1 Occupational segregation, skills, and the gender wage gap By Andres Arcila, University of Waterloo Ana Ferrer, University of Waterloo Tammy Schirle, Wilfrid Laurier University Summary 2 1. Introduction 3 2. Data 5 Labour Force Survey 5 O*NET based skills measures 5 3. Characteristics of men and women 7 4. Gender wage gaps and the net segregation wage gap 12 Methodology 12 Model Results – All industries 14 Adjusted gaps and the net segregation wage gap within industries 16 Comparing the net segregation wage gap across industries 21 5. Occupations and skills 25 6. Discussion and policy relevance 27 Appendix A. Excluded occupations 29 Appendix B. Excluded Industries 30 Appendix C. Occupation categories, average skills, and percent male. 31 Appendix D. Construction of skills indices 33 References 41
42
Embed
Occupational segregation, skills, and the gender … Tammy Schirle_Gender...References 41 2 Occupational segregation, skills, and the gender wage gap By Andres Arcila, University of
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
SummaryThisreportoffersevidenceoftheroleofskillsandoccupationalgendersegregationinexplainingthegenderwagegapinCanada.WeuseCanada’sLabourForceSurvey(1997-2015) combined with detailed skills information from the Occupational InformationNetwork(O*Net)inourassessmentofCanadians’hourlywages.Wemeasuretheextenttowhich gender differences in job skills (specifically interpersonal, analytical, physicalstrength, visual, and fine motor skills,) can explain gender wage differentials. Weestimate this economy-wide, and within industrial sectors. Moreover, we compareadjusted wage gaps that account for gender differences in skills to more standardadjustedwagegapsthataccountforgenderdifferencesinoccupations.Thedifferenceisreferred toasanetsegregationwagegap, representing thatpartof thegenderwagegap that is associated with occupational segregation, net of the gender wage gapaccountedforbygenderdifferencesinskillsusedinoccupations.Wehighlightthefollowingresults:
• Accounting for gender differences in skills does little to explain thegenderwagegap
• The net segregation gap varies across industries and is largest inmale-dominatedindustries
• Gender wage gaps within occupation are partly explained by genderdifferencesinskillsandotherproductivecharacteristics,howeververticalsegregationandrelatedpaystructuresrequirefurtherinvestigation
Wesuggestconsiderationbegiventopolicydevelopmentinregardsto:• Achievingpayequityatanindustrylevel• Addressing occupational gender segregation within broad occupation
genderwagegapoccupationsAcknowledgementsTheauthorsgratefullyacknowledgesupportfromtheOntarioPayEquityOfficeGenderWageGapGrantProgram.Theanalysispresentedinthisreportwasmadepossiblewithaccess to data at the South Western Ontario Research Data Centre as part of theStatisticsCanadaResearchDataCentresProgram.Theopinionsexpressedhereindonotrepresent the opinions of Statistics Canada. The authors thank Scott Cameron, JoelWood,andMohsenJavdanifortheirsuggestions.
3
1.IntroductionIn2016,theaveragehourlywageofwomeninOntariowas87%ofmen’saveragehourlywage rate. In Ontario, and in the rest of Canada, there has been steady progress inclosing the genderwage gap as the female-malewage ratio increases over time (seeFigure1).
Figure1.Female-MaleWageRatios,1997-2016.Source:Authors’tabulationsbasedonCANSIMtable282-0074.Wagesrepresentthoseofallworkersaged25-54.A large Canadian literature has examined the gender wage gap and attempted toaccountforfactorsdrivingthegap.1Overalltheliteraturedemonstratesthatwhilethegapbetweenmen’sandwomen’swageshasdecreasedovertime,theextenttowhichtheremaininggapcanbeexplainedasrepresentingdifferencesinproductiveskillshasfallen. Schirle (2015) has shown that across the Canadian provinces, very little of thegender wage gap reflects gender differences in family status, education, or jobcharacteristics such as union status or job tenure. For themost part, the bulk of thewagegapcannotbeexplainedbyobservablepersonalandjobcharacteristics.However,large parts of the gender wage gap are associated with occupational and industrial
1This includesFortinetal.(2017),BakerandCornelson(2016),Fortin(2004),BakerandFortin(2004), Drolet and Mumford (2012), Schirle and Vickers (2014), Arcila et al. (2016), Vincent(2013).
4
gendersegregation.Simplyput,asmenandwomentendtomoveintodifferenttypesofjobs, and the jobs largely occupied by men tend to pay higher wages, men will onaveragehavehigherwages.Whetherornot this is a satisfactoryexplanationof thegenderwagegapdependsonwhatwethinkthewagedifferentialduetogendersegregationrepresents.Ithasbeencommon to summarize the wage differential associated with such occupationalsegregation as a compensating differential. In particular, there is an expectation thatsomeoccupationsrequiremoreproductiveskills,andsuchskillswillberewardedwithahigherwage.The purpose of this study is to examine further the nature of the gender wage gapassociatedwith occupational gender segregation. Does thewage premium offered inmale-dominated occupations actually reflect a gender difference in skills used on thejob?Whatelsemightunderliethispartofthewagegap?Inwhatindustrieswillaskillsgaphelpexplainthewagegap,andinwhatindustriesistheresomethingelseunderlyingthepremiumfoundinmale-dominatedoccupations?We use data from Canada’s Labour Force Survey (LFS, 1997-2015) to investigate thisquestion.Wematch the detailed occupation information available in the confidentialfiles of the LFS to detailed information about each occupation in the OccupationalInformation Network Database (O*NET). We begin with a standard exploration ofadjustedgenderwagegaps,examininghowdifferent factorsplaya role in thegenderwagegap.Weconstructameasureoftheextenttowhichoccupationalsegregationmayexplain the genderwage gap, net of the role that differences in the skills involved indetailedoccupationscanexplainthegap.Werefertothisasthenetsegregationwagegap.This reportproceedsas follows: In thenext sectionwedescribe thedataused in thisstudyandoursampleofinterest.Wethenprovideadescriptionofmenandwomenintermsoftheirpersonalandjobcharacteristics,aswellastheskillsusedintheirjobs.Insection4wedescribeourempiricalstrategyformeasuringadjustedwagegaps,andthenet segregation wage gap. We investigate these gaps within industries and explorepotential relationships with industry characteristics. In Section 5 we examine genderwage gaps within broad occupation categories and the extent to which genderdifferencesinskillcanexplainthesegaps.Finally,wediscussthepolicyrelevanceofourestimatesandtheneedforfurtherresearchforpolicydevelopment.
5
2.Data
LabourForceSurveyWe use the confidential files of the Labour Force Survey (LFS) from 1997-2015. TheLabourForceSurvey isadministeredas longitudinalsurvey, interviewingall individualsinsampledhouseholdsfor6months.Since1997,theLFShasreportedhourlywagesofallpaidemployeesaswellaspersonalandjobcharacteristics.2To avoid repeating observations of the same individuals in our sample, we restrictourselvestousinginformationfromtheindividual’sfirst(birth)interviewintheLFS.Wethen restrictour sample toemployed individualswhoarenot self-employedbetweentheagesof25and54.Thesebasicrestrictionsleaveuswith1,507,991observations.Themost important informationwe require from the LFS isoccupation. TheavailableLFS files report the National Occupational Classification codes for 2011 (NOC-2011).Eachoccupationisassigned4digits:thefirstdigitrepresentsabroadskilltypecategory,thesecondgenerallyrepresentsaskilllevelcategory,thethirdandfourthdigitsdefineminor occupation groups. These codes are not uniformly applied internationally andhavechangedovertime.Assuch,itwasnotpossibletomatchalloccupationsfoundintheLFSwithskills information fromO*NET (describedbelow).Asa result,our sampleexcludes27,742observationsbecausetheirLFSoccupationcouldnotbematchedwithourskillsinformation.(SeeAppendixAformoredetail.)Given our interest in factors explaining genderwage gapswithin industry groups,wealsowanttoensurelargeenoughsamplesofmenandwomenwithineachindustry.Wechose to exclude any industries whereby fewer than 200 men or 200 women weresampled.Onlyahandfulofindustrieswereexcluded,dropping7080observationsfromour sample. Notably, more than half of these dropped observations representedindividualsemployedinprivatehouseholds.(SeeAppendixBformoredetail.)Given these restrictions, we are left with 1,473,168 observations in our sample. Thissample represents employedmen andwomen in Canada, aged 25-54, over the years1997-2015.
O*NETbasedskillsmeasures
The O*NET is a U.S.-based database containing information on detailed occupations,including theabilities andknowledge requiredofworkerswhohold theoccupations.32Sincewearepoolingdatasince1997,weadjustedwageratesforinflationusingtheCanadianall-itemsconsumerpriceindex(CANSIMTable326-0020)sothatwagesarein2015dollars.3Moreinformationcanbefoundathttps://www.onetonline.org
6
Increasingly,laboureconomicstudiesusetheinformationcontainedinO*Nettobettercharacterizetheimplicitheterogeneity inoccupationalchoice(seeforexampleImaietal. 2014 andAdsera and Ferrer 2014). Typically, the job skills information is collectedinto a small set of indices, using Factor Analysis, which summarizes the skillrequirementsforeachoccupationintoamoreeasilyinterpretableindex.Weconstructfive indices covering both cognitive and non-cognitive job skills that line up withpreviouswork in the area to facilitate comparison. Specifically, we use two cognitiveindices representing interpersonal (social) skills and analytical (quantitative) skills andthree indices for non-cognitive or manual skills, including fine motor skills, physicalstrength, and visual skills. We use Confirmatory Factor Analysis (CFA) (following thework of Imai et al 2014) to construct the index values; the technical details of itsimplementationcanbefoundintheAppendix.Toconnectthe informationregardingskillsrequiredofoccupationsfound inO*Nettooccupations in the LFS,we relyprimarilyona seriesof crosswalksdesigned tomatchoccupationcodesfromO*NETtoStandardOccupationCodes(SOC)usedintheUnitedStates and then from SOC to the NOC codes now used in Canada. Using thesecrosswalks, we are able to match most occupations. We then manually reviewedunmatched LFS occupations to find appropriatematches in O*NET. As noted in theAppendixA,onlyafewoccupationsareleftunmatched.
To facilitate interpretationof theskillvariablesthefactoranalysisusesasweights thedistribution of the skills in the Canadianworkforce (all ages) over our sample period(1997-2015).Assuch,askillscoreofzerodescribesthelevelofskillusedbytheaverageworkerintheCanadianworkforce.Also,aunitoftheskillscore(withmeanzero)canbeinterpretedasonestandarddeviation in theskilldistributionof theCanadianworkingpopulation.
7
3.CharacteristicsofmenandwomenInthissectionweprovideadescriptionofourmainsamplesofmenandwomenusedinthisstudy,intermsoftheiraveragecharacteristics.InTable1wesummarizesomeofthedemographiccharacteristicsofmenandwomen.The age distribution of men and women is quite similar, though men appear morepresent in the workforce at younger ages than women. In terms of the number ofchildreninone’shousehold,menaremorelikelytohaveyoungerchildrenthanwomenandwomenaremorelikelytohaveolderchildrenincludedintheirhousehold.Amongemployedmen andwomen, thiswill in part reflect themanagement of childcare (aswomen are more likely not employed when children are young) and child custodyarrangements whereby older children are more likely primarily with mothers thanfathers.Asthesefactorsmayreflectdifferencesinlifecycleworkarrangementsofmenandwomentheyareimportanttoaccountforinouranalysisofwages.
In Figure 2 we present the distribution of men and women across educationalattainmentcategories. It isclearthatwomenaremore likelytoattenduniversityandcollege than men, and are less likely to leave school before finishing high school.However,menaremorelikelytopursueatradescertificatethanwomen.Asthetrainingoffered by formal education affects a worker’s productivity, education must beaccountedforinouranalysisofgenderwagegaps.
8
Figure2.DistributionofmenandwomenacrosseducationcategoriesNote:Eachsegmentrepresentsthepercentofmenorwomenwitheachlevelofeducation.InTable3weconsidertheaveragejobcharacteristicsofmenandwomen.Onaverage,menhave slightlymore seniority in their jobs (just over fourmonths).Menaremorelikelytoworkinfulltimepositions(95.7%)thanwomenare(81.6%).Despitethis,nearlyasmanywomenareworkinginpermanentpositionsasmen.Womenareslightlymorelikelytobeunionizedthanmenare,whichinpartreflectstheirmuchhigherlikelihoodofbeingemployedinthepublicsector.Finally,womenaremorelikelythanmentoworkwithverysmallemployers(lessthan20employees)thanmenare.
InFigure3weseemenandwomentypicallyuseverydifferentskillsintheirjobs.Jobsheld by women tend to use more interpersonal skills than average, while men useslightly less interpersonalskillsthanaverage.BothmenandwomeninoursampleusemoreanalyticalskillsthantheaverageworkerinCanada.Givenoursamplingstructure,thisimpliesmenandwomen25-54areusingmoreanalyticalskillsthanworkersoutsidethisagerange.It isalsoclearthatmenusesubstantiallymorephysicalstrength,visualskills,andfinemotorskillsintheirjobsthanwomenuse.
Figure3.AveragejobskillsofmenandwomenNote:SampleincludesemployedCanadiansaged25-54,1997-2015FromFigures4and5wealsoseethatmenandwomentendtosegregateintodifferentoccupationalcategories(ata2-digitNOClevel).Menheavilydominatetheworkforceintrades-related occupations, as equipment operators, labourers, and professionaloccupationsinscience.Womentendtodominateinnursing,officeadministrationandsupport,andcareproviders. It iscleartheskillsused ineachoftheseoccupationsarequite different. (See Appendix C for a list of occupation groups with more detailedaverageskillsinformationandtheportionofemployeesthataremale.)Overall,weseeseveraldimensionsalongwhichmenandwomenappearquitedifferentin terms of their position in the labourmarket, the characteristics of their jobs, theirtraining, and the skills they use in their jobs. In the next sectionswe investigate theextenttowhichthesedifferencesincharacteristicsandskillscanexplaindifferencesinhourlywages.
MethodologyInouranalysisweusefourregressionmodels,eachaccountingfordifferentfactors,asfollows:(1) ln(wi)=α+δFi+εi(2) ln(wi)=α+γFi+XiΒ+εi(3) ln(wi)=α+λFi+XiΒ+ΣjϕjOccupji+εi(4) ln(wi)=α+φFi+XiΒ+ΣkηkSkillki+εiWhere ln(wi) is the natural logarithm of the real hourly wage and Fi is an indicatorvariableequaltoonewhenindividualiisfemale,and0whenmale.Inthefirstequation,theregressioncoefficientδwillrepresenttheunadjusted,orraw,genderwagegap:thepercentdifference inhourlywagesbetweenmenandwomen,without controlling foranythingelse.4Inthegeneralcasewheremenonaverageearnmorethanwomen,thiscoefficientwillbenegative.In thesecondequation, the termXi represents themainvariableswecontrol for,andaremodeledasaffectingwagesaccordingtotheparameterΒ.Inourstudy,thecontrolvariables include indicators for 5-year age group, marital status, the number of ownchildrenbyageofthechild,educationlevels,tenureonthejob,whetherone’sjobisfulltime,permanent,unionized,andinthepublicsector,andthesizeoftheestablishmentone works in, as well as indicators for province of residence and the year one isobserved. Whenusingourfullsampleof individualswealso includeasetof indicatorvariablesforindustry.Thecoefficientontheindicatorforfemale,γ,nowrepresentsanadjusted gender wage gap: after accounting for gender differences in these controlvariables,whatisthepercentdifferenceinhourlywagesbetweenmenandwomen.Bycomparing thevaluesofγ andδ,wecangeta senseofhowmuchof the rawgenderwagegapisduetogenderdifferencesinthesecharacteristics.In the third equation, the terms Occupj represent indicator variables for each of joccupation categories. In ourmain specificationwe include40occupation categories.WhenestimatingtheequationwithourfullCanadiansample,weareabletoincludeafiner set of categories, breaking occupations into 140 groups. The coefficient λrepresents theadjustedwagegap that furtheraccounts for the tendencyofmenand
women to enter different categories of occupations, as well as the set of covariatesdescribed in the secondequation. Thedifference (γ−λ) describes the extent towhichthegenderwagegapcanbeattributedtothisoccupationalgendersegregation.5Finally, the fourthequation replaces theoccupation indicatorvariableswithourmoredirectmeasuresofskillsusedintheindividual’soccupation(recallingfinerclassificationsof the 4-digit NOC are used when matching skills of individual jobs). From thisregression, the coefficientφ represents the adjustedwage gap that accounts for anygender differences in skills, as well as the set of covariates used in previousspecifications.Thereare twocomparisons fromthis finalequation thatare interesting.First,aswithourthirdspecification,thedifferencebetweenφandγdescribestheextenttowhichthegenderwage gap canbe attributed to genderdifferences in skills usedon the job.6Ifgenderdifferencesinskillshelpexplainthegenderwagegap,weexpectφtobecloserto zero than γ. The difference (γ−φ) indicates howmuch of the gender wage gap isattributedtogenderdifferencesinskills.Second,weareinterestedinthedifferencebetweenλandφ.Whereresultssuggestλisclosertozerothanφ is,wehaveasituation inwhichoccupationalgendersegregationappearstoexplainmoreofthegenderwagegapthangenderdifferencesinskillsusedinanoccupation.Wecreatethemeasure(λ-φ)tocapturetheextenttowhichthisistrue.Thisdifference(λ-φ)canbeinterpretedasthepartofthegenderwagegapthatcanbeexplainedbyoccupational gender segregationnetof anygenderwagegapassociatedwithgenderdifferencesintheskillsusedinoccupations.Apositivenumberwillindicatethat occupational gender segregation explains more of the gender wage gap thangenderdifferences in job skills.We refer to thisdifference (λ-φ) asanet segregationwagegap.Generallythiscouldrepresentgenderdifferencesinanyundesirablejobattributesthatemployers must compensate employees for (such as risk of injury), as well as anysystemic bias in workplaces that tend to favour wage schedules in male-dominatedoccupations.We return to a discussion of what this difference may indicate in latersections.
ModelResults–AllindustriesInTable4wepresenttheregressionresultsbasedonthefullsampleofCanadiansaged25-54, across all industries combined. Each column represents a different regressionmodel,correspondingtotheequations(1-4)describedintheprevioussection.Thefirstrow of the table provides the coefficient on the female indicator variable in theseregressions,representingtheunadjustedoradjustedwagegap.In the first columnofTable4,ourestimates indicateanunadjustedwagegap (δ)of -.175, indicating that the average wages of women are roughly 17.5% less than theaveragewagesofmen. This forms thebaseline againstwhichour estimated adjustedwagegapsarecompared.In the second column of Table 4, we include our baseline set of control variables.Perhapssurprisingly,theadjustedwagegapappearsslightlylargerthantheunadjustedwagegap.This inpartrelatestoacommonfinding inthegenderwagegap literature,wherebywomen inthissamplearemoreeducated,onaverage,thanmen,andwouldthereforeexpecttoreceivehigherwagesasaresult.Assuch,thewagegapthatisnotaccounted forby the includedvariablesappears larger. In thenext columnofTable4(2b), we extend our baseline set of control variables to include a set of indicatorvariables for industry (at the 3-digit NAICS level). This accounts for industrial gendersegregation. The adjusted wage gap estimate is much smaller as a result, at 14.7percent.InthenexttwocolumnsofTable4(3aand3b)we include indicatorsforoccupationalcategories,accountingforoccupationalgendersegregation.Theestimatessuggestthatwhenusingabroadersetof(40)occupationcategories,theadjustedwagegapis12.3percent.A finersetof (140)occupationcategoriesareused inspecification incolumn(3b) as a robustness check, resulting in a very similar estimate for the adjustedwagegap.Finally,thelastcolumnofTable4(column4)replacestheoccupationindicatorvariableswith our measures of job skills in the regression. First, we compare the resultingadjustedgenderwagegap(at14.4%)tothatincolumn(2b)(at14.7%)whichincludesallthesamecontrolvariablesexcept for skills.Accounting forgenderdifferences in skillsdoes little, if anything, to reduce the gender wage gap. Given other regressioncoefficients in Table4 and theaverage skills ofmenandwomen (Figure3),we seeasituation where men tend to use more of the skills that receive the lowest wagepremiums.Forexample,inTable4weseethatinterpersonalskillsareassociatedwithmuchhigherwages(aonestandarddeviationincreaseinskill isassociatedwithwagesthatare7.1percenthigher)andinfigure3weseethatwomenusemoreinterpersonalskills intheir jobs.Moreover,menusemorefinemotorskillsandtheseareassociatedwith lower wages on average. This balances against men’s slightly higher use ofanalyticalskillsintheirjobsandthehigherwageassociatedwiththeuseofthoseskills.
Wethencomparethedifferencebetweentheadjustedgenderwagegapaccountingforskills (φ in column 4) to the adjusted wage gap accounting for occupational gendersegregation(λincolumn3a).Thedifference,2.1percentagepoints,representsthenetsegregation wage gap. That is, we suggest that 2.1 percentage points of the genderwagegap is associatedwithoccupational gender segregationafter accounting for thepartof thegapthat isassociatedwithgenderdifferences intheskillsused indetailedoccupations.
AdjustedgapsandthenetsegregationwagegapwithinindustriesInthissectionweinvestigategenderwagegapswithinindustries,withtheexpectationthatthegapandourmeasureofthenetsegregationwagegapwillvarywidelyacrossindustries. Estimating the adjusted wage gaps and net segregation wage gap withinindustryalsoallowsustorecognizethateachtypeofskillmaybevalueddifferentlyindifferentindustries.Asdiscussedinlatersections,policymakersmaywanttorecognizeandidentifyhowthesewagegapsvaryacrossindustrialsectors.We repeat the regressions specified inequations (1)-(4) toobtainestimatesofδ, γ, λ,andφ. Thesegenderwagegapsareestimatedwithinsamplesofworkers,by industrygroup (3-digit NAICS). Results are presented in Table 5, with each row representingregressions for the specified industryandeach column representingourgenderwagegapestimatesofδ,γ,λ,andφ,respectively.Thenetsegregationwagegapcanbefoundbyfindingthedifferencebetweenthethirdandfourthcolumnresults(λ−φ).First,considertheresultsincolumn(1)fortheunadjustedwagegap(δ),whereweseeahigh degree of variation across industries. For example, in the industry of specialtytradecontractors,womenonaverageearnroughly28.9percentlessthanmen.Inrealestate,theunadjustedgenderwagegap is10.4percentand innursingandresidentialcarefacilities,thegapisonly5.5percent.Withinbroaderindustrycategories,thereisawiderangeofwagegapsobservedandfewgeneralizationscanbemade.Lookingacrossindustries there are also few generalizations, although industries dominated by thepublicsectorappeartohaveaverageorlowerthanaveragegenderwagegaps.InFigure6werelatetheextenttowhichanindustryismale-dominated(i.e.thepercentofworkers inthat industrythataremale)totheunadjustedwagegapinthe industry.Perhaps surprisingly, there is no clear correlation between the two industry levelstatistics.Moreover,weseethatevenwithinbroadercategoriesof industriesthataremale-orfemale-dominatedthereisawidevariationintheunadjustedgenderwagegap.
Figure6.Unadjustedgenderwagegapsandthepercentoftheworkforcethatismale,byindustry.Note:Eachpointinthescatterplotrepresentsanindustryatthe3-digitNAICS,labeledbythenumberforitsbroader2-digitNAICS,asprovidedinTable5.WenowturntoourestimatesoftheadjustedgenderwagegapsinTable5(columns2-4),andtheresultingnetsegregationwagegaps.Consider,forexample,theoilandgasextractionindustry.Theunadjustedgenderwagegapinthisindustryisrelativelyhigh,at21.8percent.Afteraccountingforgenderdifferencesinourmaincontrolvariablesandoccupationcategories,ourresultssuggestanadjustedgenderwagegapthatisslightlysmaller,at18.3percent.Notably,thedifferencebetweentheresultsincolumn(2)andcolumn(3),at6.1percentagepoints,suggeststhatoccupationalsegregationcanexplainasubstantialportionofthegenderwagegap.Movingfromcolumns(3)to(4),however,weseethatgenderdifferencesinskillswillaccountforlessofthegenderwagegapthangenderdifferencesinoccupation.Ourestimateofthenetsegregationwagegapfortheoilandgasextractionindustryis1.7percentagepoints.Generallyspeaking, theresultsacross industriesaresimilar in thatgendersegregationintooccupationstendstoexplainalargerportionofthegenderwagegapthangenderdifferences in skills used in detailed occupations. In some industries the resulting netsegregationwagegapisquitelarge:forexampleinspecialtytradecontractors,thenetsegregation wage gap is 8 percentage points; in professional, technical and scientific
21
services, the net segregation wage gap is 3.4 points; in fabricated metal productmanufacturingthenetsegregationwagegapis4.3points.Wenoteanunusually largenetsegregationwagegapinambulatoryhealthcareservices,at13.1percentagepoints,thatisworthyofspecialconsiderationandamoredetailedexaminationgiventherangeofservicesincludedinthisindustry.Incontrast,therearesomeindustries inwhichthenetsegregationwagegap isnearlyzero, or even positive. Accommodation services, which has a fairly high unadjustedgenderwagegapat22.1percent,hasasmallnegativenetsegregationwagegapat-0.9percent.This indicates that thepartof thegenderwagegapaccounted forbygenderdifferencesinoccupationsisalsoreflectinggenderdifferencesinskill.Otherexampleswherethisappearstobetrueincludeeducationalservices,rentalandleasingservices,several industries involved in retail trade, and support activities for agriculture andforestry.
ComparingthenetsegregationwagegapacrossindustriesWehavedescribedthenetsegregationwagegapasrepresentingthatpartofthewagegapaccounted forbyoccupational segregationnetof thewagegapaccounted forbygenderdifferencesinskills.Whatmightthispartofthegenderwagegaprepresent?Therearetwokeyfactorswewanttoconsiderhere.Thefirstfactorisgenerallyknownascompensatingdifferentials.Whilewehaveaccountedforskills,therearemanyotherjobattributes thatemployeesarecompensatedfor - suchasrisk, instability,orotherundesirablejobattributes.Thesecondfactorrepresentsasystemicbiasfavouringmalewagestructures,maybemaintaininganhistoricalrelativepositionofmeninthelabourmarket.Itmayalsorepresentatendencytorewardfulltimeworkschedulesatahigherhourlyrateofpaythanparttimework.7Forthemostpart, thefactorswearemost interested incannotbedirectlyaccountedfor given the data we have available. In this section we provide some correlationsbetween industry characteristics and the net segregationwage gap, hoping to give abettersenseofwhatthismayormaynotrepresent.InFigure7wepresentascatterplotof industries’netsegregationwagegap(basedonindustrywagegapresultsinTable5)andthepercentofworkersthataremaleineachindustry.While therewas no clear relationship between the total unadjusted gender7Typically,weseefulltimeworkersearningahigherwageratethanthoseworkingparttime(asourestimatesinTable4suggest).However,weneedtofurtherconsiderwagescheduleswithinoccupations.Forexample,GoldinandKatz(2016)studiedtheevolutionofpart-timewagepenaltiesamongpharmacistsandpointtotheimportanceoftechnologythatfacilitatesthesubstitutabilityofworkersforclosingthegenderwagegap.Furtherexaminationofworkschedulesispartoftheauthors’futureresearchagenda.
22
wagegapandthepercentmale ineach industry (Figure6), inFigure7weseeaclearpositiverelationshipbetweenthenetsegregationwagegapandthepercentmaleintheindustry.
Figure7.Thenetsegregationwagegapandpercentmalewithinindustries.Note:Eachpointrepresentsasingleindustry(3-digitNAICS),withoneoutlier(ambulatoryhealthcare)omitted.In Figures 8 and 9we consider the potential relationship between industries’ job riskand the net segregation wage gap, as risk would represent an important reason tocompensate employees apart from the skills used in their jobs.We use the availableincidenceof fatalandnon-fataloccupational injuries fromtheU.S.withineach3-digitNAICS (which isunfortunatelynot readilyavailable fromCanadiansources)asaproxyforindustryrisk.InFigure8weshowthereisasmallpositivecorrelationbetweentheriskoffatalinjuryand our measure of the net segregation gap, however we must stress that thecorrelationisnearzero.Similarly,whenweconsidertheriskofnon-fatalinjuries(Figure9),thereisvirtuallynorelationshipwithourmeasureofthenetsegregationwagegap.
Figure9.Thenetsegregationwagegapandriskofnon-fatalinjuryNote:IncidenceratesarebasedonBLSdataavailableathttps://www.bls.gov/iif/oshwc/osh/os/ostb4740.pdf,andlargelyrepresentstheprivatesector.While a more rigorous assessment of jobs in Canada is required, this suggests thatgenderdifferencesinjobriskswillnotexplainthatpartofthewagegapassociatedwithoccupationalsegregationnetofgenderdifferencesinskills.Thisisperhapsunsurprisinggivenevidencethattherisksfacingmenandwomenarenotaslargeasoncebelieved.For example,Hersch (1998) showed thatwomen face a job risk that is 71 percent ofmen’sandsimilarlyreceiveawagepremiumforthatrisk.
24
Wethenconsiderhowcompetitiveanindustryisinrelationtothenetsegregationwagegap.Generally, theoriesofdiscriminationwouldsuggestthatthemorecompetitiveanindustry, the less employers are able to discriminate against women and pay lowerwagesthanforequallyskilledmen.InFigure10werelatevaluesofthenetsegregationwage gap and the Herfindahl-Hirschman Index (HHI) for each industry (at the 3-digitNAICS level, Canada, based on total revenues and number of enterprises). Note thathighervaluesoftheHHIsuggestanindustryislesscompetitive,havingfewenterprisesholding a relatively large market share. Perhaps surprisingly, there is no correlationbetweenthismeasureofmarketconcentrationournetsegregationindex.Itisunclear,however,thatmeasuringindustryconcentrationatthishighlevelofaggregationisthemostrelevantmeasureforourpurposes.
Figure10.ThenetsegregationwagegapandindustryconcentrationNote:HHIvaluesrepresentcustomtabulationsbyStatisticsCanada.Overall,then,weareleftwiththeknowledgethatmale-dominatedindustriestendtobethose with larger net segregation wage gaps. With the information presented here,however,wearenotabletospeaktowhichfactorsunderliethenetsegregationwagegap.
25
5.OccupationsandskillsAsthenetsegregationwagegapisnoteasilyexplained,wefurther investigategenderwage gapswithin broad occupation categories. Specifically, within each of 10 broadoccupationcategories,weestimatethefollowingregressions:(5) ln(wi)=α+δFi+εi(6) ln(wi)=α+ω1Fi+ΣkηkSkillki+εi(7) ln(wi)=α+ω2Fi+ΣkηkSkillki+ΣjϕjIndustryji+εi(8) ln(wi)=α+ω3Fi+ΣkηkSkillki+ΣjϕjIndustryji+XiΒ+εiEquation (5) is directly comparable to equation (1), in that the coefficient δ willrepresent an unadjusted gender wage gap in the occupation category. The adjustedwagegapestimatefromequation(6),ω1, representsthegendergapwithinthebroadoccupationcategoryafteraccountingfortheskillsusedwithineachdetailedoccupationcategory.Theadjustedgapestimatesbasedonthemodelsinequations(7)and(8)areprovided for completeness, further adjusting the gap to account for the industry inwhichindividualsworkandotherindividualandjobcharacteristicsXi,asintheprevioussection.WealsoconductaOaxaca-Blinder(OB)decompositiontoestimatetheextenttowhichgender differences in skills are a factor driving the gender wage gap within eachoccupation group. Specifically,we estimate regressionswith the full set of covariatessimilartothatspecifiedinequation(8),butwithinsamplesofmenandwomen(ratherthan pooling the two groups together and including the indicator for female). Theresultinggenderspecificcoefficientsandaveragesofcovariates(subscriptedwithMandF)areusedtoestimatethepercentofthegenderwagegap(ln(wM)-ln(wF))attributedtogenderdifferencesinskillasfollows:(9) OB%skills=100*(Σk(SkillkM-SkillkF) ηkM)/(ln(wM)-ln(wF))The results are presented in Table 6. As a starting point, consider managementoccupations.Within thisbroad classofoccupations,we see thatwomenearnawagethatisapproximately19.4percentlessthanmenonaverage.Whenweaccountforskillsused indetailedoccupations, theadjustedwagegap ismuch smallerat16.3percent.Thissuggestsroughly3percentagepointsofthe19pointgap(or16%)canbeattributedto gender differences in skill. In the third and fourth columns, we further adjust thewage gap for Industry and other variables we can control for, noting there may beimportant interactionsbetweentheadditionalvariablesandourskillsmeasures.Afteraccounting for observable gender differences in characteristics, we find an adjusted
26
wage gap of 13.7 percent remains. In the last column, we provide the estimatedexplanatory power of gender differences in skills. The result suggests that genderdifferencesinskillsusedacrossallmanagementoccupationscanexplain4.8percentofthegenderwagegapwithinthisbroadlydefinedoccupationgroup.Table6.Wagegapswithinoccupationcategories Adjustedwagegapsaccountingfor: OB%Occupationcategory None Skills Industry Controls SkillsManagementoccupations -0.194 -0.163 -0.154 -0.137 4.8Business,finance&admin. -0.152 -0.113 -0.121 -0.102 16.8Natural&appliedscience -0.114 -0.120 -0.126 -0.123 -1.2Health -0.039 -0.087 -0.062 -0.049 -46.5Education,law,social,community,gov. -0.159 -0.078 -0.047 -0.053 17.9Art,culture,recreationandsport -0.071 -0.081 -0.064 -0.068 -0.7Sales&service -0.229 -0.212 -0.177 -0.141 3.5Trades,transport,equipmentoper. -0.289 -0.205 -0.184 -0.170 16.4Naturalresources,agriculture,related -0.417 -0.234 -0.144 -0.168 11.4Occup.Inmanufacturing&utilities -0.374 -0.305 -0.227 -0.187 10.1NOTE:Resultsinthefirstcolumnspresentedarecoefficientsonthefemaleindicatorforregressionsrepresentedbyequations(5)–(8),estimatedwithineachindustry.Resultsinthelastcolumnrepresentestimatesofequation(9).LookingacrosstheoccupationcategoriesinTable6,weseeafairamountofvariationinresults.Forexample,insomeoccupations,skilldifferentialswillaccountforfairlylargeparts of the gender wage gap. This is clearly the case for business, finance andadministration, or trades, transport and equipment operators. In others, such asoccupationsinnaturalandappliedsciences,skillswillexplainverylittleofthegap.Someoccupation groups, such as health, appear more complex as accounting for skillssubstantiallyincreasesthesizeofthegapthatneedstobeexplained.8Whatdotheadjustedwagegapstellus?TheadjustedwagegapsinthefourthcolumnofTable6willinpartreflectanyothergenderdifferencesinjobattributesthatcannotbeaccountedforinourdata.Inmanyoccupationcategories,however,asubstantialgapremains.Givenwhatweknowfromthecurrentliterature,itisdifficulttoimaginealistofjobattributeswithsufficientgenderdifferencestoaccountforthisgap.
8Aswithourdiscussionoftheambulatoryhealthservices industry intheprevioussection,webelieve health occupations require a more detailed examination than provided here. Here,AutorandHandel’s(2013)workregardingthefactthattaskswithindetailedoccupationgroupsmayvarycouldbeimportant.
27
The adjusted wage gap may reflect some degree of vertical segregation. Verticalsegregationisthesegregationofmenandwomenalonghierarchicallevelsofworkandinthiscaseoccursinsimilarlinesofwork(broadoccupationcategories).9Typically,wewould think of this hierarchy as representing different levels of skills, education, orexperience. However, in our analysis presented in Table 6 we have accounted for avarietyofmeasuresthatreflectsuchdifferences.Evidence in the literature suggests important gender differences in the propensity toreceive joboffers, training,orpromotionopportunities,unrelated toone’sproductivecharacteristics. In particular, there is a growing evidence of evaluators’ bias in theevaluation of women that reduces women’s likelihood of being hired and promotedrelativetoequallyqualifiedmalecandidates.10Thisbiasthatimpedeswomen’sabilitytoreachthetop levelsoftheoccupationalhierarchy,often labeledasthe“glassceiling”,has been cited as a determinant of the gender wage gap.11. Given the unobservablenatureofproductivityinmanyseniormanagementpositions,wewouldliketoconsiderwhethersimilarformsofevaluatorbiasalsoaffectwagestructuresinwaysthattendtofavourmen.
6.DiscussionandpolicyrelevanceClosingthegenderwagegaphas longbeenagoal forpolicymakersatmany levelsofgovernment. The results presented in this study suggest two distinct approaches topolicytoconsider:(i)thedevelopmentofpayequitypoliciesatanindustrylevel,and(ii)developmentofpoliciestoaddressverticaloccupationalsegregationwithinoccupationgroups.Thefirstsuggestedapproachtopolicyderivesfromourestimatesofthenetsegregationwagegapwithin industries. This represented thepart of the genderwagegap that isassociatedwithoccupationalsegregationnetofthegenderwagegapaccountedforbygenderdifferencesinskillsusedinoccupations.Thefactorsdrivingthenetsegregationwagegaparelikelytovarybyindustryandrequiremoredetailedstudywithinsmallergeographic regions and industry groups. Theevidenceprovided in this study suggestsindustries that could be targeted for further investigation.Here,we are interested in
9This is differentiated from vertical segregation that represents hierarchical levels of workacrossdifferentlinesofwork,andhorizontalsegregation,whichwouldbesegregationintojobswith similar skill requirements, butdifferent fields suchas teachers andengineers. See FortinandHuberman(2002)formoreonthis.10Forexamples, seeWennerasandWold (1997),BaguesandEsteve-Volart (2010),RouseandGoldin(2000)andSarsons(Forthcoming)11Seeforexample“TheGlassceiling”,TheEconomist,May5th,2009
28
industrieswith largenetsegregationgaps,particularly industriesthatareheavilymaledominated.Withinindustries,jobattributesandpaystructureswithinmaleandfemaledominatedoccupationscouldbecomparedatan industry level inamannersimilar tomethodsoutlinedincurrentpayequitypoliciesforlargeemployers.Thesecondapproachtopolicy–addressingverticaloccupationalsegregation–derivesfrom our analysis of wage gaps within broad occupation categories. Our evidencedemonstrates that within occupation categories, some of the gender wage gap willrelatetodifferencesintheproductivecharacteristicsofdetailedoccupations–suchaseducation, experience and skills.However, largeparts of the genderwage gapwithinbroadoccupationcategoriesarenotexplainedbysuchfactors.Whilesomeunobservedjob attributesmay (ormany not) rationalize the remaining genderwage gap,we aremoregenerallyconcernedthatbarriersremainpreventingmanywomenfromenteringthehigherpaidjobswithinbroadoccupationgroups.Finally, we have highlighted a need to further investigate wage schedules withinoccupations in light of evidence that the elimination of part-timewage penalties canreducethegenderwagegap.EvidencefromGoldin(2016)basedontheoccupationofpharmacist suggests that reducing employers’ costs associated with offering greater‘temporal flexibility’ (part time work schedules) is necessary to reduce the part-timewagepenalty.The importanceof flexibleworkscheduleswithinCanadianoccupationsremainsanimportantpartoffutureresearch.
AppendixB.ExcludedIndustriesWedonotincludethefollowingindustriesinourmainsample,nordoweassesstheindustryseparately,becauseofsmallsamplesofeithermenorwomen(lessthan200menor200women).Notably,justoverhalfofobservationsdroppedherearerepresentedintheindustryofprivatehouseholds.NAICS Industry
AppendixD.ConstructionofskillsindicesWe use Confirmatory Factor Analysis (CFA) to create our skills index using theinformationaboutabilitiesrequiredforjobsintheO*NET.Weweighteachindexusingthe Labor Force Survey (LFS) so that it can be interpreted as standard deviations ofdistributionsofskillinthelaborforceinCanada.ThisisexerciseisequivalenttoImaietal(2014).TheO*NETdata containsverydetailed informationabout skills required foreach job;manyof these are correlated and represents oneunderlying skill. Therefore, CFA is avery suitable technique to reduce the dimension of the information and recovereverything in one single index. We then generate five skills measure: Interpersonal,Analytical,Physical,Visual,andMotor.TomatchthecodificationoftheoccupationsintheLFS2015,whichusestheCanadianStandardOccupationClassification,andthecodificationintheO*NETdatabase,weusethe crosswalks posted in theO*NETwebpage. In some cases, twooccupations in theO*NET-SOC codificationwerematched into one in the Canadian. For these cases,weusetwoapproaches:Thefirstonewastomatchrandomlyoneoftherepeatedvaluesinthe O*NET-SOC into the Canadian one. The second was averaging the value of theinformation of those repeated values required for each job andmatched it with theCanadianpair.Theresultspresentedinthepaperdonotchangeconditionalonhowwematchedtheinformation.
34
Interpersonal
WeconstructanindexthatrecoverInterpersonalskillsusingtenvariables,thefirstsixare listed intheAbilitiessection intheO*NETdatabaseandtheremainingfour inthejob incumbent rating. Table A1 show the variables and Table A2 the principalcomponents(PC)loadings.
Translating or explaining what information meansandhowitcanbeused.
4A4a2CommunicatingwithSupervisors,Peers,or
Subordinates
Providing information to supervisors, coworkers,andsubordinatesbytelephone, inwrittenform,e-mail,orinperson.
4A4a3 CommunicatingwithPersonsOutside
Communicating with people outside theorganization, presenting the organization tocustomers, the public, government, and otherexternal sources. This information can beexchangedinperson,inwriting,orbytelephoneore-mail.
WeconstructanindexthatrecoverAnalyticalskillsusingeigthvariables,thefirstsixarelisted in theAbilitiessection in theO*NETdatabaseandtheremaining two in the jobincumbentrating.TableA3showthevariablesandA4thePCloadings:
1A1b5 InductiveReasoningTheability to combinepiecesof information to formgeneral rules or conclusions (includes finding arelationshipamongseeminglyunrelatedevents).
1A1b6 InformationOrdering
The ability to arrange things or actions in a certainorderorpatternaccordingtoaspecific ruleorsetofrules (e.g., patterns of numbers, letters, words,pictures,mathematicaloperations).
1A1b7 CategoryFlexibility The ability to generate or use different sets of rulesforcombiningorgroupingthingsindifferentways.
Weconstructan indexthatrecoverPhysicalskillsusingsixvariables, the first fourarelisted in theAbilitiessection in theO*NETdatabaseandtheremaining two in the jobincumbentrating.TableA5showthevariablesandA6thePCloadings:
1A3a1 StaticStrength The ability to exert maximum muscle force to lift,push,pull,orcarryobjects.
1A3a3 DynamicStrengthThe ability to exert muscle force repeatedly orcontinuously over time. This involves muscularenduranceandresistancetomusclefatigue.
1A3a4 TrunkStrength
The ability to use your abdominal and lower backmuscles to support part of the body repeatedly orcontinuously over time without "giving out" orfatiguing.
1A3b1 StaminaThe ability to exert yourself physically over longperiods of time without getting winded or out ofbreath.
4A3a1Performing
GeneralPhysicalActivities
Performing physical activities that requireconsiderable use of your arms and legs and movingyourwhole body, such as climbing, lifting, balancing,walking,stooping,andhandlingofmaterials.
4A3a2 HandlingMovingObjects
Using hands and arms in handling, installing,positioning, and moving materials, and manipulatingthings.
We construct an index that recover Motor skills using eight variables listed in theAbilities section in theO*NETdatabase. TableA9 show the variables andA10 the PCloadings:
1A2a1 Arm-HandSteadinessTheabilitytokeepyourhandandarmsteadywhilemoving your arm or while holding your arm andhandinoneposition.
1A2a2 ManualDexterityThe ability to quickly move your hand, your handtogetherwithyourarm,oryourtwohandstograsp,manipulate,orassembleobjects.
1A2b1 ControlPrecisionThe ability to quickly and repeatedly adjust thecontrols of a machine or a vehicle to exactpositions.
1A2b2 MultilimbCoordination
The ability to coordinate two or more limbs (forexample, two arms, two legs, or one leg and onearm)while sitting, standing,or lyingdown. It doesnot involve performing the activities while thewholebodyisinmotion.
1A2b3 ResponseOrientation
Theabilitytochoosequicklybetweentwoormoremovements in response to two or more differentsignals (lights, sounds, pictures). It includes thespeed with which the correct response is startedwiththehand,foot,orotherbodypart.
1A2b4 RateControl
Theabilitytochoosequicklybetweentwoormoremovements in response to two or more differentsignals (lights, sounds, pictures). It includes thespeed with which the correct response is startedwiththehand,foot,orotherbodypart.
1A2c1 ReactionTimeThe ability to quickly respond (with the hand,finger, or foot) to a signal (sound, light, picture)whenitappears.