotherAustralian study byVoon andMiller(2005),whichcoveredallworkers,reportedthattheincidenceofmismatchwas29.5percentformalesand32.1percentforfemales.


The Blinder-Oaxaca Decomposition Thedecompositionmethodusedintheexaminationofearningsdifferentialsbetweensub-groupsofthepopulationwasdevelopedinBlinder(1973)andOaxaca(1973).IthascometobeknownastheBlinder-Oaxacadecomposition.Initsmostwidelyusedform, the decomposition involves the estimation of a standardMincerian earningsequationseparatelyforthesub-groupsofinterest.Thisequationcanbeexpressedas:

logw ji=a j0+a j1X ji +P ji ,i=1,…n;j=mformaleandfforfemale(1)

where logwi denotes the earnings of individual i, expressed in logarithmic format,andXi denotesthevectorofcharacteristicshypothesisedtoimpactonearnings.Thesuperscriptjidentifiesthemembershipofindividualiinthemaleorfemalegroups.Thedecompositioncanthusbewrittenas:

logw m-logw f=a1

m(X m-X f )+(a1m-a f1)X f+(a m0-a f0)(2)

Theuseoftheestimatedcoefficients,aswellasthemeanvaluesofvariables

for both groups, thus allows for a decomposition of the difference between themeanwages of the sub-groups into two portions. Thefirst termon the right handsideofequation(2)attributesearningsdifferencestodifferencesinendowmentsorcharacteristics,andisconsideredthe‘explained’portionoftheearningsgap.Inthisversionofthedecompositionthedifferencesinendowmentsareevaluatedusingthemalecoefficients.Thisimpliesthatthemalewagestructureisthewagestructurethatwouldprevail in theabsenceofdiscrimination, that is, it is thenon-discriminatorynorm. The second term on the right hand side of equation (2) attributes earningsdifferencestodifferencesincoefficients,orreturnstohumancapitalcharacteristics.Thisportionthusimpliesunequaltreatmentofproductivitycharacteristicsinthelabourmarket.Thecoefficienteffects,togetherwiththedifferencebetweenthetwoconstanttermsintheestimationasexpressedbythethirdterm,formtheunexplainedportionoftheearningsdifferential.Theseeffectsaregenerallyattributedtodiscriminationinthelabourmarket.

Therearealsoalternativedecompositionmethodswhichinvolveathree-folddecompositionof the earningsdifference (see, for example, Jones andKelley1984foradiscussion).Thesedecompositionmethodsinvolvetheadditionofaninteractiontermwhichcaptures the jointeffectofdifferences inendowmentsandcoefficients.Theinterpretationofthisinteractiontermdiffers,dependingontheviewofthelabourmarketadopted.Thedifferenttreatmentsofthisinteractiontermaredescribedinmoredetailinthefollowingsection.Thetwo-folddecompositionappearstobepreferredbyresearchersintheeconomicsdiscipline,whilethethree-folddecompositionseemstobepreferredbyresearchersinthesociologydiscipline.

Measurement Issues with the Blinder-Oaxaca Decomposition Therearetwomainmeasurementissuesthathavebeenraisedinthedecompositionliterature.Thefirstissueliesinthechoiceofthenon-discriminatoryearningsstructure.Thatis,theresultsofthedecompositionofequation(2)abovewouldbedependenton

^ ^ ^ ^ ^



whether themaleor femalewagestructure isusedas thenon-discriminatorywagestructure.Thisissuehasbeenraisedbyanumberofresearchers,includingJonesandKelley(1984),Cotton(1988),Neumark(1988)andOaxacaandRansom(1994).Theavailableresearchdemonstratesthattheresultsofwagedecompositionsaresensitiveto the choice of the non-discriminatory benchmark. In Ferber and Green (1982),for instance, adopting the female wage structure as the non-discriminatory normyieldedanestimateoftwopercentoftheearningsdifferentialasthediscriminatorycomponent. When the male wage structure was adopted, however, discriminationaccountedfor70percentofthewagedifferential.

From a theoretical perspective, the choice of the non-discriminatorybenchmarkdependson thewagestructure that is thought toprevail in theabsenceofdiscrimination.Whereratesofpayinthepost-discriminationeraarelikelytobedeterminedbythewagesofthehigherearninggroup,thentheseshouldbeusedtoevaluatetheendowmenteffectinthedecomposition.3Incontrast,whereratesofpayinthepost-discriminationeraarelikelytobedeterminedbythewagesofthelowerearninggroup,thenitisthelowerearninggroup’spaythatshouldbeusedtoevaluatetheendowmenteffectinthedecomposition.4Theuseofthealternativepaystructuresinthedecompositionalsoprovidesabasisforreconcilingthetwo-wayandthree-waydecompositions.Hence,asJonesandKelley(1984)show,theuseofthepaystructureofthehigherearninggroupasthenon-discriminatorynorminthetwo-waymodelisequivalenttoaddingtheinteractiontermforthethree-waymodeltotheendowmentcomponent.Similarly, theuseof thepaystructurefor thelowearninggroupinthesimpledecompositionisequivalenttoaddingtheinteractiontermforthethree-waymodeltothediscriminationcomponent.

Theideathatthepost-discriminationwagestructurewouldbegivenbyoneofthetwoprevailingwagestructureshasbeenarguedtobeinconsistentwiththenotionof market determined pay structures. The pay structure that emerges in the post-discriminationerawouldpresumablybesomewhereinbetweentheoriginaltwosetsofpay.Neumark(1988),forexample,assumesthatbothnepotismanddiscriminationoccur at the same time, and the choice of the non-discriminatory structure shouldconsiderbothscenariosintandem.Hethussuggeststheuseofapooledwagestructureas thenon-discriminatorywagestructure,where, in this instance, thepooledwageregressiondoesnotincludeagenderinterceptshift.Cotton(1988)proposedtheuseofaweightedaverageofthetwooriginalwagestructureswiththeweightsbeinggivenby the employment shares of the groups being examined.Themethodology in theNeumark-Cotton approach thus allows for the identification of the portions in theearningsdifferentialthatareattributabletonepotismanddiscrimination.

3Atypicalapproachintheliteratureusesboththemaleandfemalewagestructuresasthenon-discriminatorywagestructureinalternativedecompositions,andtheaverageoftheresultsfromtheanalysesarereported.4JonesandKelley(1984)distinguishthesedecompositionsbyaddressinghowonewouldremovetheendowmenteffect:byreducinganygreaterendowmentofthehigherearninggroup(removingaprivilegeundertheirprivilegemodel)orbyenhancinganyinferiorsetofendowmentsofthelowerearninggroup(removingadeficitundertheirdeprivationmodel).Economistsgenerallyviewtheendowmentsofthetwogroupsasequalisingoverthelongerterminresponsetotheremovalofanydifferencesinpay(seeCotton,1988).


In a further refinement of this approach, Fortin (2008) shows that thecoefficientsinthepooledregressioncouldoverstatetheeffectsofvariableswhichvarymarkedlyacrossbothgroups.5Toovercomethis,Fortin(2008)suggeststheinclusionofagenderinterceptshiftinthepooledwageregression(seeFortin,2008,p.898).6Fortin’s(2008)methodallowsfortheattributionofthegenderwagegaptonepotismanddiscrimination,andissimilartoNeumark(1988)andCotton(1988)inthatsense.TheFortin(2008)methodappearstorepresentthepreferredapproachintherecentliterature.AnattractivefeatureofFortin’s(2008)approachisthatitisfullycompatiblewiththeclassicpooledregressionapproachwhichincludesadummyvariableforthedisadvantagedgroup.Thatis,theestimated‘discriminatory’portionoftheearningsgapfromFortin’s(2008)decompositionisequal totheestimatedcoefficientonthedisadvantagedgroupinapooledlinearregression.Fortin’s(2008)examplesshowthatNeumark’s (1988)method yields extreme results, and, empirically, her ‘regression-compatible’decompositiongivesresultsmoresimilartoCotton’s(1988)approach.

Anotherissuethathasbeenraisedwiththedecompositionliteraturelieswiththeuseofdummyvariables in theestimatingmodel.Specifically, ithasbeenshownthatthedecompositionresultsforthecoefficientseffectsforparticularsetsofdummyvariablesaresensitivetothechoiceofthebaseorreferencecategory(Jones,1983).7AsJonesandKelley(1984)note,thechoiceoftheomittedorbenchmarkcategoryisoftenarbitrary,asallavailableoptionsareusuallyequallylogical.TheyillustratethisusingdataonearningsandschoolingformalesandfemalesinAustralia.Specifically, theydistinguishedfivecategoriesofschoolinglevels,andthenomittedthehighestorlowestlevelofschoolinginalternativeestimationsofthegenderwagegap.Theestimatesandconclusionsdifferedmarkedly,dependingonthechoiceofthebenchmarkscenario.Inparticular, theestimatedresultswhereuniversityeducationwas theomittedcategoryindicated that females have favourable coefficients effects compared to men. Thisfindingwasreversedwhenprimaryschooleducationwasusedasthebenchmarkintheiranalysis. Jones andKelley (1984) argue that thedifferences are substantial, and thatthesedifferencespointtovastlydifferentconclusionsandthereforeremedialpolicy.85Fortin(2008)referredtotheestimatedeffectsforunionisationandschoolinginNeumark(1988),whichwerelargerforthepooledsamplecomparedtothecorrespondingestimatesforthemaleandfemalesamples.6Alternatively,Fortin(2008)suggestsweightingthemaleandfemaledummyvariablesbytheirpercentageinthesampletoovercomethisissue.7Theoverallwagegapcomponentsremainthesame,asdoesthecharacteristicseffectsforvarioussetsofvariables,thoughtheapparentimportanceofthecharacteristicseffectforspecificdummyvariablescanvary.8Further,JonesandKelley(1984)showthatthesamegeneralissuearisesinthecaseofcontinuousvariables,wheretherelativesizeofthecomponentsofdiscriminationinthedecomposedwagegapdependsonthelocationsofthezeropointsoftheindependentvariablesinthemodel.Theissuehereisthatthechoiceofsome,ifnotmostorall,ofthezeropointsoftheexplanatoryvariablesusedintheestimatingmodelisarbitrary.This,inturn,causestheresultsofthedecompositiontobearbitraryandbeinfluencedbythechoiceofthezeropointsratherthanactualdiscrimination.JonesandKelley(1984)illustratedtheirpointusingAustraliandataonincome,genderandschooling.Theyshowedthat substantial differences emerge in the portion of the income gap attributable to differencesin the intercept termsand the estimatedcoefficients formales and females,whenhumancapitalendowmentswerespecifiedaseitheryearsofschoolingorageleftschool.Inestimatesofthemodelemployingthelatter‘ageleftschool’specification,theyfoundthat theamountoftheincomegapattributabletotheintercepttermsdroppedby40percentwhereasthecorrespondingamountoftheincomegapattributabletothecoefficientseffectis50percenthigher.Thisissueremainsunresolvedandgenerallyservesasaremindertoexercisecautioninmakingcomparisonsacrossstudies.



Oaxaca and Ransom (1999) illustrate the same point in their empiricalexamplelookingatthegenderwagegapofcollegeprofessors.Thepartialcontributionofdegreetypetodiscriminationwas-19.3logpoints,andthecontributionofconstanttermdifferenceswas21.9logpoints,when‘NoAdvancedDegree’wasusedas theomittedcase.Correspondingfiguresfor thealternativespecificationusing‘PhD’astheomittedcasewere-1.1logpointsand3.7logpoints.Whiletheoverallimpactwasunchangedat2.6logpoints, themagnitudeandimportanceof theroledegreetypeplaysincontributingoralleviatingdiscriminationwerevastlydifferent.

More recent studies have proposed solutions to overcome the identificationproblems stated above. Yun (2005), for instance, suggests an ‘averaging approach’.Thisapproachusestheaverageoftheestimatedcharacteristicsandcoefficientseffectsobtained using all the various possible reference groups for a particular categoricalvariable.Further,Yun(2005)illustratedhowherapproachcanbeimplementedwithouttheneedformultipleestimationruns,throughanormalisationoftheregressionequation.9

To illustrate Yun’s (2005) method, consider a variable that has multiplecategories. One such variable in the analyses below is institution, which has fourcategories,GroupofEight(Go8),AustralianTechnologyNetwork(ATN),InnovativeResearch Universities (IRU) and all remaining universities (Other). In the basicestimatingequationpresentedbelow,‘Other’isusedasthereferencegroup,andtheearningseffectsrecordedbytheregressioncoefficientsareforGo8,ATNandIRU.RefertotheseasbGo8,bATNandbIRU .Alternatively,onecoulduseGo8asthereferencegroup,andfindearningseffectsforATN,IRUandOther.Thesecanbederivedfromthefirstsetofestimates,asaother=-bGo8,aATN=bATN -bGo8andaIRU=bIRU -bGo8.Theconstanttermalsochangesinthisinstance(by-bGo8)toreflectthedifferentreferencegroup.ResultscanalsobeobtainedusingATNandIRUasthereferencegroups.Inotherwords,fourseparatesetsofregressionresultscanbeobtained.

Each of these four sets of estimates can be used in a Blinder-Oaxacadecomposition. Reflecting the different estimated coefficients, the value of thecoefficients effect for institution will change (though the characteristics effect isinvariant to the choice of reference group, as shown in Jones (1983)). Yun (2005,p.768)argues ‘Therefore, Ihaveanumberofchoices for the referencegroup.The“averagingapproach”suggestsusingtheaverageofthecharacteristicsandcoefficientseffectswithvarying referencegroupsas thecontributionof individualvariables tothewage differentials.’Yun (2005) showed that this average of the characteristicsandcoefficientseffectscomputedwithall thepossible referencegroupscan infactbe obtained using the results from a single normalised regression equation,wheretheaverageofthecoefficientestimatesestimatedusinganarbitraryreferencegroup(forexample,(bGo8+bATN+bIRU+0)/4)issubtractedfromtheestimatedcoefficientsderived with the same reference group, and also added to the constant term. Thetransformedmodelcanbeviewedasarestrictedleastsquaresregressionmodelbasedonanexhaustivesetofthemutuallyexclusivesetofdummyvariables,and,followingtheconstructionofthenormalisedequation,istermeda‘deviationcontrastcoded’or‘averagingmethod’model.Thismodelisimplementedbelow.

9OtherapproacheshavebeenproposedbyGardeazabalandUgidos(2004)andNeilson(2000).However,Yun (2005)notes that identical results areproducedbyall threeapproachesand thechoiceofapproachadoptedshouldthusbedictatedbyefficacyorease.


3. Data and Methodology Data ThisstudydrawsondatafromtheGraduateDestinationSurveys(GDS),1999-2009.TheGDSisanannualcensusofallgraduateswhohavecompletedtherequirementsforaqualificationawardfromanyAustralianuniversity.TheGDSisconductedtwiceyearly,andtakesplacefourmonthsafterthecompletionofaqualification,typicallyinAprilandOctober.Whiletheresponsibilityforadministeringthesurveylieswithindividualuniversities,oversightoftheGDS,includingdatacodinginstructions,lieswithGraduateCareersAustralia.Thisdatasetcoversawiderangeofthegraduates’characteristics, such as their labour market activities after graduation, personalcharacteristics,andinformationontheireducation(past,andcurrent, ifapplicable).TheresponseratefortheGDSisreasonablyhigh,withanaverageresponserateof62percentovertheyearscoveredinthepresentstudy.AstudyconductedbyGuthrieandJohnson(1997)indicatedthattheGDScanbeconsiderednationallyrepresentativeofthegraduatelabourmarketinAustralia.

ThedatasamplewasrestrictedtograduateswhowereemployedinAustralia.At the same time,graduateswithmissing informationon thevariablesused in theanalysis,namely,salary,instituteofstudy,levelofqualification,age,tenure,double-degree, occupation, sector of employment, industry of employment, length ofemploymentcontract,hoursofwork,modeofstudy,languagebackground,residencystatus,gender,andself-employmentstatus,wereremovedfromthesample.Thefinalsampleconsistsof569,325observations.

Measurement of Education Mismatch Inthepresentstudy,twodifferentapproachesareusedtodefineeducationalmismatch.First,thejobanalysisapproachisused.Thisapproachusesrequiredlevelsofeducationasdefined ina jobdictionary.The required levelofqualifications toperformeachoccupation is derived from the Australian Standard Classification of Occupations(ASCO),whichismanagedandupdatedbytheAustralianBureauofStatistics(ABS,2011). TheGDS data from 1999 to 2005 had been coded according to theASCOclassification,whilethedatafrom2006to2009werecodedbasedontheAustralianandNewZealandStandardClassificationofOccupations(ANZSCO),alsomanagedbytheABS.Thedatafrom2006to2009wererecodedintotheASCOclassification,using the ANZSCO to ASCO correspondence table published by the ABS. Therequiredlevelofqualifications,asidentifiedintheASCOpublication,consistsoffivetypes:1)certificateii,2)certificateiii,3)certificateiv,4)diploma,and5)bachelor’spassdegree.However, the threecertificatecategorieshavebeencollapsed intoone‘certificate’category,asthenumberofobservationsineachseparaterequired-attainedcategory(suchasthosewhoattainedadiplomaandworkedinajobthatrequiredacertificateii)wastoosmall.

Thegraduatescanbegroupedintoeightlevelsofeducationalattainment:i)diploma;ii)associatedegree;iii)bachelor’spassdegree;iv)bachelor’shonoursdegree;v)graduatediploma;vi)graduatecertificate;vii)mastersdegree;andviii)doctoraldegree.Based on this information, a total of 24ORU categorieswere constructed(eight actual education levels times three required levels),with two of these beingcorrectlymatched,twoundereducated,andtheremaining20overeducated.Dummy



variablesforthesecategoriesareusedintheORUearningsequation,followingthedetaileddummyvariablespecificationproposedbyVahey(2000).

In an alternative analysis, information from the 2006 Australian CensusofPopulation andHousingwasused todetermine the required levels of educationseparately formales and females. Specifically, this realisedmatches, or statistical,approach determines the modal levels of education for the various occupations,separatelyformalesandfemales.Individualswitheducationlevelsabovethemodallevelsaredeemedtobeovereducated.Thereverseholdsfortheundereducated.TherequiredlevelsofeducationdefinedherearemoredetailedthanthatfromtheASCO,and consist of the following categories: i) year 10; ii) year 12; iii) certificate; iv)diploma,and;v)bachelor’spassdegree.Thus,a totalof40ORUcategoriescanbeconstructedinthispartoftheanalysis(thatis,eightactuallevelstimesfiverequiredlevels).While the number of correctlymatched and undereducated categories hereremains thesameasunder theASCO-basedapproach, the increasednumberof36overeducatedcategoriespermitsamoredetailedlookattheearningseffectsofbeingovereducatedbylargerextents.Furthercommentsonthisapproach,anditsbenefits,areofferedinasubsequentsection.

Estimation Models TheORUmodelofearningscanbeexpressedinthefollowingform:

logwi=b1Zi+b2Di

o+b3Dir+b4Di

u+Pi (3)

wherewrepresentsthehourlywage,usedintheanalysisinnaturallogarithmicformat,Z representsavectorofcharacteristicscorrelatedwithearnings,andDo,DuandDrare vectors of dummy variables indicating if the individual is overeducated (Do),undereducated (Du), or correctlymatched to his or her occupation of employmentintermsofeducation(Dr),asidentifiedintheprecedingsub-section.ThevariablesincludedinZ indicatethegraduates’gender,Englishspeakingbackground,residencystatus, mode of enrolment, further study status, university group, broad field ofstudy,self-employmentstatus,lengthofemployment,industryofemployment,sectorof employment, year of graduation and labourmarket experience. Two proxies forexperienceareused,namely,theageofthegraduateandtheyearsoftenure,withbothproxiesenteredintotheestimatingequationinquadraticform.Thesummarystatisticsanddescriptionsofthesevariablescanbefoundinappendix1.

InordertoobtainagreaterunderstandingoftheORUearningseffects,andtheirimpactsongenderwagedifferences,aBlinder-Oaxacadecomposition,asoutlinedinequation(2), isused.Toaccommodatethefact thatanumberofalternativenon-discriminatorywagestructurescanbeusedinthiscomputation, thedecompositionhereisbasedontheaverageofthosewhichuse,respectively,themale(b̂m)andfemale(b̂f)wagestructuresinthisregard.

10TheanalysiswillalsoincorporateYun’s(2005)‘averaging approach’ to overcome themeasurement issues raised earlier, and drawcomparisonsbetweentheresultsoftheconventionaldecompositionmethodandthosefromYun’s(2005)approach.

10Seefootnote3.


4. Results Incidence of Over-education by Gender A preliminary assessment of the importance of the difference by gender in theincidenceofover-educationandunder-educationcanbegainedfromfigure1,whichcharts,from1999to2009,theproportionofworkerswhoareappropriatelytrainedfortheirprofession.Threecommentsareofferedaboutfigure1.First,acrossallyears,anaverageof34percentofmalesareappropriatelytrainedfortheirjobs,whilethecorrespondingfigurefor females is38percent.Second,as the incidenceofunder-educationistrivialamongthetertiaryqualified(itaccountsforlessthanonepercentof the sample), it is readily apparent that the incidence of over-education is ratherhigh,ataround60to65percent,andhigherformalesthanitisforfemales.Ahigherincidenceofover-educationformalesthanforfemalesiscontrarytowhatisexpected,giventhechangesinthegendermixofuniversityenrolmentsoverthepasttwodecades,asdiscussedintheintroduction.However,thispatternisconsistentwiththefindingsofKler (2005),whoalsostudied theAustraliangraduate labourmarket.Third, thedegreeofeducation-jobmatchappearstobegraduallydecreasing,andthustheextentofeducation-jobmismatchincreasing,forbothmalesandfemales.Inotherwords,ifeducation-jobmismatchisaproblemfortheAustraliangraduatelabourmarket,itisaproblemwhichhasbeenexacerbatedinrecenttimes.

Figure 1 - Proportion of Appropriately Trained Workers by Gender, 1999-2009

45

0

25

40

30

10

Year

Percent

1999

35

5

15

Males Females

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

20



Results from the ORU Model of Earnings The results from the estimation of the ORUmodel of earnings determination arepresentedintable1.Theresultsfromestimationusingthefullsamplearepresentedinpanel(i),whileestimatesfortheseparatesamplesofmalesandfemalesarepresentedinpanels(ii)and(iii),respectively.NotethatORUvariablesaredenotedbytheword‘oru’ in thefirst portion of their names.Themiddle portion indicates the attainedqualification,while theendportiondenotes thejoblevel.Forexample, thevariableoru_hons_certisusedforgraduateswhohaveattainedabachelor’shonoursdegree,and are working in certificate level jobs. All earnings effects are evaluated withreference to the earnings of graduateswith a bachelor’s pass degree, andwho areworkinginabachelor’sleveljob.

Table 1 - OLS Estimates of the ORU Model of Earnings

Full Males FemalesVariable (i) (ii) (iii)Constant 2.305*** 2.233*** 2.274*** (175.708) (95.120) (151.039)Female -0.048*** (a) (a) (29.159)Age# 0.035*** 0.038*** 0.034*** (45.334) (26.984) (38.731)Agesquared/1000 -0.397*** -0.400*** -0.404*** (36.847) (20.527) (32.903)Non-Englishspeaking# -0.039*** -0.044*** -0.036*** (17.352) (13.066) (11.803)Non-Australian# -0.202*** -0.187*** -0.216*** (33.694) (22.988) (24.338)Tenure 0.015*** 0.016*** 0.014*** (30.517) (20.954) (22.485)Tenuresquared/1000 -0.404*** -0.471*** -0.376*** (17.551) (13.555) (12.443)Doubledegree# 0.008*** -0.001 0.014*** (3.016) (0.218) (3.939)Part-timestudy# 0.086*** 0.094*** 0.081*** (43.574) (30.379) (31.497)Furtherstudy# 0.007*** 0.012*** 0.004 (3.330) (3.603) (1.430)GroupofEight 0.026*** 0.030*** 0.024*** (13.210) (9.994) (9.378)AustraliaTechnologyNetwork 0.031*** 0.028*** 0.034*** (13.777) (7.856) (11.696)InnovativeResearchUniversity# 0.004* -0.007* 0.011*** (1.952) (1.911) (4.016)NaturalandPhysicalSciences# -0.076*** -0.084*** -0.065*** (20.422) (15.260) (12.916)InformationTechnology# -0.031*** -0.041*** -0.007 (7.643) (8.234) (0.935)Engineering# -0.008** -0.011** 0.012 (2.006) (2.456) (1.413)Architecture -0.098*** -0.097*** -0.092*** (17.467) (13.191) (10.684)AgricultureandEnvironment# -0.138*** -0.163*** -0.109*** (27.394) (22.880) (15.294)


Table 1 - OLS Estimates of the ORU Model of Earnings (continued)

Full Males FemalesVariable (i) (ii) (iii)Nursing# -0.113*** -0.152*** -0.098*** (28.156) (15.102) (21.165)Medicine# -0.028*** -0.041*** -0.019*** (8.149) (6.753) (4.510)Education# -0.062*** -0.075*** -0.052*** (16.597) (11.803) (11.240)SocietyandCulture# -0.058*** -0.065*** -0.048*** (21.762) (14.904) (14.257)CreativeArtsandOthers# -0.117*** -0.134*** -0.102*** (28.527) (18.337) (20.204)Self-employed 0.015*** 0.007 0.017** (2.679) (0.873) (2.043)Privatesector# -0.054*** -0.058*** -0.052*** (25.817) (15.709) (20.118)Short-termemployment -0.095*** -0.099*** -0.093*** (49.105) (29.550) (39.125)oru_dip_cert -0.216*** -0.186*** -0.234*** (10.674) (6.022) (8.785)oru_dip_dip -0.018 -0.018 -0.020 (1.561) (1.158) (1.177)oru_dip_bach 0.031** 0.019 0.036** (2.564) (1.033) (2.343)oru_ascdeg_cert# -0.185*** -0.147*** -0.220*** (9.440) (5.557) (7.704)oru_ascdeg_dip -0.068*** -0.072*** -0.066*** (6.195) (4.961) (3.804)oru_ascdeg_bach -0.019 -0.016 -0.035 (1.155) (0.688) (1.471)oru_bach_cert -0.156*** -0.160*** -0.150*** (55.637) (35.295) (42.038)oru_bach_dip -0.092*** -0.088*** -0.093*** (27.292) (17.170) (20.650)oru_hons_cert -0.101*** -0.096*** -0.103*** (14.606) (9.065) (11.428)oru_hons_dip# -0.037*** -0.012 -0.053*** (4.350) (0.971) (4.721)oru_hons_bach 0.027*** 0.030*** 0.029*** (7.719) (6.037) (5.719)oru_gcert_cert -0.082*** -0.072*** -0.094*** (7.423) (4.913) (5.994)oru_gcert_dip 0.015 0.009 0.012 (1.563) (0.674) (0.807)oru_gcert_bach# 0.118*** 0.100*** 0.127*** (32.361) (16.294) (28.118)oru_gdip_cert -0.117*** -0.130*** -0.110*** (12.961) (9.141) (9.513)oru_gdip_dip -0.008 -0.009 -0.015 (0.901) (0.672) (1.281)oru_gdip_bach 0.089*** 0.093*** 0.085*** (31.084) (18.749) (24.412)oru_mast_cert# -0.122*** -0.143*** -0.109*** (14.613) (12.174) (9.097)oru_mast_dip# 0.066*** 0.074*** 0.041*** (8.391) (7.129) (3.518)



Table 1 - OLS Estimates of the ORU Model of Earnings (continued)

Full Males FemalesVariable (i) (ii) (iii)oru_mast_bach 0.183*** 0.186*** 0.174*** (63.714) (41.567) (45.974)oru_phd_cert# 0.076*** 0.010 0.121*** (3.249) (0.289) (3.803)oru_phd_dip 0.084*** 0.078*** 0.089*** (4.066) (2.698) (2.995)oru_phd_bach 0.199*** 0.192*** 0.197*** (39.303) (26.622) (27.509)Industry Included Included IncludedYearofGraduation Included Included IncludedObservations 569,325 221,746 347,579AdjustedR2 0.188 0.218 0.164F-statistic 1720.73 804.87 928.1

Notes:Absolutevaluesofheteroscedasticityconsistent‘t’-statisticsarepresentedinparentheses.*,**and***indicatesignificanceattheten,fiveandonepercentlevels,respectively.Themodelsincludeddummyvariablesfor15industriesofemployment,andforthe10yearsofgraduation.(a)indicatesthatthevariablewasnotenteredintotheestimatingequation.#indicatesstatisticaldifferenceformalesandfemales,atthetenpercentlevel.Themiddleportionsoftheoruvariablesindicatetheattainedlevelofqualification,andtheendportionindicatesthejoblevel.

Mostof thepooled sample results inpanel (i) arebroadly similar to those

reported in similar studies, and attention is drawn to just two findings. First,the estimatedgenderwagegap isfiveper cent, and this ismuch smaller than thattraditionally found in the literature, of up to 15 per cent (see Borland, 1999). Asdiscussedabove,thegraduatelabourmarketwillbeaffectedbya‘youngage’effectthatwilldecreasethefemalewagedisadvantage,anda‘highlevelofeducation’effectthatwillincreasethefemalewagedisadvantage.Themodestfivepercentwageeffectinthepooledsampleanalysissuggeststhattheformeroftheseisthemoreimportantinfluence.Thisisexploredinasubsequentsection.

Second, there are distinct differences in earnings outcomes accordingto whether the graduates are correctly matched, undereducated or overeducated,andmost of these differences are statistically significant at the one per cent level.Where overeducated, the earnings outcomes vary according to the extent of theover-education.Relativetothebenchmarkgroupofbachelor’spassdegreegraduateswhoareworkinginbachelor’sleveljobs,themostseverepenalty,of22percent,isobservedfortheovereducateddiplomagraduatesworkingincertificateleveljobs.Thehighestearningpremium,of20percent, isobservedfor theovereducateddoctoralgraduateswhoworkinbachelor’spassdegreeleveljobs.Thusthereisadifferenceofover40percentagepointsbetweentheearningsofthehighestandlowestpaidORUgroups.11Inrelatedresearch(Li,2013),itisshownthattheearningseffectsassociatedwithover-educationarebroadlythesameforthegraduatesfromthevariousgroups11 Note that while there are sizeable earnings effects associated with being overeducated, theearnings effects for the correctly matched and undereducated graduates are very modest, thelargestbeingjustthreepercent.


ofinstitutions(GroupofEight,AustralianTechnologyNetwork,InnovativeResearchUniversities,andallremaininguniversities).Thiscastssomedoubtupontheempiricalrelevanceofthequalificationrecognitionargumentfortheearningseffectsassociatedwith overeducated status, as differences between well-established and less-well-establisheduniversitieswouldbeexpectedunderthatargument.12

Therearetwostrikingfeaturesoftheseestimatedearningseffectsassociatedwithover-education.First,graduateearningsaremoreclosely related to thenatureofthejobthantothequalificationpossessed.Forexample,graduateswhoworkinacertificateleveljobearnlessthanthereferencegroup,regardlessoftheeducationallevel attained. Second, earnings premiums that tend to increase with the level ofqualificationareobservedforgraduatesinbachelor’spassdegreeleveljobs.Inotherwords, despite the close linkage of earnings to jobs, there remains a payoff to theacquisitionofahigherlevelofqualification.Anexceptioninthisregardisassociatedegreegraduates.Thesepatternsarealsoobservedfortheestimationsobtainedfortheseparatesamplesofmalesandfemales.

The adjusted R2 for the male and female analyses are 0.218 and 0.164,respectively.TheORUmodel of earnings, therefore, can be said to have relativelyhigherpowerinexplainingtheearningsofmalegraduates.Thiscouldbeduetotworeasons. First, the estimated coefficients for the industry of employment variablesindicate thatemployment incertain industries, suchaseducationandmining,haveestimatedimpactsofgreatermagnitudeformalescomparedtofemales.Otherindustryvariables,suchasforhighereducation,constructionandengineering,havestatisticalsignificance formales, but not for females. Second, due to data unavailability themodel does not control formarital status or the number of children the graduateshave.Thesecharacteristics,arguably,wouldimpactonwomenmore,asthehouseholdburdengenerallyfallsonwomenmorethanmen.

AnF-testwasconductedtoseeifthereweregenderdifferencesintheestimatedcoefficients.This yielded anF-statistic of 21.254, indicating that someor all of theestimatedcoefficientsformalesandfemalesdiffer.Acomparisonofthecoefficientsofspecificvariablesformalesandfemalesrevealedanumberwheretherearestatisticallysignificantdifferences.Thesearedenotedby the #beside thevariablenames in thetable.Forinstance,maleInformationTechnologygraduatesearn4.1percentlessthanmaleManagementandCommercegraduates(thebenchmarkgroupintheestimatingequation), while the corresponding coefficient for females is insignificant. MaleAgricultureandEnvironmentgraduatesearn16.3percent less than thebenchmarkgroup,whilefemalegraduatesinthiscategoryareslightlybetteroff,withasmallerearningsdisadvantageof10.9percent.Similarly,femaleNursinggraduatesearn9.8percentlessthanthereferencecategorywhiletheirmalecounterpartsearn15.2percentless.Thesedifferencesarereasonablyminor,andaffectrelativelyfewinthesample,andsothediscussionwillfocusonthegenderdifferencesintheORUearningseffects.

Gender Differences in the ORU Earnings Impacts There are a number of differences between males and females with respect tothe estimated coefficients on theORU variables.All of these differences relate to

12Asthesamplecomprisesnewgraduates,theroleofon-the-jobtrainingversusjobsearch(seedeOliveiraet al.,2000)cannotbeexaminedrigorouslywiththesedata.



differencesinthesizeofthepointestimates,withthesignoftheestimatedimpactsonearningsbeingconsistentforbothsexes.The‘t’-testsofdifferencesbetweenmalesandfemalesonthespecificORUcoefficientsindicatedthatonlysixoftheestimatedORUcoefficientsdifferedstatisticallybygender,namely:i)associatedegreegraduatesworking in certificate level jobs; ii) bachelor’s honours degree graduates workingindiploma level jobs; iii)graduatecertificategraduatesworking inbachelor’spassleveljobs;iv)mastersdegreegraduatesworkingincertificateleveljobs;v)mastersdegreegraduatesworkingindiplomaleveljobs,and;vi)doctoralgraduatesworkingincertificateleveljobs.Twogeneralpointscanbemaderegardingthesesixdifferences.First, the ORU categories involved are heavily concentrated in the diploma andcertificateleveljobsand,hence,ORUdifferencesbygendercanbesaidtobemorelikely found for those in lower-level jobs.Second, for these six categories, femaleswereworseoffthanmalesinonlythreecategories.

The most substantial penalty to being overeducated is for the associatedegree graduates who are working in certificate level jobs (oru_ascdeg_cert):Whereasmaleassociatedegreegraduatesincertificateleveljobsearn14.7percentlessthanthebenchmarkgroup,theirfemalecounterpartsearn22percentlessthanthereferencecategory.

Malegraduateswithabachelor’shonoursdegreeworking indiploma leveljobs have earnings that do not differ statistically from the benchmark group ofcorrectlymatchedbachelor’spassdegreegraduates.Femalegraduateswiththesameeducational attainment andworking in jobs that require a diploma, however, earn5.3percent less than thefemalebenchmarkcategory. In thecategoryofgraduateswithgraduatecertificatesworking inbachelor’spass level jobs, femalesmarginallyoutperform theirmale counterparts in earnings. Female graduates in this categoryearn12.7percentmorethanthereferencegroup,whilemalegraduateshereearn10percentmore.

Mastersdegreegraduatesworkingindiplomaleveljobs(oru_mast_dip)havepositivereturnstotheirhigherqualifications.However,malesinthiscategoryhaveanearningsreturnof7.4percentwhereasfemaleshaveamoremodestreturn,of4.1percent.Mastersgraduatesworkingincertificateleveljobs(oru_mast_cert)faremuchworse than this,withmaleshavingearnings14.3percent less than thebenchmarkgroup,andfemales11percentlessthanthebenchmarkgroup.

A considerable gender difference in the magnitude of the ORU earningsimpacts is observed for doctoral graduates in certificate level jobs – the mostovereducated category. For male graduates, the impact of being overeducated inthisinstanceisstatisticallyinsignificantcomparedtothebenchmarkgroupofmalebachelor’spassdegreegraduatesworkinginmatchedjobs.Femalegraduatesinthissituation,however,getanearningspremiumcomparedtothebenchmarkgroupof12.1percent for theirsurplushumancapital.Note,however, that femalePhDgraduatesemployedincertificateleveljobshaveearningsaround8percentagepointslowerthanfemalePhDgraduatesemployedinjobsrequiringabachelor’spassdegree.

Of the six estimated coefficients with statistically significant differencesbetweenmalesandfemales,three(oru_gcert_bach, oru_mast_cert andoru_phd_cert)indicated that femaleswerebetter off compared tomales in termsof the earningseffects associatedwith over-education, which is contrary to the predictions of the


job search hypothesis outlined above.13 Clearly, the above comparison of genderdifferencesintheORUearningseffectsgiveslittlesupporttothisinterpretationofthelabourmarket.This,however,couldbeareflectionofthedominanceofworkersfrommetropolitanareasinthepresentanalysis,whoarenotasgeographicallyconstrainedintheirjobsearch.14

ORU Analyses Using Gender-Specific Required Levels of Education Thusfar, theORUvariableshavebeenconstructedusing thesamereference levelsof education for males and females. In this section the analyses are undertakenusinggender-specificreferencelevelsofeducationforeachoccupation.Thegreatervariationacross theORUcategoriesunderthisapproachprovidesabetterbasisforusingaBlinder-Oaxacadecompositioninthestudyofthegenderearningsgap.

The conventional job classification approach, such as inASCO, holds thatthesameeducationalstandardappliesforallworkersinanoccupation.Empirically,however, it often appears that standards differ betweenmales and females. In the‘reverse regression’ literature, for example,when schooling levels are regressedonincome,atypicalfindingisthattoreceivesimilarpayfemalesgenerallyrequirehigheramountsofeducation(seeGoldberger,1984;Kapsalis,1982).KamalichandPolachek(1982), for example, find that females have around 1.2 years more of schooling,comparedtomaleswithsimilarearnings.

Toaddressthisissue,themodalqualificationforeachgenderwasobtainedforeachoccupationusingdatafromthe2006AustralianCensus.Thereare17occupations,outofthetotalof103listedinthedata,inwhichthemodallevelsofeducationdifferedbygender.Amongthese17occupations,inonly3instanceswasthemodallevelofeducationhigher formales than for females.However,despite thegenerallyhighermodallevelsofeducationforfemales,theirincidenceofeducationalmatchremainedunchangedfromthefigurefoundintheearliersection,at38percent.Males,however,areless‘matched’totheiroccupations,witha30percent incidenceofeducationalmatch,fourpercentagepointslessthanintheprecedingsection.Thisislikelytobeaconsequenceofoccupationalsegregationbygenderandthelowerreferencelevelsofeducationintheanalysisofmalegraduates.

Theresultsoftheestimationoftheearningsequationusingthegender-specificmodal levelsofeducationarepresented in table2. In thissetofanalyses theORUvariableshavenamesbeginningwithcen.AswiththeORUvariablesintheprecedingsections,thecenvariablenameshavetheattainedqualificationinthemiddleportion(certtophd),andthemodallevelsofeducationattheendportion(y10tobach).

13Notethatoccupationswerenotusedasregressorsinthemainestimatingequation,astheywereusedtoconstructtheORUvariables.However,separateregressionswererunwith11occupationalgroupsascontrolstoestimatetheimpactofoccupationalsegregationonORUearningseffects.Formales, the negative earnings effects associated with the ORU variables are reduced onceoccupations are taken account of. On the contrary, females have their earnings disadvantageexacerbatedonceoccupationiscontrolledfor.Thissuggeststhatfemalesareabletooffsetsomeoftheearningsdisadvantagesassociatedwithover-educationthroughoccupationalmobility.14Thedatasetcontainsinformationontheresidentialandemploymentpostcodesofthegraduates.However,thesewerenotenteredintotheestimatingequation,asthereweremissingvaluesforasubstantialnumberoftherespondents.



Table 2 - OLS Estimates of the ORU Model, Gender-specific Required Education

Full Males FemalesVariable (i) (ii) (iii)Constant 2.228*** 2.120*** 2.199*** (171.798) (91.246) (148.410)Female -0.049*** (a) (a) (29.650)Age# 0.041*** 0.044*** 0.040*** (53.392) (32.029) (45.767)Agesquared/1000 -0.455*** -0.466*** -0.461*** (42.603) (24.032) (37.899)Tenure# 0.019*** 0.019*** 0.017*** (39.528) (25.898) (28.936)Tenuresquared/1000# -0.538*** -0.609*** -0.502*** (23.251) (17.259) (16.594)Non-Englishspeaking# -0.044*** -0.050*** -0.040*** (19.578) (14.648) (13.149)Non-Australian -0.214*** -0.213*** -0.227*** (35.822) (26.568) (25.612)Furtherstudy# 0.006*** 0.010*** 0.003 (2.832) (3.034) (0.895)GroupofEight# 0.021*** 0.030*** 0.021*** (11.219) (9.954) (8.350)AustralianTechnologyNetwork 0.030*** 0.032*** 0.033*** (13.278) (9.041) (11.519)InnovativeResearchUniversity# -0.000 -0.011*** 0.008*** (0.148) (2.943) (2.754)NaturalandPhysicalSciences -0.090*** -0.085*** -0.082*** (24.251) (15.413) (16.351)InformationTechnology# -0.033*** -0.032*** -0.012 (8.136) (6.436) (1.585)Engineering -0.021*** -0.013*** 0.002 (5.394) (2.670) (0.223)Architecture -0.116*** -0.095*** -0.110*** (20.724) (12.870) (12.745)AgricultureandEnvironment# -0.137*** -0.158*** -0.113*** (27.158) (22.057) (15.675)Nursing# -0.124*** -0.155*** -0.113*** (30.761) (15.201) (24.178)Medicine -0.038*** -0.042*** -0.032*** (11.018) (6.886) (7.479)Education# -0.070*** -0.076*** -0.060*** (18.962) (11.950) (13.056)SocietyandCulture# -0.060*** -0.064*** -0.049*** (22.313) (14.735) (14.487)CreativeArtsandOthers# -0.126*** -0.137*** -0.109*** (30.672) (18.820) (21.485)Self-employed 0.019*** 0.014* 0.018** (3.285) (1.720) (2.215)Privatesector -0.054*** -0.056*** -0.050*** (25.374) (15.084) (19.336)Short-termemployment# -0.101*** -0.110*** -0.098*** (52.295) (32.906) (41.398)cen_dip_y10# -0.374*** -0.228*** -0.523*** (6.841) (3.438) (3.761)


Table 2 - OLS Estimates of the ORU Model, Gender-specific Required Education (continued)

Full Males FemalesVariable (i) (ii) (iii)cen_dip_y12 -0.148*** -0.122** -0.172*** (7.326) (2.453) (7.221)cen_dip_cert# -0.016 0.013 -0.109*** (1.394) (0.993) (3.334)cen_dip_dip# -0.055 -0.350*** 0.102*** (1.255) (3.237) (7.991)cen_dip_bach 0.039*** 0.021 0.040*** (3.230) (0.932) (2.613)cen_ascdeg_y10 -0.312*** -0.275*** -0.232** (5.548) (3.523) (2.276)cen_ascdeg_y12# -0.178*** -0.063* -0.197*** (7.608) (1.953) (6.408)cen_ascdeg_cert# -0.035*** -0.018 -0.165*** (3.014) (1.514) (3.989)cen_ascdeg_dip -0.258*** -0.549 0.016 (2.979) (1.169) (1.358)cen_ascdeg_bach -0.008 -0.029 -0.018 (0.578) (0.900) (0.821)cen_bach_y10 -0.292*** -0.263*** -0.285*** (30.049) (22.118) (14.824)cen_bach_y12# -0.132*** -0.102*** -0.141*** (48.965) (21.102) (39.460)cen_bach_cert# -0.109*** -0.040*** -0.085*** (22.742) (9.197) (17.352)cen_bach_dip# -0.028*** -0.207*** 0.018 (3.453) (10.521) (1.393)cen_hons_y10# -0.284*** -0.217*** -0.379*** (8.568) (5.463) (5.795)cen_hons_y12# -0.081*** -0.054*** -0.098*** (12.468) (5.139) (11.080)cen_hons_cert# -0.061*** -0.016 -0.046*** (4.753) (1.485) (3.729)cen_hons_dip# -0.026 -0.011 0.117** (1.020) (0.246) (2.416)cen_hons_bach 0.022*** 0.028*** 0.023*** (6.419) (5.564) (4.605)cen_gcert_y10 -0.228*** -0.224*** -0.212* (4.958) (4.480) (1.919)cen_gcert_y12# -0.017* 0.019 -0.055*** (1.866) (1.387) (3.837)cen_gcert_cert# -0.000 0.131*** 0.031* (0.001) (16.014) (1.820)cen_gcert_dip# 0.040 -0.218*** 0.066 (1.441) (3.240) (0.907)cen_gcert_bach# 0.145*** 0.115*** 0.155*** (40.735) (15.438) (34.943)cen_gdip_y10 -0.318*** -0.362*** -0.264*** (8.523) (7.801) (3.780)cen_gdip_y12# -0.064*** -0.046*** -0.086*** (8.260) (3.327) (7.733)cen_gdip_cert# -0.018 0.148*** -0.013 (1.255) (16.097) (1.003)



Table 2 - OLS Estimates of the ORU Model, Gender-specific Required Education (continued)

Full Males FemalesVariable (i) (ii) (iii)cen_gdip_dip# 0.001 -0.108*** 0.118*** (0.038) (3.539) (4.831)cen_gdip_bach# 0.103*** 0.082*** 0.099*** (36.836) (15.462) (29.132)cen_mast_y10 -0.351*** -0.354*** -0.317*** (11.979) (10.111) (6.113)cen_mast_y12# -0.048*** -0.010 -0.082*** (6.525) (0.867) (6.963)cen_mast_cert# 0.055*** 0.238*** 0.065*** (4.732) (38.381) (5.670)cen_mast_dip# 0.088*** -0.140*** 0.060* (4.820) (3.711) (1.955)cen_mast_bach# 0.201*** 0.163*** 0.191*** (71.672) (32.360) (51.265)cen_phd_y10 -0.256* -0.283*** -0.370 (1.823) (2.956) (0.933)cen_phd_y12 0.107*** 0.077** 0.132*** (6.304) (2.365) (5.198)cen_phd_cert# 0.096** 0.228*** 0.145*** (2.430) (13.918) (6.343)cen_phd_dip 0.188*** -0.020 0.534 (4.476) (0.169) (1.066)cen_phd_bach# 0.192*** 0.172*** 0.190*** (38.078) (22.609) (26.514)Observations 569,325 221,746 347,579AdjustedR2 0.186 0.215 0.162

Notes:Absolutevaluesofheteroscedasticityconsistent‘t’-statisticsarepresentedinparentheses.*,**and***indicatesignificanceattheten,fiveandonepercentlevels,respectively.Themodelsincludeddummyvariablesfor15industriesofemployment,andforthe10yearsofgraduation.#indicatessignificanceforthet-testofdifference.(a)indicatesthatthevariablewasnotenteredintheestimatingequation.#indicatesstatisticaldifferenceformalesandfemales,atthetenpercentlevel.Themiddleportionsofthecenvariablesindicatetheattainedlevelofqualification,andtheendportionsindicatethejoblevel.

The adjustedR2 for the analysis of the full sample is 0.186,which is very

similartotheearlieranalysisreportedintable1.TheadjustedR2valuesforthemaleandfemaleestimationsinpanels(ii)and(iii)are0.215and0.162,respectively.Thus,thereisnoadvantage,andperhapsevenaslightdisadvantage,fromusingtheadditionaldetailavailableinthesealternativemeasuresofrequirededucation.Atfacevalue,thissuggeststhatthelabourmarketisnotoverlydiscerninginthisregard.However,theresultsofthe‘t’-testsofdifferencesintheestimatedearningscoefficientsbygenderrevealedamuchlargernumberofstatisticallysignificantearningseffectsdifferencesinthetable2analyses,ascomparedtothoseintable1.Theseareagaindenotedbythe # beside the variables’ names.Out of the 87 variables in themodel, estimatedcoefficientsfor41ofthem,oralmosthalf,werefoundtodifferstatisticallybygender.Areviewoftheestimatedcoefficientsinpanels(ii)and(iii)revealsthatthemagnitudesofearningseffectsaregenerallylargerformales,andinsomecasesarestatistically


significantformalesbutnotforfemales.Thisissimilartothepatternevidentintable1.Looking at fields of study, for example, it is observed that the estimates for theInformationTechnologyandEngineeringgraduatesarestatisticallysignificantattheonepercentlevelformales,butareinsignificantforfemales.Moreover,theestimatesonAgricultureandEnvironment,Nursing,andCreativeArtsandOthersarelarger(inabsoluteterms)formales.

The analyses disaggregated by gender, and using gender-specific modallevelsofqualifications,areassociatedwithgreatervariability in theORUearningsimpactsbetweenmalesandfemales.Thus,usingthe‘t’-testsofdifference,26ofthe39ORUvariablesheredifferstatisticallybetweenmalesandfemales.Thisisalargerproportionthanthatfoundfortheearlieranalyseswhichdidnotusegender-specificrequiredlevelsofeducation,andthisispresumablylinkedtothegreaterdetailusedintheconstructionoftheORUvariablesinlower-skilledjobs.Themagnitudeoftheearningsdifferencesbygenderaresubstantiallylarger,comparedtothosefoundintheearliersection.Forexample,maleswithadiploma,employedinajobwithamodaleducationallevelofaYear10qualification,earnedasubstantial23percentlessthanthebenchmarkgroupoftheirmalecounterpartswithabachelor’spassdegreeworkinginajobwherethemodalqualificationisabachelor’spassdegree.15Incomparison,females are much worse off if they are in the same situation, with the earningseffect being negative 52 per cent. Earnings effects differences betweenmales andfemalesexceedingtenpercentagepointsareevidentin18categories,thoughthesearereasonablyevenlydividedbetweencaseswheremalesareatanearningsadvantageandcaseswherefemalesareatanearningsadvantage.

Further,ofthe39ORUearningscoefficientsinthissection,only17exhibitedgenderdifferencesinearningimpactsthatareconsistentwiththejobsearchhypothesis.Whilethisisalargerproportioncomparedtothatfoundintheprevioussection(3outof23),itstillaccountsforlessthanhalfoftheestimatedORUearningscoefficients.Generally,itcanbesaidthatthejobsearchhypothesisdoesnotappeartobevalidatedbythefindingsherefortheAustraliangraduatelabourmarket.

Blinder-Oaxaca Decomposition Asnotedinrelationtoequation(2),aBlinder-Oaxacadecompositioncanbeusedtoprovideabetterunderstandingofthereasonswhyfemaleshavealowermeanrateofpay thanmales.Thefindingsfromthisdecomposition,basedon theaverageof theresultsobtainedwhenthefemaleandmalewagestructuresareemployedasthenon-discriminatorynorm,arepresentedintable3.Inthecurrentdatathereisarawgenderwagedifferentialof9.6percentagepointsinfavourofmales.Ofthis9.6percentagepoints, theBlinder-Oaxacadecomposition revealed that4.4percentagepointswereattributable to thedifference in theendowmentsof themaleandfemalegraduates.Theremaining5.1percentagepoints,orslightlyoverhalfofthewagedifference,canbeattributed to thedifference incoefficients.16,17TheORUvariablesaccounted for15ThisORUcategoryconsistsoffarmmanagers,andproductionortransportlabourers.16This‘coefficient’effectisalsoknownasthe‘unexplained’wagedifferential,ordiscrimination,inthegenderwagegapliterature.17Atwo-folddecompositionisperformedhere,forsimplicityinthediscussionofresults,aswellascompatibilitywithmoststudiesintheeconomicsliterature.Performingathree-folddecompositionrevealsthatthe‘third’interactioncomponentissmall,at2.5percentagepoints.Theendowmentandcoefficienteffectswere3.2and3.9percentagepoints,respectively.



acombined,andverysmall,0.3percentagepointsofthe‘endowment’effect.18Themajority(32outof39)oftheestimatedendowmenteffectsfortheORUvariableswere,however,statisticallysignificantatthe10percentlevelorhigher.

Table 3 - Estimates from the Blinder-Oaxaca Decomposition

PredictedMaleWage 3.0658*** (0.0013)PredictedFemaleWage 2.9701*** (0.0010)RawWageGap 0.0957*** (0.0017)Explained 0.0443*** (0.0012)Unexplained 0.0514*** (0.0018)Constant -0.0788*** (0.0221)Observations 569,325

Note: Standarderrorsarepresentedinparentheses.***,**and*denotesignificanceattheone,fiveandtenpercentlevels,respectively.

ThenegligiblecombinedendowmenteffectfortheORUvariablesmayseem

inconsistentwiththefindingsreportedearlier,ofmalesbeinglesslikelytobe‘matched’totheiroccupation(30percentincidenceofeducation-occupationmatchcomparedwith38percentforfemales)andoftherebeingsubstantialvariationinwagesacrossthe ORU categories entered into the estimating equation. An examination of theindividualendowmenteffectsforthe39ORUvariablesshowsthattheseeffectsareminute,with the largestestimatebeingonly0.7percentagepoints, for thecategoryofgraduateswithamastersdegreeworkingincertificate level jobs.19Moreover,21oftheORUestimatedcoefficients,oraroundone-halfoftheORUvariables,wereofnegativesign,while theremaining18wereofpositivesign.Anegativesignmeansthattheremovalofthatcomponentwouldleadtoawidergenderwagegap,whereasapositivesignindicatesthatremovalofthatcomponentwouldlessenthegenderwagegap,ceterisparibus.Thus,theseeffectscancelout,withthenetresultbeingthattheORUendowment effectsdonot favour eithergender.Thisfindingprovides abasisfor further evaluation of the job search hypothesis. Specifically, as there does notappeartobeaclearwageadvantage(detriment)causedbytheORUendowmentsformales(females),thejobsearchhypothesisisnotvalidated,atleastfortheAustraliangraduatelabourmarket.Itwouldbeofinterest,however,toknowifthesefindingsholdafterthemeasurementissuesthatwerehighlightedintheliteraturereviewsectionareaccountedfor.Theseissuesareaddressedinthefollowingsection.

18Fortheunexplained‘coefficient’component,ORUeffectsaccountforamodest1.5percentagepoints,outofthe5.2percentagepoints.19Theindividualendowmenteffectsarenotreportedhere,butareavailableonrequest.


The Blinder-Oaxaca Decomposition and the ‘Averaging Approach’ As an extension to the Blinder-Oaxaca decomposition analysis from the precedingsection, andbearing inmind themeasurement issues raised in the literature reviewsectionearlier,equation(2)isre-estimatedutilisingthe‘averagingapproach’suggestedby Yun (2005). Selected results from this decomposition are presented in table 4.Panels(i)and(ii)presentstheresultsfromthepreviousdecompositiondiscussedintheprecedingsection,whilepanels(iii)and(iv)presenttheresultswhichhaveincorporatedYun’s(2005)‘deviantcontrastcoding’or‘averagingapproach’.Notethatastheresultsoftheoveralldecompositiondonotchange,theywillnotbepresentedhere.Instead,thefocuswillbeontheORUeffectsandthechangeintheconstantterms.

Table 4 - Selected Results from the Blinder-Oaxaca Decomposition under the Averaging Approach

Single Benchmark Averaging Approach Explained Unexplained Explained UnexplainedVariable (i) (ii) (iii) (iv)cen_bach_bach -0.0042*** 0.0098** (0.0005) (0.0045)cen_dip_y10 -0.0001*** 0.0001*** -0.0001*** 0.0001*** (0.0000) (0.0000) (0.0000) (0.0000)cen_dip_y12 0.0001*** 0.0001 0.0000*** 0.0001* (0.0000) (0.0001) (0.0000) (0.0001)cen_dip_cert -0.0002** 0.0003*** 0.0000 0.0004*** (0.0001) (0.0001) (0.0001) (0.0001)cen_dip_dip 0.0001** -0.0002*** 0.0000 -0.0002*** (0.0000) (0.0000) (0.0000) (0.0000)cen_dip_bach 0.0000 -0.0000 0.0000 0.0000 (0.0000) (0.0001) (0.0000) (0.0001)cen_ascdeg_y10 -0.0001*** -0.0000 -0.0000** -0.0000 (0.0000) (0.0000) (0.0000) (0.0000)cen_ascdeg_y12 0.0001*** 0.0001** 0.0000** 0.0001*** (0.0000) (0.0000) (0.0000) (0.0000)cen_ascdeg_cert -0.0005*** 0.0005*** -0.0002* 0.0006*** (0.0001) (0.0001) (0.0001) (0.0001)cen_ascdeg_dip 0.0003*** -0.0003*** 0.0002*** -0.0003*** (0.0001) (0.0001) (0.0001) (0.0001)cen_ascdeg_bach -0.0000 -0.0000 0.0000 0.0000 (0.0000) (0.0001) (0.0000) (0.0001)cen_bach_y10 -0.0040*** 0.0003 -0.0032*** 0.0006** (0.0001) (0.0002) (0.0002) (0.0003)cen_bach_y12 0.0084*** 0.0049*** 0.0044*** 0.0086*** (0.0002) (0.0008) (0.0005) (0.0017)cen_bach_cert -0.0040*** 0.0038*** -0.0003 0.0062*** (0.0002) (0.0005) (0.0004) (0.0012)cen_bach_dip -0.0006*** -0.0013*** -0.0002*** -0.0011*** (0.0001) (0.0001) (0.0001) (0.0002)cen_hons_y10 -0.0003*** 0.0002*** -0.0003*** 0.0002*** (0.0000) (0.0001) (0.0000) (0.0001)cen_hons_y12 0.0005*** 0.0006*** 0.0001* 0.0009*** (0.0001) (0.0002) (0.0001) (0.0002)cen_hons_cert -0.0002*** 0.0003* 0.0002** 0.0005*** (0.0001) (0.0002) (0.0001) (0.0002)cen_hons_dip 0.0000 -0.0001 0.0001** -0.0000 (0.0000) (0.0000) (0.0000) (0.0000)



Table 4 - Selected Results from the Blinder-Oaxaca Decomposition under the Averaging Approach

Single Benchmark Averaging Approach Explained Unexplained Explained UnexplainedVariable (i) (ii) (iii) (iv)cen_hons_bach 0.0002*** 0.0003 0.0008*** 0.0016** (0.0000) (0.0004) (0.0001) (0.0007)cen_gcert_y10 -0.0001*** -0.0000 -0.0001*** 0.0000 (0.0000) (0.0000) (0.0000) (0.0000)cen_gcert_y12 -0.0000 0.0005*** 0.0000* 0.0006*** (0.0000) (0.0001) (0.0000) (0.0001)cen_gcert_cert 0.0014*** 0.0013*** 0.0023*** 0.0016*** (0.0001) (0.0002) (0.0002) (0.0003)cen_gcert_dip -0.0000* -0.0001*** -0.0000 -0.0001*** (0.0000) (0.0000) (0.0000) (0.0000)cen_gcert_bach -0.0027*** -0.0015*** -0.0039*** -0.0004 (0.0001) (0.0003) (0.0002) (0.0006)cen_gdip_y10 -0.0003*** -0.0001* -0.0002*** -0.0001 (0.0000) (0.0000) (0.0000) (0.0000)cen_gdip_y12 0.0002*** 0.0004** 0.0000 0.0006*** (0.0000) (0.0002) (0.0000) (0.0002)cen_gdip_cert 0.0009*** 0.0021*** 0.0016*** 0.0025*** (0.0001) (0.0002) (0.0001) (0.0003)cen_gdip_dip 0.0000 -0.0001*** 0.0000* -0.0001*** (0.0000) (0.0000) (0.0000) (0.0000)cen_gdip_bach -0.0030*** -0.0012** -0.0049*** 0.0008 (0.0001) (0.0005) (0.0003) (0.0010)cen_mast_y10 -0.0012*** -0.0001 -0.0010*** -0.0000 (0.0001) (0.0001) (0.0001) (0.0001)cen_mast_y12 -0.0002*** 0.0012*** 0.0000 0.0017*** (0.0000) (0.0002) (0.0000) (0.0003)cen_mast_cert 0.0079*** 0.0061*** 0.0109*** 0.0071*** (0.0003) (0.0004) (0.0004) (0.0006)cen_mast_dip -0.0000 -0.0002*** 0.0000 -0.0002*** (0.0000) (0.0001) (0.0000) (0.0001)cen_mast_bach -0.0017*** -0.0027*** -0.0022*** 0.0000 (0.0001) (0.0006) (0.0002) (0.0013)cen_phd_y10 -0.0000** 0.0000 -0.0000** 0.0000 (0.0000) (0.0000) (0.0000) (0.0000)cen_phd_y12 0.0000** -0.0001 0.0000** -0.0000 (0.0000) (0.0000) (0.0000) (0.0000)cen_phd_cert 0.0008*** 0.0002** 0.0010*** 0.0003*** (0.0001) (0.0001) (0.0001) (0.0001)cen_phd_dip 0.0000 -0.0000 0.0000* -0.0000 (0.0000) (0.0000) (0.0000) (0.0000)cen_phd_bach 0.0013*** -0.0004* 0.0017*** 0.0002 (0.0001) (0.0003) (0.0001) (0.0004)Total 0.0030 0.0149 0.0067 0.0328Constant -0.0788*** -0.1296*** (0.0221) (0.0249)Observations 569,325 569,325Notes:Standarderrorsarepresentedinparentheses.*,**and***denotesignificanceattheten,fiveandonepercentlevels,respectively.Panels(iii)and(iv)presentresultsofthetwo-folddecompositionunderthe‘averaging’or‘deviationcontrastcoding’method.


Thefirstkeyobservation thatcanbemade is that thevalueof the interceptterm is markedly different. The estimate for the intercept term in the previousdecompositionwasaround7.9percentagepoints,whilethecorrespondingestimateforthecurrentdecompositionwithdeviationcontrastcodingis13percentagepoints.Thissuggeststhatgroupmembershipperseismoreimportantincontributingtothegenderwagegapthanthatindicatedfromthepreviousdecomposition,wheretheconstanttermreferredtotheoutcomesforasinglereferencegroup,namely,thebachelor’spassdegreegraduatesworkinginjobsthatrequirethelevelofqualificationthattheypossess.

LookingattheestimatedcoefficienteffectsassociatedwiththeORUvariables,however,indicatesverymodestchanges.31outofthe40ORUvariableshavecoefficienteffectsinthedecompositionthatarestatisticallysignificantatthetenpercentorhigherlevels,similar to the32outof39significantORUcoefficienteffectsreportedunderthe‘singlebenchmark’approachfromtheprevioussection.Thesignontheestimatedcoefficienteffectslargelyremainsunchanged–theonlyexceptionisforgraduateswithanhonoursdegreeworking incertificate level jobs.Theendowmenteffect for thesegraduatesalsochanged,frombeingofanegativesigntobeingpositive.Theabsolutevalueoftheendowmenteffect,however,remainssmall,at0.2percentagepoints.

Further,anexaminationof theendowmenteffects in thedecompositionforthedifferentlevelsofrequirededucationrevealsfourgeneral(thoughnotuniversal)patterns.First,therearefewsignificantgenderendowmenteffectsinjobsthatrequireadiploma,and thegenderendowmenteffects in jobs that requireabachelor’spassdegreearemixed.Asthesearejobrequirementsthatareclosesttothequalificationsofthegraduatepopulation,thisempiricalresultindicatesthatsimilarsortingoutcomesformaleandfemaleuniversitygraduatesoccurforsuchjobs.

Second, jobs that require certificate level qualifications have endowmenteffectsthatfavourmorehighlyqualifiedmales.Thatis,fewermorehighlyqualifiedmalesthanfemalesworkinthesejobsthatattractawagepenalty.Whenthisresultiscombinedwiththefirstfeaturementionedabove,theimplicationisthatmaleandfemalegraduatesaredifferentiallysortedintojobsthatrequirelower-levelqualifications.

Third,jobsthatrequireYear12educationtypicallyhaveendowmenteffectsthatfavourmales.Asgraduatesinthesejobsareovereducated,andtheirover-educationstatusisassociatedwithlowerwages,thisendowmenteffectinfavourofmalesmustarisebecausefemalesaremorelikelythanmalestobeintheseintermediateleveljobs.

Fourth,jobsthatrequireYear10educationtypicallyhaveendowmenteffectsthatfavourfemales.Applyingthereasoningadvancedabove,thissuggeststhatmalesaremorelikelythanfemalestobeintheselow-skilledjobs.

Inthecaseofthecoefficientseffect,therearethreefindingsofnote.First,thecoefficienteffectsforjobsthatrequireeitheradiplomaorbachelor’spassdegreetendtobeassociatedwithanegativeeffect.Inotherwords, thiscomponentofthewagedecompositionactstolessenthemalewageadvantagethatwouldotherwiseoccur.

Second,thecoefficienteffectsforjobsthatrequireeitheracertificateorYear12 schooling tend to be positive. In other words, the over-education wage effectsassociatedwithuniversitygraduatesworkinginthesejobstendtofavourmales,andleadtoawideningoftheirwageadvantage.

Third, the low-skilled jobs requiring onlyYear 10 are typically associated



with similar wage effects formales and females, so that the coefficient effects inthewagedecompositionforthesejobsareusuallynotstatisticallysignificant.Thus,thedecompositionindicates thatwhile theoverallwageeffectsareslight, thereareinterestingpatternsinthedatathatsuggestthattherearesystematicfactorsimpactingthewagedeterminationprocesswhenitisexaminedfromtheORUperspective.

A comparison of the overall contribution of ORU earnings effects to thegenderwagegaprevealschangesofamoderatescale.RecallfromtheprevioussectionthatORUearningseffectsaccountedfora total0.3percentagepoint impactontheendowmenteffect.Further, theunexplainedportionof thegenderwagegap,or thecoefficient effect, attributed to ORU effects was 1.5 percentage points. Under the‘averagingapproach’,thesevaluesare0.7and3.3percentagepoints,respectively.Theendowmentandcoefficienteffectsof thegenderwagegapattributable toORUcanthusbe said tohavedoubled. In thecaseof theendowmenteffects,ORUearningseffectscanbesaidtoplayaminorrole-theystillaccountforonly17percentoftheoverall endowment effect for themodel. The coefficient effect of the genderwagegapattributable toORU,however, is substantial.ORUearningseffects account forroughly two-thirds of the (modest) estimated coefficient effect of five percentagepoints. Nevertheless, the finding from the previous section, that the ‘job-search’hypothesis is not validated in theAustralian graduate labourmarket, is reinforcedby the decomposition utilisingYun’s (2005) ‘averaging’ approach, given the smallendowmenteffectsassociatedwithORU.

The Gender Wage Gap and Age TheBlinder-Oaxacadecompositionindicatedastandardisedgenderwagegapof5.2percentagepoints.Thisgenderwagegap isofsimilarsize to thatestimatedby thefemale dummyvariable in theORUmodels of earnings in the preceding sections.Further,recallthatinanearliersectionitwasnotedthattherelativelysmallgenderwagegapobservedinthepresentstudymightbeduetothefocusonlabourmarketentrants.Thus,thepooledregressionwillbeusedinamoredetailedexaminationofthegenderwagegapeffectbyage.Twoapproachesareconsideredinthisexamination.First, thesamplewasdisaggregatedbyagegroups,andequation (3)wasestimatedseparatelyforeachagegroup.20Theresultsindicatedthatforgraduatesaged35yearsandbelow,thegenderwagegapwasaround4.5percent.Thereafter,thegenderwagegapwidenedconsiderably.Femalesaged36to40yearsearned6percentless.Thoseaged46to50yearsexperiencedearnings8percentlowerthantheirmalecounterparts.Femaleworkersagedmorethan56yearsoldearned13percentless.Theseestimateswereallsignificantattheonepercentlevel.Asageincreases,bothmalesandfemalesenterhigher-wagepositions,butthefemalewagedisadvantagewidens.Thisprovidessomesupportforthe‘glassceiling’foundinotherstudies.

Second,equation(3)wasestimatedonthefullsample,withaninteractiontermbetweengenderandage(genage=female*age).Theinclusionofthisterminthemodelyieldedanestimateof-0.013forfemale,andanestimateof-0.004ongenage,bothsignificantattheonepercentlevel.Theseestimatescanbeinterpretedasfollows.20 The results for the fullmodel on these analyses by age groups, and the subsequent pooledregressionarenotpresentedinthisstudy,butareavailableonrequest.


Thegenderwagegapisthreepercentwhenevaluatedat25yearsofage.Forgraduatesaged40yearsold,thegenderwagegapismuchwider,atninepercent.Thisincreaseinthegenderwagegapfollowsthroughforincreasingyearsofage,andthegapisasubstantial19percentforgraduatesatthe(retirement)ageof65years.

Thesefindingslendsupporttothethesisthattherelativelysmallgenderwageeffectamongyoungerworkers,andthelargergenderwageeffectamongolderworkers,isduetothemeasureofworkexperience.MincerandPolachek(1974)fortheUS,andRummery(1992)forAustralia,showthatchangingfromameasureofpotentialworkexperiencetoameasureofactualworkexperiencecanreducethestandardisedwagegapby40to70percent.21Thecompetinghypothesis,thatthesmallgenderwagegapforyounggraduatesisduemainlytominimumwageeffects,doesnotseemcrediblewhenthegradualchangesinthegenderwagegapwithageareconsidered.

5. Conclusion This paper has examined gender differences and educational mismatch in theAustraliangraduatelabourmarket,usingvariousanalysesandperspectives.Anumberof conclusionscanbedrawn from this analysis.First, thegenderwagegap for thehighereducatedlabourmarketentrantsissmallerthanthatreportedinotherAustralianstudies.Analysesofthechangeinthiswithagesuggestthatthischangeismostlikelylinkedtothemeasureofworkexperienceincludedintheestimatingequation.

Second, themost substantial penalties to being overeducated are found atthelowest joblevels.MostORUearningseffectsdonotdifferstatisticallybetweenmales and females. Greater earnings penalties and gender differences are foundwhengender-specificandmoredetailed required levelsofeducationareused fromtheCensusdatathanwhenthegender-neutralASCO-basedstandardsareemployed.Nevertheless, the absence of evidence in either set of analyses that females incurgreater earnings penalties thanmales from their overeducated status suggests thatfemales’over-educationdoesnotariseduetotheirmorelimitedjobsearch.

Athirdfindingreinforcesthatofthesecondpoint.Addingbroadcontrolsforoccupationtothemodelimpactsnegatively(positively)ontheORUearningseffectsfor female (male) graduates. This indicates that females are more mobile acrossoccupationscomparedtomales.Again,thisdoesnotsupportthetheorisedoutcomesunderthejobsearchhypothesis.

Fourth,theBlinder-OaxacadecompositionrevealedthatORUeffectsaccountedfor only a negligible portion of the gender wage gap. However, the decompositionrevealed interesting trends regarding sorting outcomes and ORU earnings effectsformalesandfemales.Withregardtojobsthatrequirelowereducationlevels,malesweremorelikelythanfemalestobesortedintojobsrequiringYear10schoolingorcertificates.At thesame time, thecoefficienteffects in thedecomposition indicatedthattheestimatedORUeffectsforhigher-leveljobsthatrequireadiplomaorbachelor’spassdegreetendtonarrowthegenderwagegap.Incontrast,theestimatedORUeffectsforjobsthatrequireacertificateorYear12tendtowidenthegenderwagegap.

Insummary, there isagenderwagegap in theAustraliangraduate labour21InRummery(1992)themeasureofactualexperiencewasconstructedasthenumberofyearsworkedfulltimeplusathirdoftheyearsworkedparttime.Alldatawerecollectedretrospectively.



market,thoughthisgapissmallerthanthatfoundfortheaggregate-levelAustralianlabour market. These findings thus favour education as a tool of eliminatingdiscrimination in the labourmarket.As females are less overeducated thanmales,despitethelargerrepresentationoftheformerinhighereducation,thereshouldnotbeconcernthatexpandinghighereducationwilldisadvantagefemales.ThisprognosisisreinforcedbythefindingthatthemajorityoftheestimatedORUpenaltiesdonotdifferstatisticallybetweenmalesandfemales,andthefindingthatdifferentlevelsofover-educationofmalesandfemalesmakeaminutecontributiontothe‘endowmenteffect’ in thegenderpaygapdecomposition.At thesametime,however,awordofcautionisneeded.Thegenderwagegapislargerforgraduatesintheolderagegroupsandwhoareinmoreadvancedstagesoftheircareer.This‘glassceiling’effectappearssubstantial.However,whetheritisapure‘glassceiling’effect,orsimplyastatisticalartefactattributabletotheuseofapoormeasureofworkexperienceintheearningsequation,isamootpoint.Thecollectionofdetailedworkhistorieswillbeneedediftheunderstandingofthislabourmarketoutcomeisapriority.

Appendix 1 Table A1 - Summary Statistics and Description of Explanatory Variables

Variable Mean Std. Dev.Dependent VariableLoghourlywage=Hourlywage,expressedinlogarithmicformat 3.007 0.620University GroupGroupofEight=Go8university 0.279 0.449AustraliaTechnologyNetwork=ATNuniversity 0.190 0.393InnovativeResearchUniversity=IRUuniversity 0.130 0.336Other=Otheruniversity(omittedcategory) 0.400 0.490Personal CharacteristicFemale=Femalegraduates(omittedcategory=Malegraduates) 0.611 0.488Age=Age,expressedinyears 29.409 9.428Agesquared=Agesquared,expressedinyears 953.792 673.722NESB=Non-Englishspeakingbackground(omittedcategory=English-speakingbackgound) 0.174 0.379Non-Australian=NoAustralianresidencystatus(omittedcategory=Australianresidencystatus) 0.045 0.208Study CharacteristicsDoubledegree=Doubledegreequalification(omittedcategory=Nodoubledegree) 0.091 0.287Part-timestudy=Studiedonapart-timebasis(omittedcategory=Studiedfull-time) 0.353 0.478Furtherstudy=Engagedinfurtherstudy(omittedcategory=Nofurtherstudy) 0.196 0.397Broad field of studyNaturalandPhysicalSciences 0.062 0.241InformationTechnology 0.050 0.217Engineering 0.054 0.226ArchitectureandBuilding 0.021 0.142AgricultureandEnvironment 0.021 0.144Nursing 0.071 0.258


Table A1 - Summary Statistics and Description of Explanatory Variables (continued)

Variable Mean Std. Dev.Medicine 0.100 0.300Education 0.145 0.353SocietyandCulture 0.171 0.376CreativeArtsandOthers 0.063 0.242ManagementandCommerce(omittedcategory) 0.242 0.428Employment CharacteristicsSelf-employed=Self-employed(omittedcategory=Notself-employed) 0.039 0.193Privatesector=Employedinprivatesector(omittedcategory=Publicsector) 0.597 0.490Short-termemployment=Employmentoncontractslessthanoneyear,oronacasualbasis(omittedcategory=Long-termemployment) 0.306 0.461Tenure=Jobtenure,expressedinyears 2.371 3.998Tenuresquared=Jobtenuresquared,expressedinyears 21.606 80.220Industry of EmploymentAccounting 0.034 0.180Wholesaleandretail 0.078 0.268Accommodation 0.033 0.178Manufacturing 0.041 0.198Forestryandmining 0.011 0.105Legalservices 0.024 0.153Government 0.092 0.289Education 0.145 0.352Highereducation 0.063 0.243Healthandcommunityservices 0.158 0.365Medicineanddentistry 0.040 0.196Construction 0.013 0.112Otherservices 0.077 0.267Transportandcommunication 0.031 0.173Engineeringconsulting 0.021 0.143Financialservices(omittedcategory) 0.139 0.345Year of Graduation1999=Graduatedin1999(omittedcategory) 0.075 0.2642000=Graduatedin2000 0.078 0.2682001=Graduatedin2001 0.065 0.2472002=Graduatedin2002 0.069 0.2532003=Graduatedin2003 0.085 0.2792004=Graduatedin2004 0.095 0.2942005=Graduatedin2005 0.098 0.2972006=Graduatedin2006 0.089 0.2842007=Graduatedin2007 0.109 0.3112008=Graduatedin2008 0.116 0.3212009=Graduatedin2009 0.120 0.325

Note:Valuesofmeansinsomecategoriesmaynotsumtounityduetorounding.



References Australian Bureau of Statistics (2004), ‘Education & Work: Higher Education

GraduatesintheLabourMarket’,AustralianBureauofStatistics.Availablefrom: http://www.abs.gov.au/AUSSTATS/[email protected]/2f762f95845417aeca25706c00834efa/e1a27d207c960e79ca256e9e00286295!OpenDocument [17 June2011].

AustralianBureauofStatistics(2011),‘ASCO–AustralianStandardClassificationofOccupations’,AustralianBureauofStatistics.Availablefrom:http://www.abs.gov.au/ausstats/>[17June2011].

Battu,H.,Belfield,C.R.andSloane,P.J.(2000),‘HowWellCcanWeMeasureGraduateOver-EducationanditsEffects’,National Institute Economic Review,171(1),82-93.

Blinder,A.S.(1973),‘WageDiscrimination:ReducedFormandStructuralEstimates’,The Journal of Human Resources,8(4),436-455.

Borland, J. (1999), ‘TheEqualPayCase–ThirtyYearsOn’,Australian Economic Review,32(3),265-272.

Buchel,F.andBattu,H.(2003),‘TheTheoryofDifferentialOverqualification:DoesitWork?’,Scottish Journal of Political Economy,50(1),1-16.

Chiswick,B.R.(1978),‘TheEffectofAmericanizationontheEarningsofForeign-bornMen’,Journal of Political Economy,86(5),897-921.

Chiswick.B.R.andMiller,P.W.(2008),‘WhyisthePayofftoSchoolingSmallerforImmigrants?’, Labour Economics,15(6),1317-1340.

Chiswick. B.R. andMiller, P.W. (2010), ‘Does the Choice of Reference Levels ofEducationMatterintheORUEarningsEquation’,Economics of Education Review,29(6),1076-1085.

Cotton,J.(1988),‘OntheDecompositionofWageDifferentials’,Review of Economics and Statistics,70(2),236-243.

deOliveira,M.M.,Santos,M.C.andKiker,B.F.(2000),‘TheRoleofHumanCapitaland Technological Change in Over-education’, Economics of Education Review,19(2),199-206.

Dolton,P.J.andVignoles,A.(2000), ‘TheIncidenceandEffectsofOver-educationintheUKGraduateLabourMarket’,Economics of Education Review,19(2),179-198.

Duncan,G. andHoffman, S.D. (1981), ‘The Incidence andWage Effects ofOver-education’,Economics of Education Review,1(1),75-86.

Ferber,M.A. andGreen,C.A. (1982), ‘Traditional orReverse SexDiscrimination?ACaseStudyofaLargePublicUniversity’,Industrial and Labor Relations Review,35(4),550-564.

Fortin,N.M.(2008),‘TheGenderWageGapamongYoungAdultsintheUnitedStates:TheImportanceofMoneyversusPeople’,The Journal of Human Resources,43(4),884-918.

Frank,R.H.(1978),‘WhyWomenEarnLess:TheTheoryandEstimationofDifferentialOverqualification’,The American Economic Review,68(3),360-373.

Freeman,R.B.(1976),The Over-educated American,AcademicPress,NewYork.


Gardeazabal, J. and Ugidos, A. (2005). ‘More on Identification in DetailedWageDecompositions’,Review of Economics and Statistics,86(4),1034-1036.

Goldberger,A.S.(1984),‘ReverseRegressionandSalaryDiscrimination’,The Journal of Human Resources,19(3),293-318.

Gregory, B. (1999), ‘Labour Market Institutions and the Gender Pay Ratio’, The Australian Economic Review,32(3),273-278.

Guthrie, B. and Johnson, T. (1997), Study of Non-response to the 1996 Graduate Destination Survey, Department of Employment, Education, Training andYouthAffairs,Canberra,Australia.Availablefrom:<http://www.dest.gov.au/archive/highered/eippubs/eip9710/front.htm#contents>[1November2011].

Hartog,J.(2000),‘Over-educationandEarnings:WhereAreWe,WhereShouldWeGo?’,Economics of Education Review,19(2),131-147.

Jones,F.L.(1983),‘OnDecomposingtheWageGap:ACriticalCommentonBlinder’sMethod’,The Journal of Human Resources,18(1),126-130.

Jones, F.L. and Kelley, J. (1984), ‘Decomposing Differences between Groups: ACautionaryNote onMeasuringDiscrimination’,Sociological Methods and Research,12(3),323-343.

Kamalich, R.F. and Polachek, S.W. (1982), ‘Discrimination: Fact or Fiction? AnExaminationUsinganAlternativeApproach’,Southern Economic Journal,49(2),450-461.

Kapsalis,C.(1982),‘ANewMeasureofWageDiscrimination’,Economics Letters,9,287-293.

Kee,H.J.(2006),‘GlassCeilingorStickyFloor?ExploringtheAustraliaGenderPayGap’,The Economic Record,82(259),408-427.

Kler,P.(2005), ‘GraduateOver-education inAustralia:AComparisonof theMeanandObjectiveMethods’,Education Economics,13(1),47-72.

Le,A.T.andMiller,P.W.(2002),‘TheRisingEducationLevelofFemalesinAustralia’,Education Economics,10(1),1-24.

Leuven,E.andOosterbeek,H.(2011),‘Over-educationandMismatchintheLaborMarket’,InstitutefortheStudyofLaborDiscussionPaperNo.5523,Bonn,Germany.

Li, I.W. (2013), Over-education in the Australian Graduate Labour Market,unpublishedPhDthesis,CurtinUniversity,Perth.

McGoldrick,K.andRobst,J.(1996),‘GenderDifferencesinOver-education:ATestoftheTheoryofDifferentialOverqualification’,Economic Behavior and the Family,86(2),280-284.

McGuinness,S.(2006),‘Over-educationintheLabourMarket’,Journal of Economic Surveys,20(3),387-418.

McGuinness, S. and Bennett, J. (2007), ‘Over-education in the Graduate LabourMarket:AQuantileRegressionApproach’,Economics of Education Review,26(5),521-531.

Miller,P.W.(1994),‘EffectsonEarningsoftheRemovalofDirectDiscriminationinMinimumWageRates:AValidationoftheBlinderDecomposition’,Labour Economics,1(3/4),347-363.

Miller,P.W.(2007),‘Over-educationandUnder-educationinAustralia’,The Australian Economic Review,40(3),292-299.



Mincer,J.andPolachek,S.(1974),‘FamilyInvestmentsinHumanCapital:EarningsofWomen’,Journal of Political Economy,82(2),76-108.

Neumark, D. (1988), ‘Employer’s Discriminatory Behavior and the Estimation ofWageDiscrimination’,The Journal of Human Resources,23(3),279-295.

Nielsen,H.S.(2000),‘WageDiscriminationinZambia:AnExtensionoftheOaxaca-BlinderDecompostion’,Applied Economics Letters,7(6),405-408.

Oaxaca, R. (1973), ‘Male-Female Wage Differentials in Urban Labor Markets’,International Economic Review,14(3),693-709.

Oaxaca,R.andRansom,M.R.(1994),‘OnDiscriminationandtheDecompositionofWageDifferentials’,The Journal of Econometrics,61(1),5-21.

Oaxaca,R.andRansom,M.R.(1999),‘IdentificationinDetailedWageDecompositions’,Review of Economics and Statistics,81(1),154-157.

Preston,A.C.(2001),The Structure and Determinants of Wage Relativities: Evidence from Australia,Ashgate,Aldershot.

Robst, J. (2007), ‘Education,CollegeMajor,andJobMatch:GenderDifferences inReasonsforMismatch’,Education Economics,15(2),159-175.

Rummery,S.(1992),‘TheContributionofIntermittentLabourForceParticipationtotheGenderWageDifferential’,The Economic Record,68(202),351-364.

Strategy and Research Policy in Education (2011), The Gender Agenda: Gender Differences in Higher Education.Availablefrom:<http://www.spre.com.au/download/SPREGenderAgenda2011.pdf>[15May2012]

Weichselbaumer, D. and Winter-Ebmer, R.W. (2005), ‘A Meta-Analysis of theInternationalGenderWageGap’,Journal of Economic Surveys,19(3),479-511.

Vahey,S.P.(2000),‘TheGreatCanadianTrainingRobbery:EvidenceontheReturnstoEducationalMismatch’, Economics of Education Review,19(2),219-227.

Voon, D. and Miller, P.W. (2005), ‘Under-education and Over-education in theAustralianLabourMarket’,The Economic Record,81(1),22-33.

Yun,M.(2005),‘ASimpleSolutiontotheIdentificationProbleminDetailedWageDecompositions’,Economic Inquiry,43(4),766-772.

Gender Discrimination in the Australian Graduate Labour...

Documents

Transcript of Gender Discrimination in the Australian Graduate Labour...