The gender wage gap and occupational segregation in the Mexican labour market

Transcription

1 The geder wage gap ad occupatioal segregatio i the Mexica labour maret Pedro Orraca 1 Fracisco-Javier Cabrera 2 Gustavo Iriarte 3 Abstract: This study examies the role of occupatioal segregatio i explaiig the geder wage gap i Mexico. Usig cesus data ad based o the Brow et al. (1980) decompositio, it is observed that male-female wage differetials icreased betwee 2000 ad For both years, withi occupatio wage differetials geerally icreased the geder wage gap, whereas betwee occupatio wage differetials provided the opposite effect. Sice withi occupatio wage differetials are drive by the uexplaied compoet, the results suggest that the geder wage gap is primarily a product of differeces i the average returs to productivity related characteristics withi occupatios. Occupatioal segregatio does ot icrease malefemale wage differetials sice wome do ot appear to ecouter barriers ito high payig occupatios. This result is robust to the use of differet levels of occupatioal aggregatio ad whe correctig the wage equatios for selectivity-bias. Keywords: Occupatioal segregatio, geder wage differetials, Mexico. jel Classificatio: J31, J24, J71. Resume: Este estudio examia el papel que desempeña la segregació ocupacioal sobre la brecha salarial de géero e México. Mediate iformació cesal y co base e la descomposició propuesta por Brow et al. (1980), se observa que la brecha salarial de géero se icremetó etre 2000 y Las diferecias salariales detro de las ocupacioes tiede a icremetar la brecha salarial total, mietras que las diferecias salariales etre ocupacioes preseta el efecto opuesto. Dado que las diferecias salariales detro de las ocupacioes so producto del compoete o explicado, los resultados sugiere que la brecha salarial es primordialmete producto de diferecias e el redimieto de las características productivas de los trabajadores. La segregació ocupacioal o icremeta la brecha salarial dado que las mujeres o efreta barreras para acceder a ocupacioes altamete remueradas. Este resultado es robusto al uso de distitos iveles de agregació ocupacioal y a la correcció por sesgo de auto-selecció. 1 Departmet of Ecoomics. Uiversity of Sussex. p.orraca-romao@sussex.ac.u 2 Divisió de Admiistració Pública. Cetro de Ivestigació y Docecia Ecoómicas A. C. fracisco.cabrera@cide.edu 3 Departmet of Ecoomics. Uiversity of Sussex. c.iriarte-rivas@sussex.ac.u EcooQuatum / Vol. 13. Núm. 1 51

2 52 EcooQuatum Vol. 13. Núm. 1 Palabras clave: Segregació ocupacioal, diferecias salariales de géero, México. Clasificació jel: J31, J24, J71. Recepció: 09/07/2014 Aceptació: 16/02/2015 Itroductio I recet decades, due to a series of factors icludig demographic ad cultural chages, a opeig of its ecoomy, a rise i the levels of formal educatio, the implemetatio of structural reforms which restructured the coutry s ecoomy towards a more service orieted oe, ad a series of crises that pushed wome ito the labour force, amog others, female participatio rates i Mexico have substatially icreased. Accordig to iformatio from Mexico s populatio cesuses, the percetage of wome betwee 18 ad 65 years of age participatig i the labour force grew from 19.4% i 1970, to 24.2% i 1990 ad to 42.3% i Cosequetly, aalysis of the geder wage gap have become icreasigly importat, where a large umber of studies have examied the evolutio of male-female wage differetials i Mexico (e.g. Alarcó ad McKiley, 1994; Pagá ad Ullibarri, 2000; Popli, 2013). The rise i female labour participatio rates has bee accompaied by costat chages i the coutry s occupatioal structure ad distributio of wages betwee occupatios (see, e.g. Brow, Pagá ad Rodríguez-Oreggia, 1999; Redó, 2003; Calóico ad Ñopo, 2009). The preset study attempts to examie these pheomea ad cotribute to the literature by examiig the role of occupatioal segregatio i explaiig the geder wage gap i Mexico. 4 A uderstadig of the factors drivig male-female wage differetials is relevat sice it helps direct public policies ad evaluate if previous efforts to reduce geder wage iequalities have bee rewarded. The majority of studies that have examied geder wage differetials i Mexico have doe so uder the framewor developed by Oaxaca (1973) ad Blider (1973), focusig o the issue of wage discrimiatio or uequal pay for equal wor. Commoly referred to as the idex umber methodology, uder Oaxaca-Blider (OB) type decompositios, the discrimiatio effect is calculated as the uexplaied differece betwee two reduced-form wage equatios estimated separately for me ad wome. The estimatio of these two wage equatios is ormally performed cotrollig for a stadard set of productivity related variables, icludig variables deotig the idividual s occupatioal level, with the purpose of separatig the effects 4 Followig the related literature (e.g. Liu, Zhag ad Chog, 2004), occupatioal segregatio is said to exist if worers are assiged to differet occupatios based o o-productivity related characteristics. This is ot to be cofused with occupatioal structure, which refers to the distributio of worers betwee occupatios. Differeces i the occupatioal structures of me ad wome do ot ecessarily suggest that there is discrimiatio i the labour maret. Noetheless, if wome face barriers to etry ito occupatios or if their occupatioal choices are limited by lower returs to their productivity related characteristics withi occupatios, the differet occupatioal structures do ideed reflect discrimiatio (Ehreberg ad Smith, 2012).

3 The geder wage gap ad occupatioal segregatio of discrimiatio from those of legitimate compesatig differetials (Reilly, 1991: 1). While icludig occupatioal cotrols allows iterpretig the effect of the differet occupatioal distributios of me ad wome, doig so is problematic sice it assumes that occupatios are exogeously give. Cosequetly, if occupatioal determiatio is subject to labour maret discrimiatio, the this approach would be iappropriate (Liu et al., 2004). Sice the OB approach does ot properly separate the effects of wage ad job discrimiatio from those of occupatioal segregatio, Brow, Moo ad Zoloth (1980) suggest a alterative methodology that treats occupatioal choice as edogeously determied, ad allows a decompositio of the wage gap ito withi ad betwee occupatioal compoets. 5 Uder the Brow, Moo ad Zoloth (BMZ) framewor, the withi ad betwee compoets ca be further decomposed ito effects attributable to differeces i observable characteristics ad to differeces i the estimated coefficiets. I the cotext of betwee occupatioal effects, differeces i the estimated coefficiets provide a explicit estimatio of the effect of occupatioal segregatio o the geder wage gap. Idetifyig the relative importace of betwee agaist withi occupatioal effects is importat from a policy perspective, sice the results may shed light o whether policies directed more at promotig equal pay withi occupatios should be prioritised, istead of measures that ecourage a more equal access to differet occupatios (Liu et al., 2004). 6 The preset study cotributes to the literature i several ways. First, a updated aalysis based o the BMZ decompositio for the period is performed. Furthermore, the Duca ad Duca (1955) idex of dissimilarity is also preseted i order to provide a objective measure of the degree to which the occupatioal structures of males ad females differ. Secod, ulie previous studies that have emphasised the role of occupatioal segregatio, we use cesus iformatio which allows us to iclude rural localities ad therefore study the etire Mexica labour maret. Followig Kidd ad Shao (1996) who raise awareess cocerig the use of broadly defied occupatioal categories whe performig wage decompositios, the much larger sample size of the cesus eables us to aalyse the sesitivity of the results to the use of differet levels of occupatioal aggregatio. Third, i additio to the poit estimates, we preset stadard errors for the differet compoets of the Brow et al. (1980) decompositio. Fially, the sesitivity of the results is aalysed where the 5 I recet years, a variety of decompositio techiques that focus o aalysig wage differetials beyod the mea have bee developed (e.g. Machado ad Mata, 2005; Firpo, Forti ad Lemieux, 2009). Nevertheless, ulie the Brow et al. (1980) methodology, these decompositios are uable to properly idetify the effects of occupatios o wages sice they treat occupatios as beig exogeously determied. Sice the preset study examies the role of occupatioal segregatio i the geder wage gap, the Brow et al. (1980) decompositio represets our preferred methodology. Cocerig the alterate method developed by Ñopo (2008), it has bee recetly applied to the Mexica case by Calóico ad Ñopo (2009). 6 The iteratioal literature that has aalysed geder wage differetials usig the BMZ methodology icludes the origial wor of Brow et al. (1980) for the U.S., Miller (1987) ad Dolto ad Kidd (1994) for the U.K., Reilly (1991) for Irelad, Kidd ad Shao (1994, 1996) for Caada, ad Salardi (2012) for Brazil, amog others. These studies have geerally cocluded that most of the geder wage gap is a product of withi occupatio effects, where occupatioal segregatio teds to play oly a mior role.

4 54 EcooQuatum Vol. 13. Núm. 1 estimated wage equatios are corrected for self-selectio ito each occupatio usig the methodology outlied by Lee (1983). Amog the mai results it is observed that geder wage differetials icreased betwee 2000 ad For both years, withi occupatio differetials positively cotributed to the geder wage gap, whereas betwee occupatio differetials provided a egative effect. Sice withi occupatio differetials are drive by the uexplaied compoet, the results suggest that the geder wage gap is primarily a product of differeces i the average returs to productivity related characteristics withi occupatios. Furthermore, occupatioal segregatio does ot icrease the geder wage gap sice wome do ot appear to ecouter barriers ito high payig occupatios. This result is robust to the use of differet levels of occupatioal aggregatio ad whe correctig the wage equatios for selectivity-bias. The study proceeds as follows. Sectio 2 presets a review of the related literature. Sectio 3 outlies the empirical ad ecoometric methodologies. Sectio 4 describes the data ad presets descriptive statistics. Sectio 5 discusses the mai results. Sectio 6 tests the sesitivity of the results. Sectio 7 cocludes. Literature review The literature that has examied geder wage differetials i Mexico is extesive. These studies usually report that wome ear less tha me, where the magitude of the geder wage gap fluctuates depedig o the period of study ad the geographical coverage of the data source used. Amog the studies that have used the OB framewor to examie geder wage differetials, Alarcó ad McKiley (1994), based o data from the Ecuesta Nacioal de Igresos y Gastos de los Hogares (ENIGH, Natioal Household Icome ad Expediture Survey) from 1984, 1989 ad 1992, observe that the geder wage gap icreased from 23.3% i 1984, to 28.4% i 1989 ad to 25.3% i I additio, for all three periods, the scholars fid that at least 70.0% of the wage gap is attributed to discrimiatio, i.e. to differeces i the ecoomic returs to average characteristics. 7 Based o data from the ENIGH from 2000, Martíez ad Acevedo (2004) observe that approximately 85.0% of the geder wage gap is a product of differeces i the ecoomic returs to average characteristics. Alteratively, Medoza ad García (2009), usig data from the Ecuesta Nacioal de Ocupació y Empleo (ENOE, Natioal Occupatio ad Employmet Survey) from 2006, observe that the majority of the geder wage gap, i.e. 87.6%, is a product of differeces i average characteristics. Moreover, the authors report that this figure varies sigificatly by regio, where geder wage discrimiatio is smallest i the orth of Mexico ad largest i the south. Usig a alterate methodology, Pagá ad Ullibarri (2000), based o data from the Ecuesta Nacioal de Empleo Urbao (ENEU, Natioal Urba Employmet Survey) from 1995, aalyse the role of group heterogeeity, i.e. differeces by educatio level, 7 A commo icoveiece with OB type decompositios is that all of the variables that affect wages are rarely available, i which case the extet of labour maret discrimiatio will ted to be overstated.

5 The geder wage gap ad occupatioal segregatio age, firm size, sector of employmet, ad regio of residece, o the geder earigs gap. The authors compute selectivity corrected earigs equatios ad estimate a additively decomposable idex of the extet of geder uexplaied wage iequality based o Jeis (1994). The study shows that the uexplaied wage gap is larger amog worers with high educatio levels, which have full-time jobs, wor i micro-firms, ad are employed i the iformal sector. O the other had, Pagá ad Sáchez (2000) use the Ecuesta Regioal de Servicios Fiacieros a Uidades de Protecció Rural (SREFS, Survey of Rural Etrepreeurs ad Fiacial Services) to study the factors that explai geder differeces i worforce participatio ad i the icidece of selfemploymet i rural Mexico. Estimatig a bivariate probit to aalyse the propesity of idividuals to eter the worforce ad examie the icidece of self-employmet amog the employed, the authors show that if the female employmet rate is predicted uder the assumptio that wome face the male labour maret structure, geder employmet rate differetials almost disappear. While 88.2% of geder employmet rate differeces ca be explaied by structural demad ad supply factors, oly 11.8% is attributable to differeces i idividual edowmets. Although the authors do ot explicitly explore geder wage differetials, the results suggest that geder-related labour maret discrimiatio is liely to be preset i the wage or salaried sector. Studies that emphasise the role of occupatios whe examiig the geder wage gap i Mexico iclude Sáchez (1998), where the author employs the BMZ decompositio to evaluate the impact of betwee ad withi sectoral factors o male-female earigs differetials amog rural ad urba microeterprises. Based o data from the SREFS from 1994 ad the Ecuesta Nacioal de Microegocios (Natioal Survey of Urba Microeterprises) from 1992, it is observed that female-headed microeterprises i rural ad urba areas ear 64.0% ad 50.0% less, respectively, tha male-headed microeterprises. I additio, a estimated 35.0% of the earigs gap i urba areas ca be attributed to differeces i average characteristics, whereas i rural areas this figure rises to 42.0%. Oly a small share of the geder earigs gap is explaied by differeces i sectoral distributios. I a descriptive aalysis, Redó (2003) examies geder differeces i employmet, segregatio ad wages. Based o various data sources icludig the 1990 ad 2000 Populatio Cesuses ad the Ecuesta Nacioal de Empleo (ENE, Natioal Employmet Survey) from 2000, the author reports that male hourly wages surpass those of females by approximately 9.0%, where this gap varies sigificatly depedig o the sector of employmet. I additio, the scholar reports that wome ted to be cocetrated i specific occupatios. Whe aalysig the sector of employmet of wome, a icrease i segregatio by productio sector betwee 1990 ad 2010 is observed. Moreover, the study reports that there is high degree of vertical segregatio i the Mexica labour maret, where me ted to occupy higher raig ad better payig positios tha wome. 8 8 The evet whe me ad wome are clustered i differet occupatios is commoly referred to as horizotal or occupatioal segregatio, ad is ormally reflected i wage disadvatages betwee occupatios. The evet whe geder differeces are preset withi a sigle occupatio where worers occupy differet categories is commoly ow as vertical or hierarchical segregatio, ad is ormally reflected i wage disadvatages withi occupatios.

6 56 EcooQuatum Vol. 13. Núm. 1 Usig data from the ENEU, Brow et al. (1999) examie the effect of occupatioal attaimet o the icrease i the geder wage gap betwee 1987 ad The study uses the Welligto (1993) decompositio to disaggregate wage gap chages ito relative chages i persoal characteristics ad chages i the returs to these characteristics. The authors also decompose geder differeces i earigs betwee occupatios ad aalyse the role of occupatioal attaimet i the overall earigs gap by usig the BMZ methodology. The results show that durig the period of study there was a icrease i the geder earigs gap attributed to relative chages i huma capital edowmets sice, all else equal, wage coefficiet chages would have slightly reduced the geder earigs gap. Moreover, while i 1987 the explaied compoet accouted for 68.0% of the earigs gap, i 1993 it represeted close to 73.0%. Furthermore, the geder earigs gap was mitigated by a substatial declie i geder differeces i occupatioal attaimet. The scholars also use the OB decompositio to study the effects of the occupatioal structure o the wage gap. It is observed that the majority of the earigs gap is a product of uexplaied differeces. Whe occupatioal cotrols are icluded, the authors fid that occupatioal segregatio reduces earigs differetials. I aother study, Calóico ad Ñopo (2009), based o data from the ENEU from 1994 to 2004, use a matchig comparisos techique to examie the role of idividual ad family characteristics i determiig geder segregatio ad wage gaps. The scholars report that the elimiatio of vertical segregatio would decrease the geder wage gap by 5.0 percetage poits. O the other had, the elimiatio of occupatioal segregatio would icrease geder wage differetials by 6.0 percetage poits. Furthermore, it is observed that, durig the period of study, the role of occupatioal segregatio i determiig the geder wage gap icreased, whereas the relative importace of vertical segregatio decreased. Recet studies have focused o studyig the geder wage gap at differet parts of the wage distributio. Amog these, usig primarily cesus data from 1990 to 2010, employig the methodology put forward by DiNardo, Forti ad Lemieux (1996) ad correctig the aalysis for selectio of wome ito the labour maret, Campos- Vázquez ad Arceo-Gómez (2014) observe a cosistet patter of sticy floors ad a decreasig patter of glass ceiligs over the period of study. I additio, the authors report that the majority of the geder wage gap is a result of differeces i prices. Moreover, the scholars show that the geder wage gap becomes larger whe correctig for selectio, which suggests that wome are positively selected ito the labour maret. Fially, Popli (2013) also examies geder wage differetials throughout the etire distributio i both the formal ad iformal sectors based o data from the ENIGH from 1996 ad Taig ito accout the probability of employmet ito the wage or salaried sector, the study fids that the geder wage gap differs cosiderably by sector of employmet. Cotrary to raw geder wage differetials, the selectio-adjusted geder wage gap is higher i the formal sector compared to the iformal sector. Usig differet techiques icludig the OB decompositio, the measure proposed by Jeis (1994) ad o-parametric techiques, the scholar shows that the uexplaied geder wage gap varies accordig to the sector of employmet, where the uexplaied term is

7 The geder wage gap ad occupatioal segregatio higher i the formal sector relative to the iformal sector. The study also reports that i both sectors the uexplaied geder wage gap decreased durig the period of aalysis, where there is evidece of a glass ceilig i the formal sector, with the uexplaied male-female wage gap beig higher i the upper tail of the distributio. Methodology The decompositio developed by Brow et al. (1980), i.e. the BMZ decompositio, requires the estimatio of occupatio-specific wage regressios for male (m) ad female ( f ) worers: (1) w m m m i = X W ib + f (2) w f f f i = X W b + f m i f i i where w i represets the atural logarithm of hourly wages for idividual i employed i occupatio = 12,,..., K; bw is a vector of wage coefficiets specific to occupatio, ad f i is the error term. Moreover, let w m ad w f deote mea male ad female f wages, ad P m ad P represet the proportio of male ad female worers employed i occupatio. Thus, the overall mea geder wage differetial ca be calculated from Eqs. (1) ad (2) i the followig maer: K m f (3) w - w = / ^ P m w m - P f w fh = / / 1 K m f w w P f X m X f m K m m X P m f - = ^ - hwb + Wb ^ - WP h = 1 = 1 / / K (4) PX f f m f K m m f m + `Wb -Wb j + X Wb ^WP - P h = 1 = 1 where addig ad subtractig terms o the right had side of Eq. (3) ad maipulatig terms yields Eq. (4). I the decompositio, X is a vector of mea values of the f exogeous variables, ad W P deotes the proportio of female worers who would be employed i occupatio if they were to face the same occupatioal structure as that ecoutered by male worers. The first compoet o the right had side of Eq. (4) captures wage differetials derived from differeces i average characteristics withi occupatios, while the secod compoet explais differetials geerated by dissimilarities i participatio shares betwee occupatios. These two terms costitute the withi explaied (WE) ad betwee explaied (BE) compoets, respectively. The third term reflects compesatig differetials withi occupatios, while occupatioal segregatio ad differet prefereces betwee the two groups are captured by the fourth term. These two terms costitute the withi uexplaied (WU) ad betwee

8 58 EcooQuatum Vol. 13. Núm. 1 uexplaied (BU) compoets, respectively. The WU compoet ca be uderstood as the wage effect of vertical or hierarchical segregatio, sice it captures uexplaied differeces i the structure of occupatioal achievemet or the uobserved factors that determie why certai groups are uder- or overrepreseted i higher payig positios withi idividual occupatios (Salardi, 2012). O the other had, the BU term may be iterpreted as the wage effect of horizotal or occupatioal segregatio, sice it provides a idicatio for the equality of access betwee me ad wome to various occupatios (Liu et al., 2004). I additio to the occupatio specific wage regressios, the BMZ decompositio f requires a method for predictig W P. Due to data costraits, the latter is doe through a estimatio of a reduced form multiomial logit (MNL) model. The MNL captures how differet variables affect the probability of a idividual worig i a give occupatio, while treatig the occupatioal choice as edogeously determied. This probability may be defied as: exp^ziwc h (5) Pi = Pr^yi = Occh = K-1 + h 1 + exp^ziwc h where Z i is a vector of labour supply ad demad related exogeous variables presumed to determie occupatioal attaimet; cw is a vector of coefficiets correspodig to occupatio, ad h i is the error term. The reduced form MNL captures both supply drive differeces i group prefereces ad demad drive costraits. Both of these factors are liely to geerate differeces i the estimated coefficiets betwee me ad wome, where the limited access to some occupatios for female worers suggests the presece of occupatioal segregatio (Démurger et al., 2009: 613). 9 f To obtai W P, the estimates of the parameters of Eq. (5) are calculated for males ad the data for females is subsequetly substituted ito the estimated equatios. This produces for every female observatio a vector of predicted probabilities of belogig to each occupatio. The, these probabilities are added over observatios to produce the predicted or o-discrimiatory occupatioal distributio of each female worer. / = 1 i Data descriptio The data used i this study comes from a 10.6% ad 10.0% radom sample of Mexico s 2000 ad 2010 Cesus of Populatio ad Housig, respectively. The iformatio is collected by the Istituto Nacioal de Geografía y Estadística (INEGI, Natioal Istitute of Geography ad Statistics) ad provided by IPUMS-Iteratioal (2013). The sample icludes full-time salaried worers betwee the ages of 18 ad 65 years 9 Potetial demad factors drivig the BU compoet ca be for example employers affiity to wor with people from the same or the opposite geder, a preferece for either male or female worers based o perceived geder differeces i the ability to lead or follow or relate to other worers, or the belief that certai occupatios should be occupied by me ad other by wome, amog others.

9 The geder wage gap ad occupatioal segregatio that report o-zero wages. 10 Furthermore, worers located at the bottom ad top 0.5% of the wage distributio are omitted i order to elimiate the effects geerated by the presece of outliers. All figures are i real Jue 2010 Mexica pesos (MXP). Give its substatial sample size, the use of cesus data provides a clear advatage over Mexico s traditioal labour force surveys, i.e. the ENEU, ENE ad ENOE. The cesus allows havig a larger umber of observatios for each idividual occupatio ad allows usig a more disaggregated occupatioal classificatio scheme. The use of smaller samples has its drawbacs as it may lead to the use of more broadly defied occupatioal categories, which i tur may hide wage differeces ad heterogeeity withi geerally defied occupatios. Descriptive statistics Table 1 presets descriptive statistics, where sigificat differeces betwee male ad female characteristics are observed. Amog these, for both 2000 ad 2010 mea log hourly wages of me are larger tha those of wome. Specifically, this figure is.021 log poits higher i 2000 ad.026 log poits higher i A similar relatioship is observed regardig the age structure, where i both periods me are slightly older tha wome. Give that age is positively correlated with labour maret experiece, this suggests that oe of the reasos why male wages are higher is because o average they have more experiece tha their female couterparts. With respect to huma capital levels, o average female worers have higher educatio levels tha males. This differece slightly icreased durig the period of study, goig from 1.34 years i 2000 to 1.36 years i The fact that wome have higher levels of educatio but lower hourly wages tha me implies that their returs to educatio may be smaller. O the other had, me have substatially loger wor-wees tha wome ad have a lower lielihood of egagig i part-time wor. The differece i the size of the wor-wee is oe of the mai factors behid the observed geder wage gap whe focusig o weely wages. Moreover, for both 2000 ad 2010, wome have higher participatio rates i the formal sector tha me, which teds to be associated with higher productivity levels ad therefore higher wages. I additio, male salaried worers have a higher probability of beig married tha their female couterparts. Sice labour force participatio rates differ substatial betwee male ad female worers, this may be sigallig that wome are liely to drop out of the labour force oce they egage i marriage. I summary, taig the above differeces ito cosideratio, it caot be ruled out that the geder wage gap may be drive by socioecoomic ad demographic characteristics istead of other factors such as occupatioal segregatio Full-time employmet is defied as worig 30 or more hours per wee. 11 These figures ca be easily coverted to percetage terms. Let x represet the differece i mea log hourly wages betwee me ad wome, the percetage wage differetial is the e x Sice the sample used i the study oly icludes full-time salaried worers betwee the ages of 18 ad 65 years that report o-zero wages ad omits worers located at the bottom ad top 0.5% of the wage distributio, some of the descriptive statistics preseted i Table 1 seem couterituitive ad ot i lie with those commoly observed i the literature. This arises because other studies frequetly do ot distiguish betwee worers accordig to their icome level, umber of hours wored or status as salaried or selfemployed worers.

10 60 EcooQuatum Vol. 13. Núm. 1 Table 1 Descriptive statistics, meas ad stadard errors Variable Me Wome Me Wome Log hourly wages (0.706) (0.711) (0.639) (0.665) Age (11.14) (10.13) (11.49) (10.74) Schoolig (4.407) (4.253) (4.337) (4.229) Hours wored per wee (6.085) (6.514) (6.115) (6.497) Formal Married Rural Part-time employmet Labour force participatio Uemploymet rate Observatios 780, , , ,410 Source: Authors elaboratio based o the 2000 ad 2010 Mexica Cesus. Data icludes idividuals betwee 18 ad 65 years of age. Hourly wages are i atural logarithms ad i real Jue 2010 MXP. Stadard errors are i paretheses. Turig our attetio to mea log hourly wages withi occupatios, some iterestig results emerge. As see i Table 2, for both periods withi all occupatioal categories me have higher wages tha wome. I the year 2000, the geder wage gap was the largest at the Crafts ad related trades worers category, risig to.303 log poits, ad it was the smallest at the Techicias ad associate professioals category, stadig at.048 log poits. At the highest payig occupatio, i.e. the Legislators, seior officials ad maagers category, the male-female wage differetial was also substatial, where it was equal to.210 log poits. I 2010, the geder wage gap was the largest at the Plat ad machie operators ad assemblers category, risig to.277 log poits, ad it was the smallest at the Silled agricultural ad fishery worers category, stadig at.004 log poits. Fially, i 2010 at the top of the occupatioal distributio the malefemale wage gap decreased to.083 log poits. Results The first-stage of the Brow et al. (1980) decompositio requires the estimatio of K occupatio specific wage regressios for male ad female worers as defied i Eqs. (1) ad (2), ad a MNL occupatioal attaimet model as defied i Eq. (5). While we

11 The geder wage gap ad occupatioal segregatio Table 2 Log hourly wages by occupatio, meas ad stadard errors Occupatio Me Wome Me Wome Legislators, seior officials ad maagers (0.730) (0.698) (0.655) (0.648) Professioals (0.626) (0.595) (0.629) (0.617) Techicias ad associate professioals (0.600) (0.601) (0.596) (0.595) Clers (0.650) (0.590) (0.550) (0.523) Crafts ad related trades worers (0.573) (0.537) (0.533) (0.718) Service worers ad shop ad maret sales (0.631) (0.628) (0.590) (0.561) Plat ad machie operators ad assemblers (0.488) (0.360) (0.513) (0.389) Elemetary occupatios (0.491) (0.519) (0.463) (0.464) Silled agricultural ad fishery worers (0.539) (0.557) (0.493) (0.484) Observatios 780, , , ,410 Source: Authors elaboratio based o the 2000 ad 2010 Mexica Cesus. Data icludes idividuals betwee 18 ad 65 years of age that wor 30 or more hours per wee. Hourly wages are i atural logarithms ad i real Jue 2010 MXP. Stadard errors are i paretheses. do ot preset the first-stage estimatios due to space costraits, we briefly discuss the mai fidigs. Furthermore, these results are available upo request. The estimatio of ie occupatio specific wage regressios show that there are sigificat differeces i the returs to educatio betwee me ad wome. I 2000, i the Legislators, seior officials ad maagers category, males had higher returs to educatio tha females. I the rest of the distributio, the returs to educatio of males ad females were similar with the exceptio of the Plat ad machie operators ad assemblers category, where the returs for males were substatially larger. By 2010, female returs to educatio surpassed those of males i the top payig category ad were similar amog all worers i the rest of the occupatioal distributio. Based o the previous outcomes, it ca be iferred that geder differeces i returs to schoolig are i large part drive by dissimilarities i the occupatioal structures. Regardig the returs to potetial experiece, i both 2000 ad 2010, wome were better compesated tha me for low experiece levels i the highest payig occupatios. O the other had, i the lower part of the occupatioal distributio, the returs to experiece were

12 62 EcooQuatum Vol. 13. Núm. 1 larger for me. Additioal results are as expected, where beig i formal employmet, married ad worig i a urba locality are positively related with higher hourly wages. Moreover, the state uemploymet rate is egatively correlated with hourly wages, particularly i the highest payig categories. I summary, the results show that there is substatial variatio i the effects of the covariates o male ad female wages depedig o the occupatio of employmet. Attetio ow turs to the MNL occupatioal attaimet model, where the Elemetary occupatios category was used as the referece group. Amog the explaatory variables, educatio plays a importat role i predictig employmet ito the two highest payig occupatios, i.e. the Legislators, seior officials ad maagers ad Professioals categories. The effect of educatio is strogest amog wome ad is sigificat i both 2000 ad A differet story emerges i low silled categories, where a additioal year of schoolig has either a much smaller effect or reduces the probability of attachmet. Regardig potetial experiece, i 2000 this was geerally associated with a icrease i the probability of attachmet at the top of the distributio. Moreover, its effect was larger amog males ad egative for females i low-payig categories. While a similar relatioship is observed amog males i 2010, for females experiece it was positively associated with attachmet i the highestpayig occupatio ad egatively associated with attachmet i other occupatios. O the other had, variables related to the worer s family situatio show that, i 2000, the married populatio had a higher probability of beig employed outside of the Elemetary occupatios category, where the size of this coefficiet was fairly stable across occupatios. I 2010, whereas a comparable relatioship was observed amog wome, o clear patter emerged amog males. Pertaiig to the umber of childre uder age six, i the year 2000 this had a egative effect o the probability of worig i the Legislators, seior officials ad maagers category for all worers, although the coefficiets are ot statistically sigificat. A differet relatioship is observed i 2010, where the coefficiet was positive for both geders at the top of the occupatioal distributio, although it was oly sigificat for wome. Regardig family size, for both periods ad geders, the variable had a egative effect o the probability of attachmet i the highest payig occupatios. Altogether, the effect of the explaatory variables o occupatioal attaimet varies accordig to the geder ad betwee periods. This suggests that me ad wome are allocated differetly betwee occupatios based o their observed characteristics. Brow-Moo-Zoloth decompositio The results of the BMZ decompositio of log wage differetials betwee me ad wome disaggregated by year ad locality size are preseted i Table 3. I 2000, the hourly log wage differetial betwee me ad wome stood at.021 log poits. The sum of the WE ad BE compoets, which equalled log poits, suggests that accordig to their average characteristics wome actually should have had higher hourly wages tha me. Noetheless, aalysig these two terms separately shows substatial heterogeeity i their effects. Specifically, the WE compoet, which rose to.023 log

13 The geder wage gap ad occupatioal segregatio poits, was roughly equal to the total geder wage gap. This implies that the etirety of the geder wage gap was accouted for by differeces i average characteristics. O the other had, the BE compoet stood at log poits idicatig that, o average, wome were allocated i better remuerated positios tha me. With respect to the uexplaied compoets which capture differeces i coefficiets, the sum of the WU ad BU terms was equal to.064 log poits or 304.7% of the geder wage gap. This sigals that based o the returs to their average characteristics, the size of the geder wage gap should have bee larger tha what is i fact observed. Noetheless, whe separatig the two uexplaied compoets sigificatly differet patters emerge. The WU term, equivalet to.167 log poits, suggests that wage differetials are largely drive by vertical segregatio, i.e. by differeces i the returs to average characteristics withi each occupatio. O the other had, the BU compoet which captures the effect of occupatioal segregatio egatively cotributes to the geder wage gap. Somewhat surprisigly, this implies that wome ecoutered fewer barriers to etry ito high payig occupatios tha me. With respect to differeces betwee urba ad rural localities, the geder wage gap varies sigificatly whe maig this distictio, risig to.081 ad.017 log poits i urba ad rural settigs, respectively. I additio, i urba localities the sum of the WE ad BE compoets equalled log poits. O the other had, i rural areas the WU ad BU compoets totalled.101 log poits. Table 3 BMZ decompositio of log hourly wage differetials betwee me ad wome Total Urba Rural Total Urba Rural Total log hourly wage differetial (.001) (.001) (.002) (.001) (.001) (.002) Explaied: Differeces i mea characteristics Withi (.001) (.001) (.001) (.001) (.001) (.001) Betwee (.001) (.001) (.001) (.001) (.001) (.001) Uexplaied: Differeces i coefficiets Withi (.001) (.001) (.002) (.001) (.001) (.002) Betwee (.001) (.001) (.002) (.001) (.001) (.002) Note: Male OLS coefficiets are tae as the o-discrimiatory vector. Source: Authors elaboratio based o the 2000 ad 2010 Mexica Cesus. Stadard errors are i paretheses. Decompositio based o OLS regressios which iclude as idepedet variables years of educatio, potetial experiece, potetial experiece squared, the state uemploymet rate, ad dummy variables for marital status, formal-iformal wor ad regio of residece.

14 64 EcooQuatum Vol. 13. Núm. 1 I 2010, the geder wage gap stood at.026 log poits. With the exceptio of the WE term which equalled log poits, the sig ad overall iterpretatio of the differet compoets of the BMZ decompositio are geerally similar to those observed i The egative sig of the WE compoet suggests that accordig to their average characteristics withi occupatios, female hourly wages should have bee higher tha those of males. Moreover, the BE compoet which stood at log poits deotes that, o average, wome were allocated i better remuerated positios tha me. Relative to 2000, the reductio i absolute value of the BE term sigals that, while i 2010 wome still occupied better compesated positios tha me, malefemale occupatioal structures became more alie durig this period. Cocerig the uexplaied compoets, the sum of the WU ad BU terms totalled.074 log poits or 284.6% of the wage gap. While the WU compoet stood at.132 log poits, its reductio with respect to 2000 implies that the differeces i the returs to average characteristics betwee me ad wome decreased by 21.0% durig the period of study. The declie i the WU term betwee 2000 ad 2010 sigals a positive chage towards a more equal wage compesatio withi occupatios amog me ad wome. O the other had, the cotributio of occupatioal segregatio to the wage gap captured through the BU compoet was agai egative, although smaller i absolute value relative to This suggests that i 2010, wome still ecoutered fewer barriers to etry ito high payig occupatios tha me. Moreover, the statistically sigificat effect of the BU term i 2000 ad 2010 implies that, i both periods, worers were assiged to differet occupatios based o o-productivity related characteristics. Fially, geder wage differetials oce agai differ whe maig the distictio accordig to the size of the locality of residece. Specifically, the wage gap stood at.066 ad.007 log poits i urba ad rural localities, respectively. I urba areas, the sum of the WE ad BE terms equalled log poits, while i rural localities the sum of the WU ad BU compoets totalled.089 log poits. Differeces i the occupatioal structure The observed ad predicted occupatioal distributios of me ad wome across the ie occupatioal categories for 2000 ad 2010 are preseted i Table 4, where the allocatio rule of males, derived from Eq. (5), is used to obtai the female simulated occupatioal structure. This sceario ca be iterpreted as me ad wome havig similar prefereces ad the labour maret providig equal opportuity ad occupatioal access to both groups. Furthermore, the Duca ad Duca (1955) idex of dissimilarity is used i order to provide a better uderstadig of the degree of geder differeces i the occupatioal structures. The idex is give by: (6) D 1 K f = 2 P -P ` j/ = where Eq. (6) is iterpreted as measurig the proportio of male worers required to chage occupatios i order to obtai the same occupatioal distributio geerated by female worers. A idex value equal to zero reflects that there is o occupatioal 1 m

15 The geder wage gap ad occupatioal segregatio dissimilarity betwee both groups. O the other had, a idex value equal to oe idicates that me ad wome are ever i the same occupatios. Table 4 Distributio by occupatio ad dissimilarity idices Occupatio m P f P f WP m P f P WP f Legislators, seior officials ad maagers Professioals Techicias ad associate professioals Clers Crafts ad related trades worers Service worers ad shop ad maret sales Plat ad machie operators ad assemblers Elemetary occupatios Silled agricultural ad fishery worers D D male Source: Authors elaboratio based o the 2000 ad 2010 Mexica Cesus. Predicted distributio by occupatio represets the proportio of female worers who would be i each occupatio if they were to follow the same occupatioal sortig fuctio as me, based o the estimatio of a MNL model of occupatioal attaimet for male worers. With respect to the observed occupatioal distributio, i 2000, at the extremes of the distributio, me had higher participatio rates i both the highest payig occupatio, i.e. i the Legislators, seior officials ad maagers category, ad i the lowest payig oe, i.e. i the Silled agricultural ad fishery worers category. Noetheless, at the middle of the occupatioal distributio, a differet picture emerges. I geeral, wome had higher participatio rates tha me at the top-half of the distributio, whereas at the bottom-half o clear patter is observed. Focusig o the Duca idex of dissimilarity, it shows that, i 2000, approximately 32.1% of wome would have eeded to chage occupatios i order to obtai the same occupatioal distributio as me. O the other had, i 2010, wome had higher participatio rates tha me i the four highest payig occupatios. Furthermore, while the share of female worers employed i the Elemetary occupatios category remaied relatively stable betwee 2000 ad 2010, a much larger proportio of me were employed i this category at the ed of this period. As a result, the Duca idex slightly decreased to.305 i Comparig observed distributios to predicted oes, accordig to the male allocatio rule, female worers were uder-represeted at the Legislators, seior officials ad maagers category i 2000, but ot i I both 2000 ad 2010, wome were uderrepreseted at the bottom of the occupatioal distributio. This result ca be iferred sice uder the male occupatioal sortig fuctio female worers had a much larger

16 66 EcooQuatum Vol. 13. Núm. 1 participatio rate i the Silled agricultural ad fishery worers category compared to their actual share. O the other had, i both years, wome were over-represeted i the Elemetary occupatios category, which is oe of the lowest payig occupatios ad employs a much larger share of male ad female worers. Whe wome follow the same allocatio rule as me, the Duca idex is reduced sigificatly to 7.7% i 2000 ad 5.0% i Joitly, the previous fidigs imply that oly a small part of the differeces i the occupatioal structures betwee me ad wome are drive by differet sortig fuctios. These dissimilarities are liely a product of differeces i productivity related characteristics ad cultural factors which assig differet roles i the labour maret for me ad wome. Sesitivity checs Differet levels of occupatioal aggregatio Because of computatioal problems ad the presece of small cell sizes withi certai occupatios, the majority of studies that employ the BMZ decompositio use a very aggregated level of occupatioal classificatio. 13 Nevertheless, the use of broadly defied categories coceals the importace of occupatios whe performig wage decompositios, sice it liely coceals sigificat wage differeces ad heterogeeity withi the geerally defied occupatios. To examie the sesitivity of our results to the umber of categories, the aalysis is performed uder differet levels of occupatioal aggregatio. Table 5 Sesitivity of BMZ decompositio of log hourly wage differetials betwee me ad wome to differet levels of occupatioal aggregatio Number of occupatios WE BE WU BU WE BE WU BU Oe-digit (.001) (.001) (.001) (.001) (.001) (.001) (.001) (.001) Two-digit (.001) (.001) (.002) (.001) (.001) (.001) (.002) (.001) Note: Oe-digit deotes ie occupatioal categories for 2000 ad Two-digit deotes 18 occupatioal categories for 2000 ad 49 categories for Source: Authors elaboratio based o the 2000 ad 2010 Mexica Cesus. Stadard errors are i paretheses. The results of the BMZ decompositio at the oe-digit ad two-digit occupatioal aggregatio levels are preseted i Table For 2000, augmetig the umber of 13 Amog early studies, the origial aalysis by Brow et al. (1980) for the U.S. used eight categories; Miller (1987) used six occupatios for the Uited Kigdom; Kidd (1993) used six categories for Australia; Kidd ad Shao (1994) used seve occupatios for Caada; ad Brow et al. (1999) used eight categories for Mexico. 14 We origially iteded to iclude the decompositio results at the three-digit level. Noetheless, this greater disaggregatio led to small cell sizes, i.e. less tha 50 observatios, i certai occupatios particularly for

17 The geder wage gap ad occupatioal segregatio categories did ot icrease the proportio of the wage gap attributed to the explaied compoet, i.e. the sum of the WE ad BE terms. O the cotrary, it decreased from log poits at the oe-digit level to at the two-digit level. Moreover, the WE, BE ad WU compoets were margially reduced, while the BU term was subject to a icrease. For 2010, expadig the umber of categories led to a slight icrease i the proportio of the wage gap attributed to the explaied compoet, where the sum of the WE ad BE terms rose from log poits at the oe-digit level to log poits at the two-digit level. Moreover, while the WE compoet grew by.019 log poits, the BE term was reduced by.007 log poits. I geeral, the mai results for the year 2000 are ot very sesitive to the umber of categories, sice augmetig the umber of occupatios does ot icrease the proportio of the geder wage gap attributed to the explaied compoet. O the other had, for 2010 the WE term varies cosiderably betwee the oe ad two-digit levels, suggestig that this compoet is sesitive to the umber of occupatioal categories. Noetheless, the fact that the BU term is always egative ad remais relatively stable provides further evidece that occupatioal segregatio does ot icrease the geder wage gap i Mexico. Cotrollig for self-selectio Sice occupatioal attaimet is determied by the iteractio betwee demad ad supply factors where idividuals may differ amog uobservable characteristics betwee occupatios, the sample of worers observed i each occupatio may ot be radom. I this sectio, we use the iformatio obtaied from Eq. (5) to adjust the occupatio specific wage equatios for potetial effects geerated by selectio bias. Followig Lee (1983), the wage equatios are modified accordigly ad coditioal o occupatio beig chose are give by: (7) w = X Wb + Wi m + p i i i i Zi where W zx 6 ^ W c h@ i = vt, mi =-, ad p i is the error term. FZ ^ iwc h I Eq. (7), z is the stadard ormal probability desity fuctio, v is the stadard error of the disturbace term, ad t is the correlatio betwee the error terms of Eqs. (5) ad (7). The fuctio x is a strictly icreasig trasformatio that coverts the radom variables associated with occupatioal attaimet ito a stadard ormal 1 variat, i.e. x = U - ^Fh, where U is the stadard ormal cumulative distributio fuctio, ad F is the distributio fuctio of the MNL as defied i Eq. (5). wome. As a result, we were forced to either group together occupatios to icrease their cell size or drop the occupatios with small cell sizes from our sample. Neither of these alteratives is ideal sice creatig artificial groupigs may geerate misleadig results, while the elimiatio of low-populated occupatios would mae our results o-comparable to those previously obtaied.

18 68 EcooQuatum Vol. 13. Núm. 1 The procedure is carried out i two stages. First, estimates of the coefficiet vector i the MNL equatio are obtaied from Eq. (5). Secod, these estimated coefficiets are used to calculate m i. The variables icluded i the vector Z i are expected to affect the worer s desire for a particular occupatio as well as the willigess of employers to hire the idividual. The aalysis uses as excluded istrumets that idetify selectivity idicators for the umber of childre below six ad household size. It is assumed that these demographic variables shift the probability of beig employed i occupatio but do ot affect wages. Noetheless, attaiig idetificatio i the selectivity model is challegig give that o available variable ca be cosidered a completely exogeous istrumet. The results of the BMZ methodology cotrollig for self-selectio ito occupatios are preseted i Table I these estimatios, the correctio term is ot cosidered as costitutig part of the explaied or uexplaied compoets ad istead is examied separately. Moreover, the four previously defied terms of the decompositio are grouped ad demarcated as represetig the wage offer gap or ucoditioal wage differetial. This is equivalet to settig the selectivity effects to zero i each equatio, where the wage offer gap is tae to idicate the wage a worer radomly draw from the populatio would receive if she or he were selected ito the occupatioal category i questio (Gyouro ad Tracy, 1988). Amog the results, it ca be see that the selectio term differetial is always positive, implyig that the wage offer gap is smaller tha the total wage gap, ad that geder wage differetials are smaller ad egative after correctig for self-selectio. Moreover, for the full sample i both 2000 ad 2010, the BE term ow stads at ad log poits, respectively. This reiforces the result obtaied i Sectio 4 which idicates that, o average, wome are allocated i better remuerated positios tha me. A similar fidig is observed whe dividig the sample betwee urba ad rural worers. O the other had, whe cotrollig for self-selectio the WU term is o loger the mai driver of the geder wage gap, sice this compoet is ow cosiderably smaller for both years ad across differet samples. Istead, the geder wage gap is mostly a product of differeces i the selectio term. Furthermore, there are cosiderable differeces i the magitude of the WU across time, where this compoet stood at.002 ad log poits i 2000 ad 2010, respectively. These results suggest that whe cotrollig for self-selectio, wome do ot appear to ecouter vertical or hierarchical segregatio i the labour maret. Regardig the BU compoet, it is oce agai observed for the full sample i 2000 ad 2010 that this term egatively cotributed to the total wage gap. This implies that wome ecoutered fewer barriers to etry ito high payig occupatios tha me, where the effect of the BU term is ow bigger i absolute value. Additioally, the WE term wet from positively cotributig to the geder wage gap i 2000 to havig a egative effect i Moreover, the BE compoet icreased from to log poits, whereas the BU term icreased from to log poits The first stage results which iclude the occupatio specific wage regressios corrected for self-selectio are available upo request. 16 I results ot preseted, usig additioal iformatio from a 10.0% radom sample of the 1990 Cesus, the