[237] * The authors are members of GRANECO, research group from Instituto de Estudios Económicos del Caribe, Universidad del Norte. OAXACA-BLINDER WAGE DECOMPOSITION: METHODS, CRITIQUES AND APPLICATIONS. A LITERATURE REVIEW Carlos G. Ospino* Paola RoldánVasquez Nacira Barraza Narváez Fecha de recepción: noviembre de 2009 Fecha de aceptación: diciembre de 2009 REVISTA DE economía DEL CARIBE n° 5 issn: 2011-2106 ARTíCULO DE REFLEXIóN La descomposición salarial de Oaxaca- Blinder: Métodos, críticas y aplicaciones. Una revisión de la literatura
38
Embed
O -B decOmpOsitiOn: methOds applicatiOns. a literature · Oaxaca-Blinder wage decomposition: Methods, critiques and applications. ... independent variables depend on the microeconomic
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.
a b s t r a c tThis document outlines a series of works developed over severaldecades–fromtheoriginalandindependentworksofBlinder(1973)andOaxaca(1973)-withtheaimoffacilitatingthecomprehensionof the technique for wage decomposition and its methodologicalimprovements and applications over time. Similarly, the analyzedapplications serveasaguide for thedevelopmentofworkrelatedto theestimationofwagegapsbetweengroupsof individualsandtheircauses.
k e y wO r d s :WageDecomposition,Blinder,Oaxaca,Heckman,GapWage.j e l c O d e s :J33,J71.
r e s umenEstedocumentorevisaunaseriedetrabajosdesarrolladosalolargodevariasdécadas-desdelostrabajosoriginaleseindependientesdeBlinder(1973)yOaxaca(1973)-conelfindefacilitarlacomprensióndelatécnicaparaladescomposiciónsalarial,susmejorasmetodo-lógicasyaplicacionesa lo largodel tiempo. De igualmanera, lasaplicacionesanalizadassirvendeguíaparaeldesarrollodetrabajosrelacionadosconlaestimacióndebrechassalarialesentregruposdeindividuosysuscausas.
p a l a B r a s c l av e :Descomposiciónsalarial,Blinder,Oaxaca,Hec-kman,BrechaSalarial.c l a s i f i c a c i ó n j e l :J33,J71.
In thisorderof ideas, it is said thatwagediscriminationoccurswhenanindividualsimilartoanother,whoonlydiffersinrace,sexorotherpersonalcharacteristics,receivesalowerwageforotherreasonsthantheperformanceintheirwork.Ingeneral,issociallyandeconomicallyimportanttoidentifywhethertherearewagedifferencesbetweenminoritiesandmajoritiesandwhatitscausesare;Therefore,itcomesasnosurprisethatresearchershavedevelopedmethodstoestimatethedifferencesinwagesandhavedevelopedtheoriestoexplaintheoriginofthesedifferentials.
ThestandardmethodforwagedecompositionoftheOaxacaandBlindermethodologyhasbeenwidelyusedtoexaminediscrimination in the labor market.The technique breaksdowntheaveragewagegapexistingbetweentwodemographicgroupsintotwosummands:thefirstoneshows,differencesinqualifications,thedifferencesthatareexplainedbythemodel;andthesecondoneshows,differencesinthestructureofthemodel,i.e,thosedifferencesnotexplained;theunexplaineddifferenceisanestimateofdiscriminationinthelabormarket.(Oaxaca:1999)
a. TheoreticalFoundations
Ingeneral,thereisconsensustosaythatdiscriminationagainstwomen takesplacewhen the relativewageearnedbymenexceedstherelativewagethatwomenwouldhaveobtained
IntheparticularcaseoftheUSeconomy-asBlinder(1973)mentioned-wasverywellknownthatwhitesearnedwageshigherthanblacksandmenearnedwagessubstantiallyhigherthan women.Therefore, the assumption that earned wageswereequaltothemarginalproductivitycouldnothold.
intercepts of the equation corresponds to discrimination2; how-ever,Blinder(1973)proposesthattheunexplainedportionofthedifferencecomesfromboth:thedifferencesincoefficients,asthedifferencesinaveragecharacteristicsoftheminoritygroup.Blinder(1973)modeleditsproposalasfollows:
D = Portion of the differential attributable to discrimination = C +U”4
Duringestimationitmustbetakenintoaccountthatitiscriticaltouselaborincomesofindividualsasthedependentvariable;thisisduetothefactthat,ingeneral,surveysincludeincomesotherthatlaborincomes.Ontheotherhand,selectedindependent variables depend on the microeconomic wagedeterminationmodelchosen.Bearinginmind,ofcourse,thatcorrelationdoesnotimplycausality,andiftheinterestistofindthecausesofthewagegapyoumustspecifyaquiteaccuratemodel.(Blinder:1973)
c. Firstapplications
i. Blinder(1973):
Blinder(1973)inhisarticle,Wage discrimination: reduced form and structural estimates useswageregressionsofwhitemen,blackmenandwhitewomentoanalyzethewagedifferentialbetweenblackandwhitemenandthewagedifferentialbetweenwhitemenandwomen.
equationincludesBandZvariablesthatareomitted-explic-itly-inthestructuralequation.Overall,theestimationofthereduced form takes into account individuals’ conditions ofbirthtoshowtheconditionalexpectationofwagesearnedbythem.Foritspart,thestructuralestimatesreflectthecondi-tionalexpectationofincomeearnedbytheindividualtakingintoaccountthecurrentsocio-economicstatus.
Byanalyzingtheincomegapbetweenwhitemenandblackmen,Blinderfindsthatcomparingstructuralcalculationsanoverallwageadvantageof50.5%ofwhitesoverblackswasobserved; another finding revealed that 30.7% of the dif-ference (around60%of the total) is attributable toblacks’lowerendowments.Similarly,Blinder(1973)analyzeswagedifferentialsbetweenwhitemenandwomenandfinds thatmostfamilyhistoryvariablesusedtoestimatethestructuralequationoperate in favorofwomen reflecting its influencethroughthecoefficients.
According toBlinder, inregressionsdirect labormarketdiscrimination accounts for around two-thirds of the wagedifferentialandaboutonethirdofthedifferenceofthedis-crimination is due to other endogenous variables, such aslaborsituationandlabortenure.Inotherwords,itdoesnotattribute anypart of the labordifferential todifferences inlaborcharacteristics.
ageandEi thenumberofyearsofstudyapprovedbythei-thindividual. Potential experience and actual experience willmatchonlyifindividualsobtainedtheiremploymentsimme-diatelyaftercompletingtheirstudies;thisdefinitionimpliesthat individuals do not experience unemployment, or thatindividualsdidn’tworkwhiletheystudied.Forwomen,thecalculationofpotentialexperiencehasalimitationassociatedwiththetimespentonhouseholdandchildbearingactivities.Thereforethepotentialexperienceisareasonableapproxima-tiontotherealexperienceofmenbutexaggeratestheactualyearsofworkforwomen.(Oaxaca:1973).
Other explanatory variables used by Oaxaca (1973) aredefinedasfollows:
Education:yearsofschoolingcompleted(linearandquadraticterms);Class of Worker: dummy variables for union membership(privatelyemployedwageandsalaryworker),governmentemployed,andself-employedwithnonunionprivatewageandsalaryworkersasthereferencegroup;
Industry:dummyvariablesforU.S.Censustwodigitindustrieswithretailtradeasthereferencegroup;Occupation:dummyvariablesforU.S.Censustwodigitoc-cupationswithsalesworkersasthereferencegroup;Health Problems:dummyvariable=1iftheindividualreportshealthproblemsthataffectthekindoramountofworkheorshecanperform,and‘0’otherwise;Part-Time:dummyvariable=1iftheindividualworkslessthanthirty-fivehoursaweek,and‘0’otherwise;Migration:a)dummyvariableIiftheindividualhasmaintainedaresidencemorethanfiftymilesfromhisorhercurrentad-dresssincetheageofseventeen,and‘0’otherwise,b)YRSM:numberofyearssincetheindividuallastmigrated(linearandquadraticterms);Marital Status:dummyvariablesforspousepresent,spouseabsent,widowed, anddivorced (or separated)withnevermarriedindividualsasthereferencegroup;Size of Urban Area:dummyvariablesforresidenceinStandardMetropolitanStatisticalAreaslessthan250,000(SMSA<250),greaterthanorequalto250,000butlessthan500,000(SMSA250-500),greaterthanorequalto500,000butlessthan750,000(SMSA500-750), and greater than or equal to 750,000 (SMSA750+)withurban,non-SMSA’sasthereferencegroup;andRegion:dummyvariablesforU.S.CensusregionsNorthEast,NorthCentral,andWestwithSouthasthereferencegroup8.
tion reveals researcher’s attitude towards what constitutesdiscriminationinthelabormarket(Oaxaca:1973).
ThedatausedbyOaxaca(1973)camefromtheSurveyofEconomic Opportunity of 1967.The used sub-sample wasformedbytheintersectionof:individualswhoreceivedasalaryperhourintheweekprevioustothesurvey;adultsaged16orolder,thoselivinginurbanareas,andthosewhoreportedtheirraceaswhiteorblack.
However, theproblemgoes beyond identifying the interceptcomponent.Ingeneral,conventionaldecompositionmethodologiescannotidentifythesinglecontributionfromadummyvariabletothedecompositionofwages,becauseitisonlypossibletoestimatetherelativeeffectsofadummyvariable.(Oaxacaandransom:1999)Finally, Oaxaca and Ransom (1999) claim that almost all wageregression models contain categorical variables, so it is unlikelythatadetailedapplicationofdecompositionscanescapetheaboveidentificationproblem.
iii. Other criticism that can be made to the Oaxaca -Blinderapproachisthatitonlymeasuresthediscriminationinthelabormarket.Iftherearedifferencesinaccesstoendow-mentsthatarerewardedinthelabormarket,-forexample;ifwomenhaveworseaccess tohighereducationthanmen,orevenif,ceterisparibus,menaremorelikelytoworkthanwomen-thentheOaxaca-Blinderstandardapproachtendstounderestimatethediscriminationdegree(Madden:1999).
iv. Furthermore,Atal,ÑopoandWinder(2009)althoughthe decomposition Blinder - Oaxaca is the approach mostappliedinresearchaboutwagedifferentials,ithasthreenote-worthyflaws:First,thedecompositiononlygivesinformation
abouttheaveragewagegapomittingthedifferentdistributionofthisgapamongindividualsofasamegroup.Secondly,au-thorsmentionthatithasbeenobservedthattherelationshipbetween characteristics and wages is not necessarily linear,andfoundthatrecentdataviolatefundamentalimplicationsoftheMincermodel,whichisthekeyinputofthedecomposi-tions.Finally,thedecompositiondoesn’trestricttheanalysistocomparableindividuals,whichcanleadtoanupwardbiasofthecomponentassociatedwithdiscrimination.
v.Fromanempiricalperspective,themostseriousproblemthatthismethodologyhasisthatsinceestimatesofthecoef-ficientscapturebiasesgeneratedfrominformationproblems,errorsinthevariablesandselectivityprocesses,theinterpre-tationofthisresidualasameansofdiscriminationisdebat-able.However,itisworthmentioningthatnoempiricalworkis without problems and methodological questions. (Tenjo,RiveroandBernat:2002)
Heckman(1979)discussesthebiasresultingfromtheuseofnon-randomlyselectedsamplestoestimatetherelationshipsbetween variables.The author mentions, in contrast to theusualanalysisof“omittedvariables”inEconometrics,thatintheanalysisofsampleselectionbiasissometimespossibletoestimate the variables that give rise to specification errors.Theestimatedvaluesoftheomittedvariablescanbeusedasregressorsallowingestimatebehavioralfunctionswhicharetheinterestobjectoftheinvestigation.
The final assumption is a consequence of the randomsamplingscheme.JointdensityofU1i,U2i is (U1i,U2i).Theregressorsmatrixisoffullranksoifalldatawereavailable,eachequationparameterscouldbeestimatedbyOLS.
MaradonayCalderón(2000)exposedthattheerrortermhas conditional expectation with zero mean, however, nonconditionalexpectation13,whichisusedwhenweuseOLS,hasanon-zeromeanandcorrelateswiththeindependentvariables;
Thefirst corresponds to the ratioof thedensity functionandthecumulativedensityassessedonthevalue foreachindividualfunction.ThistermisknownastheinverseoftheMill’sratioandistheexcludedvariableintheanalysisofthemarketwageequationuncorrected for selectionbias.Thesecondtermisthecoefficientofthetheoreticalregressionequations(A)and(B)errors.Thus,themarketwageequa-tioncorrectedforthepresenceofselectionbiasinempiricaltermsisstatedasfollows:
Theterm correspondstothepartofthewagegap that can be explained by differences in the observedcharacteristics of individuals.On theotherhand, the term
reflectstheunexplainedpartofthegap,whichisduetodifferencesinthecoefficientsofXi,thistermisregardedastheeffectsofdiscrimination.Theterm isduetoselectionbiasandisgeneratedbythedifferencesbetweenthe incorporation pattern into the labor market men andwomenhave.
Selectionbiascorrectionhasnotbeenfreeofcriticism.AsinLewis(1986)inmostcasesthereisnotatheoreticalmodeltoexplain the specific selectionprocess, and to specify thevariablestoexplainit.Intheabsenceofthistheory,whatisusuallydoneistoincludeadhocvariableswhicharethought
If the only reason for not reporting income is the factthatpeoplearenotinvolved,youmightthinkofapplyingtheHeckmancorrectiontoanequationofalaborparticipation,whosetheoreticalsupportisprettysolid.However,inpractice,nonreportofincomemaybeduetootherthings(otherthannon-participation)suchasopen-wideunemploymentorem-ploymentinoccupationsoffamilyassistantwithoutadefinedremuneration.Thisgeneratesadditionalcomplicationswhicharenotclearlydiscernible.(Tenjo,RiveroandBernat:2002)
f. Applications
i. Genderwagegap
Attheinternationallevel,severalauthorshaveexaminedthewage differential between men and women and its causesthrough wage decomposition a la Blinder–Oaxaca; findingrevealing resultsof theexistenceofdiscrimination inmostcases.Basicallyalloftheseauthorsestimatedmincerianincomeequationsasfollows:
Inthewordsoftheauthor,thefundamentalpropertyofthisvariableisthatit“punishes”incomeofthoseworkingmorethan40hoursperweekand“rewards”incomeofpart-timeworkers, i.e., homogenizes the equivalent to 40 hours perweekforthemarketincome.
TheobtainedresultsshowedthatthedifferencesinthegapbetweenGranBuenosAiresandtheArgentineNorthwestweresmallanditislikelythattheyrespondmoretotheroundingof numbers rather than economically relevant issues. Ob-servedgapislocatedinnotlessthan0.70ormorethan0.85,dependingonthespecificationofempiricalmodelsusedto
The authors estimated Mincer equations for men andwomen,withandwithoutselectioncorrection,foremployedandunemployed,foreachofthesixcountriesinthestudy.Thedependentvariableusedwasthehourly income;tocorrectforselection,anestimationofHeckman’stwo-stepmethodwasused.
Similarly,Tenjo, Rivero and Bernat (2002) observed asignificant level of job segregation by branch of economicactivityandoccupations.Howevertheymakeclearthatthissegregationisnotreflectedinlowerwagesforwomen.Thus,hourlywagedifferentialsbetweenmenandwomenappeartobeassociatedwithlaborremunerationpatternswithinsectorsandoccupation.
• DiPaolaandBerges (2000),calculated thegenderwage gap estimating income Mincerian functions forbothsexesandapplyingthetechniqueofBlinderandOaxaca.
The dependent variable, the logarithm of the monthlyhours,wascalculated fromthehoursperweekdeclared intheSurvey17bytheworker,dividedby7andmultipliedby30.Theyexpectedthatsignsandapproximatevaluesofthecoefficientsof the income functionmatched thehypothesisofhumancapital.Theauthorsalsostatedthatthecoefficientofthenaturallogarithmofmonthlyhoursworkedshouldbepositiveanditwouldbeinterpretedasanelasticity,indicatinginwhatproportionvariestheincometoapercentagechangeinhoursworked(DiPaolaandBerges).Theycorrectedtheproblem of selection bias by estimating labor participationequationasmentionedabove.
The model suggested by Dipaola and Berges (2000) forselectionbiascorrectionwasasfollows:
• Johanson,KatzandNyman(2005)discusstheevolu-tionofwagedifferentialsand the factors thatmayberelated,yearbyyearthroughtheanalysisofcrosssectiondataoftheDepartmentofstatisticsofSweden(HEK)fortheyears1981and1983-98;thereforedecomposethewagedifferentialgenderaccordingtothestandardmethodofOaxaca(1973).
Thedependentvariableonestimatesis(thelogarithmof)imputedhourlywage.Variablesused in thewageequationsareagesquared,fictionalvariablesfor:theleveloftraining,requirementstobeablue-collarorwhite-collarworker,in-dustry,region,thecentralgovernment,theprivatesector,orlocalgovernmentemployeesandtoworkinmen-dominatedorwomen-dominatedoccupations(Johanson,KatzandNyman:2005).Authorsviewageasameasureoftheexperienceoflifeingeneral,thatcanaffectthewages.
Bernat(2005)estimatedwagedecompositionofOaxaca-Blindercorrectingforselectionbias,therefore,firstestimatesthe participation equation, using variables such as maritalstatus,headofthehousehold,theunemploymentrateofthefamily and the number of children, characteristics whichclearly affect theparticipationof an individual in the labormarket.Theincomefortherestofthefamilyisincludedasanapproximationoftheindividual’sreservationwage.Thehigherthisincome,thehigheritisexpectedthattheindividualhasmoretimetoanalyzeemploymentopportunitiesinthelabormarket,thecrossovereffect(cc)thatitincludesinvestigateswhetherbeingamarriedheadofthehouseholdorunmarriedhouseholdheadaffectsthedecisionoflabormarketparticipa-tion.(Bernat:2005).
Racial discrimination by industry estimates are calculatedby,meaning;thewagegapbetweenwhitesandblacks inthesame industry,afterdiscountingthedifferences in theindividualcharacteristics18
According to the results found by Johnson, employersclassifiedascostsminimicersdiscriminatemoreinwagesthatemployerswhocanoperatemoreindependentlyofthecom-petitivemarketdiscipline.Theregulatedcompaniesseemtodiscriminatelessthanotherfor-profitemployers.
According to the author, estimated results revealed thatbetween75and100%ofthedifferentialaverageincomewascaused by differences in occupational level, indicating the
problemreferstothepolicyofentrytoworkinsteadofdif-ferentpaymentswithinasameoccupationlevel. Inspiteofchanging economic conditions, no evidence of movementsontheaveragewagedifferentialintheperiod1970-1975wasfound.Finally,notedthatduringthisperiodblackimmigrantshaveprogressedconsiderablyless,professionally,thanwhitesbornintheUnitedKingdom.(Stewart:1983)
• Reimers(1983)attemptstomeasurethedegreeofwage discrimination against Hispanics and Blacks intheUnitedStates.TheauthorusesmicrodatafromtheSurveyofIncomeandEducationfrom1976toestimateseparate wage functions, selection bias corrected foreach ethnic group -Mexican, Puerto Rican, Cuban,CentralAmericanandSouthAmerican,otherHispanic,non-Hispanicandwhitenon-Hispanicblacks.Samplesconsistofwagesforemployedmalecivilians,14orolder,whowerenotself-employedorfull-timestudents.
ReimersfoundthatformenofPuertoRico,discrimina-tionmayberesponsiblefor18%ofthewagedifferential,onits part, for“other Hispanics”, discrimination can cause upto12%ofthewagegap.ForMexicans,however,accountsof
In this article we make an effort to illustrate a techniquefortheanalysisofwagediscrimination.Wefirstexposetheindependent works of Blinder (1973) and Oaxaca (1973),followed by criticism and methodological improvements ithas experienced over time.We also summarize empiricalimplementationsofthetechniquefortheanalysisofgenderandracialdiscrimination.
This methodology has been widely used, in spite of itslimitationsanddisadvantages,especiallytocalculatethewagedifferentialbetweenmenandwomenandtodemonstratetheexistenceofdiscriminationasakeycauseofthosedifferences.Amethodologicalimprovementbymanyresearchersistoincludeathirdsummandrepresentingtheexistenceofselectionbiasinthesample;thistermisenteredusingtheestimateproposedbyHeckman’stwo-stepmethodin1979.
Subsequently,severalauthorshavedevelopedafewexten-sionsoftheoriginalmethodproposedbyOaxacaandBlinder;however,theanalysisofthesemethodologieswasnotaddressedinthisarticledueto its lengh.Therefore,ananalysisofthe
methodologiesthatextendtheoriginalconceptionofwagedecompositionofOaxaca-Blinderisproposedasatopicofinterestandwillbecarriedoutinasubsequentpaper,wherewewillreviewtechniquesthatfocusnotonlyonaveragedif-ferences but that also decompose, indexes and distributionfunctions.
REFERENCES
Atal, J., Ñopo, H. &Winder, N. (2009). New Century, OldDisparities:GenderandEthnicWageGaps inLatinAmerica.IDB Working Papers,109,76.
Bernat, L. (2005). Análisis de género de las diferencias salariales en las siete principales áreas metropolitanas colombianas: ¿Evidencias de discriminación?CuadernosPNUD–DANE.
Blinder,A.S. (1973).Wage Discrimination Reduced Form andStructuralEstimates.Journal of Human Resources,8,(4),436-455.
DiPaola,R.&Berges,M.Sesgo de selección y estimación de la brecha por género para Mar del Plata. CentrodeInvestigacionesEconómicas-UniversidadNacionaldeMardelPlata.
Paz,J.(1998).Brecha de ingresos entre géneros. (Comparación entre el Gran Buenos Aires y el noroeste argentino).
Stewart, M. (1983). Racial Discrimination and OccupationalAttainmentinBritain.The Economic Journal,93,(371),521-541
Tenjo,J.,RiberoR.&Bernat,L.(2002).EvolucióndelasdiferenciassalarialesporsexoenseispaísesdeAméricaLatina.Colección Documentos de Economía1,59.
Tenjo,J.&Herrera,P.(2009).Losensayossobrediscriminación:discriminaciónsalarialydiscriminaciónenaccesoalempleopororigenétnicoyporgénero,Colección Documentos de Economía1.