The effect of motherhood on wages and wage growth: evidence for Australia

Transcription

1 Unversy of Wollongong Research Onlne Faculty of Commerce - Papers (Archve) Faculty of Busness 2010 The effect of motherhood on wages and wage growth: evdence for Australa Tanya Lvermore Reserve Bank of Australa Joan R. Rodgers Unversy of Wollongong, jrrodger@uow.edu.au Peter Smnsk Unversy of Wollongong, smnsk@uow.edu.au Publcaton Detals Lvermore, T. (2010). The effect of motherhood on wages and wage growth: evdence for Australa. 39th Australan Conference of Economsts (pp. 1-26). Sydney: Economc Socety of Australa. Research Onlne s the open access nstutonal reposory for the Unversy of Wollongong. For further nformaton contact the UOW Lbrary: research-pubs@uow.edu.au

2 The effect of motherhood on wages and wage growth: evdence for Australa Abstract Labour market theory provdes several reasons why mothers are lkely to earn lower hourly wages than nonmothers. However, the sze of any motherhood penalty s an emprcal matter and the evdence for Australa s lmed. Ths artcle examnes the effect of motherhood on Australan women s wages and wage growth usng a seres of panel-data models. Based on data from the Household, Income and Labour Dynamcs n Australa survey, an unexplaned motherhood wage penalty of around 5 per cent for one chld, and 9 per cent for two or more chldren, s found. Further analyss suggests that the wage penalty emerges over tme through reduced wage growth, partcularly when the youngest chld s an nfant, rather than through an mmedate wage declne after the brth of a chld. Ths reducton n wage growth s consstent wh dscrmnaton, but also wh a reducton n mothers work effort. Dscplnes Busness Socal and Behavoral Scences Publcaton Detals Lvermore, T. (2010). The effect of motherhood on wages and wage growth: evdence for Australa. 39th Australan Conference of Economsts (pp. 1-26). Sydney: Economc Socety of Australa. Ths conference paper s avalable at Research Onlne:

3 The Effect of Motherhood on Wages and Wage Growth: Evdence for Australa Tanya Lvermore 1 Unversy of Wollongong Prelmnary paper prepared for the Australan Conference of Economsts 2010 ABSTRACT Labour market theory provdes numerous reasons why mothers may earn lower hourly wages than non-mothers; however the emprcal evdence for Australa s lmed. Ths paper examnes the effect of motherhood on Australan women s wages and wage growth usng a seres of panel data models whch account for unobserved factors affectng both wage levels and wage growth. Usng data from the Household, Income and Labour Dynamcs n Australa (HILDA) survey, an unexplaned motherhood wage penalty of around fve percent for one chld, and nne percent for two or more chldren s found. Further analyss suggests that the wage penalty emerges over tme through reduced wage growth, rather than through an mmedate wage declne after brth. Ths reducton n wage growth s consstent wh flatter wage profles of part-tme workers, dscrmnaton and a reducton n mothers work effort. KEY WORDS Motherhood Wage Penalty, Wage Growth, Fxed Effects, HILDA JEL Classfcaton: J31, J13, C33 1 I would lke to express my upmost apprecaton for the advce and assstance provded by Joan Rodgers and Peter Smnsk and thank two thess examners for ther comments. I also gratefully acknowledge fnancal assstance provded by the Unversy of Wollongong where ths work was performed. Ths paper uses un record data from the Household, Income and Labour Dynamcs n Australa (HILDA) Survey. The HILDA Project was nated and s funded by the Australan Government Department of Famles, Housng, Communy Servces and Indgenous Affars (FaHCSIA) and s managed by the Melbourne Instute of Appled Economc and Socal Research (Melbourne Instute). The fndngs and vews reported n ths paper, however, are those of the author and should not be attrbuted to the Unversy of Wollongong, FaHCSIA or the Melbourne Instute. Any remanng errors are my own. Correspondence: Tanya Lvermore, School of Economcs, Unversy of Wollongong, Northfelds Avenue, New South Wales, Australa, Emal: tanyalvermore@gmal.com

4 I. INTRODUCTION A large body of nternatonal lerature has found an unexplaned wage dfferental between mothers and non-mothers (e.g. Anderson, Bnder and Krause, 2002; Bulgescu et al., 2009; Drolet, 2002; Waldfogel, 1995); however the presence of a motherhood wage penalty n Australa s less clear. No study to date has examned the effect of motherhood on Australan women s wages takng account of unobserved factors, and the effect of motherhood on wage growth has not been assessed. Understandng the effect of chldren on women s wages and wage growth however s mportant to current Australan polcy. The polcy optons for counteractng the fscal effects of an ageng populaton centre on ncreasng brth rates whlst mantanng hgh levels of female labour force partcpaton (LFP) (Australan Treasury Department, 2007). However to encourage both outcomes, s mportant to determne how the presence of chldren affects mothers wages, whch n turn affects LFP. The effect of motherhood on wages s also mportant to the study of gender wage equaly. The drect and ndrect effects of chldren are often ced as a cause of the gender wage gap, and thus measurng the motherhood wage dfferental wll shed lght on ths hypothess (Waldfogel, 1998a). Fnally, the opportuny costs born by mothers are of socal mportance; snce good parentng provdes posve externales to the communy, has been argued that mothers dsproportonately share the costs of chldbearng (Budg and England, 2001). Ths s exacerbated when the costs of chldren go beyond the drect costs of food, clothng, health, educaton and shelter and nclude ndrect opportuny costs of wages foregone. Motvated by the mportance of the ssue to Australan polcy and the lack of pror research, ths paper examnes whether mothers, on average, earn lower hourly wages than

5 non-mothers. Fxed effects estmates robust to tme nvarant unobserved factors are obtaned usng an unbalanced panel from the frst seven waves of the Household Income and Labour Dynamcs n Australa (HILDA) survey. The fxed effects results are compared to OLS and Heckman-corrected cross-sectonal models to gauge the extent of heterogeney and selecton bas. Due to the absence of a suable nstrument, the analyss does not account for reverse causaly. Whether the motherhood wage dfferental arses mmedately after brth, or develops over tme through wage growth s also nvestgated. The frst-dfference and fxed effects models of Loughran and Zssmopoulos (2009), whch account for the effect of unobserved heterogeney on both the wage level and wage growth, are appled to the Australa data. Whle nsghtful, the methodology s sensve to outlyng wage growth observatons, and s lkely to be more useful as addonal waves of HILDA are avalable. The paper s structured as follows. Secton II revews the emprcal lerature and dscusses the theoretcal reasons for a motherhood wage penalty. Secton III presents the emprcal approach adopted n ths paper and Secton IV descrbes the HILDA data and sample. Fnally, Secton V presents the emprcal results and Secton VI concludes. II. EXISTING EVIDENCE AND THEORETICAL BACKGROUND A large body of nternatonal lerature has studed whether mothers earn lower wages than non-mothers. A sgnfcant wage penalty has been found n the Uned States 2, Bran (Waldfogel, 1995; 1998b), Canada (Drolet, 2002) and Germany (Bulgescu et al., 2009). 3 Whle other studes n Denmark (Gupta and Smh, 2002) and Sweden (Albrecht et al., 1999) have found motherhood or maternal leave to have no drect effect on wages. 2 Such as: Anderson, Bnder and Krause (2002; 2003), Baum (2002), Budg and England (2001), Loughran and Zssmopoulos (2009), Lundberg and Rose (2000), Tanguch (1999) and Waldfogel (1997; 1998b). 3 Most studes whch found a motherhood wage penalty control for maral status n ther analyss (for example, Anderson, Bnder and Krause, 2002; 2003; Baum, 2002; Budg and England, 2001) whch demonstrates that motherhood has a negatve effect on wages, ndependent of the effects of marrage.

6 Although the effects of chldren on wages may work ndrectly through characterstcs such as experence, tenure, educaton and job choces, most studes focus on estmatng the drect effect of chldren. To that end, the resdual motherhood penalty s typcally estmated wh human capal, job characterstcs and tme nvarant unobserved heterogeney held constant. In studes whch have found a resdual penalty, estmates range from two percent (Baum, 2002; Loughran and Zssmopoulos, 2009) to nne percent (Waldfogel, 1995; 1998b) for one chld and a further two (Anderson, Bnder and Krause, 2002; 2003) to nne percent (Budg and England, 2001) for addonal chldren. In contrast to the large nternatonal lerature, the effect of motherhood on wages or wage growth has receved ltle attenton n Australa. In fact, only two Australan studes (Krepp, 2007; Whehouse, 2002) have sought to examne whether mothers earn lower hourly wages than non-mothers, and no Australan study to date has examned the effect of motherhood on wage growth. Usng cross-sectonal data from the Australan Workplace Industral Relatons Survey n 1995, Whehouse (2002) found no sgnfcant effect of dependant chldren on Australan women s wages. Smlarly, Krepp (2007) found no resdual wage penalty usng a Heckman-corrected cross-sectonal model to account for selecton nto employment. Although these Australan studes have found no drect effect of motherhood on wages, ther results are potentally based by unobserved heterogeney and the omsson of key controls. 4 Economc theory provdes a number of non-mutually exclusve reasons why a raw wage dfferental between mothers and non-mothers may exst. Fore mostly, the greater frequency 4 In Whehouse s model, the coeffcent of motherhood may be based upward by the omsson of experence f mothers n the sample are older than non-mothers, as the two categorcal age varables ncluded may not successfully control for the effect of agng on experence. Smlarly, Krepp s models do not control for parttme employment status; f mothers are more lkely to work part-tme jobs, and part-tme work attracts a pay premum (Booth and Wood, 2008), the motherhood coeffcent may be based towards zero. Although Krepp accounts for selecton nto employment, her estmates may be based by other unobserved factors correlated wh both motherhood and wages (Anderson, Bnder and Krause, 2002; 2003; Korenman and Neumark, 1992).

7 and duraton of work nterruptons among mothers may reduce wages through foregone experence, deprecaton of sklls and loss of tenure. 5 The antcpaton of work nterruptons may also slow wage growth f nduces lower nvestment n human capal (Ben-Porath, 1967). 6 The propensy of mothers to seek employment offerng flexbly, part-tme hours, materny leave entlements or lmed travel tme may also result n lower wages as a compensatng dfferental (Rosen, 1986). Part-tme work explans a large porton of the motherhood wage penalty n the Uned States (Waldfogel, 1997) and Bran (Josh, Pac and Waldfogel, 1999), however Australan studes of part-tme wages (Booth and Wood, 2008; Rodgers, 2004) suggest that part-tme work may not drectly cause (or ncrease) a motherhood wage penalty. Nevertheless, workng part-tme may stll affect Australan mothers wages n the long run through lower human capal accumulaton (due to less tme spent workng) and fewer opportunes for promoton (Abhayaratna et al., 2008). The effort requred to rase chldren may also reduce mothers wages followng Becker s (1985) worker effort hypothess (Budg and England, 2001). However even between equally productve mothers and non-mothers a wage dfferental may arse through statstcal dscrmnaton f employers assume chldcare and housework responsbles make mothers less productve (Hyclak, Johnes and Thornton, 2005, p.384). It s also possble that employers wh a prejudce aganst mothers wll pay a wage below productvy by an amount suffcent to compensate for ther taste for dscrmnaton (Becker, 1957). Even n the absence of any drect or ndrect effect of chldren on wages, a motherhood wage dfferental may be observed n OLS models f there are unobserved factors, such as work motvaton, whch are negatvely correlated wh the desre for chldren, and posvely 5 See Anderson, Bnder and Krause (2002; 2003), Baum (2002), Budg and England (2001), Hll (1979), Josh, Pac and Waldfogel (1999), Lundberg and Rose (2000), and Waldfogel (1995; 1997). 6 Ths effect s accentuated for mothers who expect to work part-tme hours followng the brth of a chld.

8 correlated wh wages. On the other hand, the effect of motherhood on wages wll be underestmated f those women most lkely to experence a motherhood wage penalty are also the least lkely to be employed. To account for these sources of bas, nternatonal studes have typcally used panel data to estmate fxed effects models (e.g. Anderson, Bnder and Krause, 2002; 2003; Budg and England, 2001; Lundberg and Rose, 2000). A few of studes have also addressed reverse causaly between motherhood and wages wh nstrumental varables methods. Ths paper does not account for reverse causaly as potental nstruments are eher not avalable n Australa or not possble gven the sample sze. 7 The drecton of possble reverse causaly bas s not entrely clear. The motherhood penalty may be overstated f those women wh lower wages, and hence a lower opportuny cost of leavng the workforce, are more lkely to become mothers (Lundberg and Rose, 2000, p.692). Alternatvely, an ncrease n the wage rate may have a posve ncome effect on havng a chld (Arza and Ugdos, 2007). III. EMPIRICAL APPROACH Women s wages are assumed to be determned accordng to the human capal model: lnw Chld 1 Chld HC JC MS Year u t (Equaton 1) where ln s the natural logarhm of the hourly wage (n 2007 dollars) of woman W ( 1,2,..., N ) n year t ( t 1 2,,...,T) ; Chld1 and Chld2 are dummy varables equal to one f the woman has one or more chldren (and has a vald wage rate), and two or 7 Instruments used n the motherhood penalty lerature nclude father s and mother s educaton, whether the woman lved wh her parents n adolescence (Amuedo-Dorantes and Kmmel, 2005; Neumark and Korenman, 1994), the parent s educatonal goals for the woman, whether the woman s mother worked at age 14, the number of sblngs (Neumark and Korenman, 1994; Smonsen and Skpper, 2006), the woman s past attudes and her mother s famly buldng hstory (Josh, Pac and Waldfogel, 1999). Instruments used n studes of motherhood and other labour market outcomes nclude state and county ndcators of the cost of fertly and fertly control and laws on pregnancy termnaton (Klepnger, Lundberg and Plotnck, 1999), mscarrage (Hotz, Wllams McElroy and Sanders, 2005), sex-mx of the frst two chldren (Angrst and Evans, 1998) and twn brths (Jacobsen, Wshart Pearce and Rosenbloom, 1999).

9 more chldren respectvely; HC s a vector of human capal varables, namely, work experence, experence squared, educaton and tenure wh the current employer; JC s a vector of job characterstcs, namely, part-tme and casual employment status, ndustry, occupaton, sector, frm sze and unon membershp; MS s a vector of dummy varables representng maral status, namely, partnered and separated; Year t s a vector of year dummes; s an ndvdual specfc ntercept; and u s a random error term. To remove the ndvdual specfc ntercepts, and hence control for tme nvarant unobserved characterstcs, the varables n Equaton 1 are demeaned, obtanng fxed effects estmates of the motherhood wage dfferental. A sgnfcantly negatve (posve) 1 ndcates that there s a resdual motherhood wage penalty (premum). Snce Chld1 and Chld2 are equal to one f the woman has one or more chldren and two or more chldren respectvely, the coeffcent 2 gves the ncremental effect of a second chld. 8 To understand whether the wage dfferental arses mmedately after brth or develops over tme through wage growth, the methodology of Loughran and Zssmopoulos (2009, pp ) s employed. Modfyng Equaton 1 we obtan: lnw b Chld 1 1 b2chld2 b3ychld1 b4ychld 2 HC JC MS Year Gap Exp Exp u t 5 2 (Equaton 2) where HC represents educaton and tenure, wh experence ( Exp ) and the quadratc n experence ( Exp ) ncluded separately. In addon to 2 Chld 1 and Chld 2, two 8 To test for selecton bas a Heckman selecton model (Cameron and Trved, 2009, p ) s estmated wh the pooled sample usng maxmum lkelhood estmaton (see Table A.1, Appendx B for results). Nonlabour ncome s used as an excluson restrcton followng other studes (such as Amuedo-Dorantes and Kmmel, 2005; Baum, 2002; Booth and Wood, 2008; Josh, Pac and Waldfogel, 1999; Korenman and Neumark, 1992; Krepp, 2007). Snce self-employed women are excluded from the sample, the selecton correcton accounts for selecton nto beng an employee rather than employment n general.

10 varables YChld 1 and YChld 2 are ncluded whch count the number of years snce returnng to the workforce after gvng brth to the frst and second chld respectvely (equal to one n the frst year of return to work after brth). 9 Moreover, Gap, whch measures the number of years not n the labour force around the frst and second brths, s ncluded to capture the effect of human capal deprecaton. 10 In ths specfcaton, motherhood s allowed to affect both wage levels and wage growth. The mmedate effect on the wage level n the frst year of work after gvng brth s gven by 1st b b Gap for the frst chld and the ncremental effect of a second chld s nd gven by b b Gap. The effect of the brth on subsequent annual wage growth s gven by b 3 for a frst chld and b 4 gves the ncremental effect of addonal chldren. As well as allowng the wage equaton to have separate ntercepts ( ) the above model also allows dfferent slopes n experence ( ) for each ndvdual, enablng ndvdual specfc unobserved factors to affect both wage levels and wage growth. By takng frst-dfferences of each varable and assumng that experence ncreases by one every year, we obtan: 11 9 Instead of constructng frst-dfferences of YChld1 and YChld2 and dvdng by the number of years between ntervews, ths paper follows Loughran and Zssmopoulos (2009) n usng the conceptually equvalent Chld1 and Chld2 varables nstead. Chld1 and Chld2 are the more relable measure as the nformaton requred to construct YChld1 and YChld2 s not avalable n the general release HILDA data. 10 For women who gave brth and had an employment break, Gap s set equal to the number of waves not workng, and remans at ths value n the years thereafter. The coeffcent 5 s constraned to be the same for frst and second chldren. 11 In the panel used by Loughran and Zssmopolous, ndvduals are not observed every year so experence does not necessarly ncrease by one between observatons. To account for ths, Loughran and Zssmopoulos re-ntroduce the square of experence n Equatons 3 and 4 and dvde the change n the log wage, experence and experence squared by the tme elapsed between the respondent s ntervews. Instead of dvergng from the specfcatons, the sample and experence measure used n ths paper to estmate Equatons 3 and 4 wll be constructed n such a way to ensure that experence ncreases by one between wage observatons. In dong so, s not necessary to nclude any addonal varables n Equatons 3 or 4, or to dvde any varables by the tme between ntervews.

11 lnw b 1 Chld1 b2 Chld2 b3chld1 b4chld2 1 HC 2 JC 3 MS 4 Yeart 5 Gap Exp u (Equaton 3) where 2. The tme nvarant has been removed, however the effect of unobserved heterogeney on wage growth,, remans. Ths allows for the possbly that wages have dfferent growth rates over tme for dfferent ndvduals. If unobserved productvy tras such as career motvaton are posvely correlated wh wage growth and also nduce women to delay or forego chldbearng, coeffcents b 1 4 wll be negatvely based. Smlarly, f there are unobserved factors whch ncrease wage growth and nduce women to become mothers b 1 4 wll be posvely based. Demeanng Equaton 3 elmnates, obtanng a frst-dfference wh fxed effects model: lnw lnw b 1 b Chld 1 Chld1 b2 Chld 2 Chld 2 Chld 1 Chld1 b4 Chld 2 Chld 2 HC HC JC JC MS MS Year Year Gap Gap Exp Exp u u t t (Equaton 4) Estmaton results wll be presented for Equaton 1 (as a pooled cross-secton and wh fxed effects), Equaton 3 (frst-dfference) and Equaton 4 (frst-dfference wh fxed effects). IV. DATA The wage equatons are estmated usng un-record data from the frst seven waves ( ) of the Household, Income and Labour Dynamcs n Australa (HILDA) survey. 12 Over the seven waves of HILDA, a total of 9,792 women were surveyed resultng n 47,727 woman-year observatons Table A1 defnes the varables used. 13 For more nformaton on the HILDA survey see Wooden and Watson (2007).

12 Sample A: TABLE 1 Sample Constructon: Remanng Woman-Year Observatons Women Aged between 21 and Employed Employees (not self-employed) Not studyng full tme Not mssng wage data Not an outlyng wage Not nconsstent chld data No deceased chldren Not mssng human capal data Not mssng job characterstc data Not mssng maral status Sample B: Part of a vald block encompassng 3 or more wages Droppng the frst observaton from each block n makng frst-dfferences Rato of consecutve wages no more than To estmate the effect of motherhood on wages (Equaton 1) the sample s restrcted to women between the ages of 21 and Snce each woman-year observaton must have a vald wage rate, 7,320 observatons n whch a woman was unemployed or out of the labour force were dropped. Other observatons were excluded f the woman was self-employed, n full-tme educaton, or had mssng or n-consstent data. 15 Furthermore, outlyng wage 14 Women aged 21 years or less were excluded as wage rates of young workers are generally determned by junor pay rate scales and as such, do not reflect ndvdual worker productvy. A further 2,549 women older than the usual chldbearng age-range were also dropped. Inspecton of the HILDA data shows that pregnancy rates fall greatly beyond the age of 45. To allow a woman aged close to 45 to have a chld n the frst of HILDA and be ncluded n the sample n the waves that follow, only women aged 52 years or older were excluded. 15 The self-employed were excluded as ther wages are not determned n the same way as employees. Fulltme students were excluded as ther decsons about occupatons and wage contracts are lkely to dffer from ndvdual s no longer n full-tme educaton. Observatons were excluded f the total number of chldren the woman currently has exceeded the total number of chldren ever gven brth to or adopted. Ths should not occur as foster and step chldren are not counted as chldren currently have n HILDA. Women wh deceased chldren were also dropped as they may have wage effects stemmng from chldbearng and rearng n the past and therefore dffer to women who have never had chldren. Observatons wh mssng data on salary or hours worked, human capal, job characterstcs or maral status were excluded.

13 observatons n the top or bottom 0.5 percent of the remanng sample were excluded. 16 After applyng all restrctons, a sample of 4,255 women (2,596 wh one or more chldren, 1,998 wh two or more chldren) makng 14,972 woman-year observatons was obtaned (Sample A). Descrptve statstcs for Sample A are provded n Table 2. To estmate Equatons 3 and 4, the change n each varable between the years a woman s workng must be constructed. To ensure that experence changes by (around) one year between consecutve wage observatons (as requred to obtan Equaton 3) frst-dfferences are not taken over non-respondng years. Instead, blocks of consecutve years are dentfed where a woman was workng and had a vald wage observaton or was not employed. 17 To be ncluded n the sample, each woman must have a vald block of respondng years encompassng at least three wage observatons to allow frst-dfferences to be constructed, and the frst-dfference wh fxed effects model to be estmated. 18 Excludng outlyng wage growth observatons where the rato of consecutve wages exceeded a factor of 1.5, a fnal sample of 2,247 women makng 7,989 woman-year observatons remaned (1,460 wh one or more chldren, 1,137 wh two or more chldren). In ths sample, a total of 132 frst brths and 96 second brths were observed (Sample B) Testng revealed the results are generally nsensve to the wage cut-off chosen, and changng the wage cutoff has ltle effect on the conclusons drawn. 17 Ths method should ensure that for the women n the sample, experence does not ncrease n the years no wage s observed. 18 To make full use of the data, addonal second blocks were dentfed. 82 women have 2 vald blocks (Waves 1-3 and Waves 5-7). Frst-dfferences are only taken whn each block, that s, not across waves 5 and That s, any wage whch ncreased by more than 50 percent or decreased by more than one thrd was excluded. Ths corresponds to a change n the natural log of the hourly wage of As dscussed n Secton VI, the results are sensve to outlyng wage growth observatons.

14 TABLE 2 Descrptve Statstcs for Sample A Mother Non- Non- Mother Mother Mother Hourly wage Industry Contnued Log-hourly wage Constructon Age (years) *** Retal/Hospaly Experence (years) *** Transport Tenure (years) *** Culture *** Educaton Fnance/Scence *** Post-Graduate *** Educaton/Health *** Bachelor Degree *** Other (Omted) Dploma *** Sector Certfcate *** Prvate Sector *** Year *** Publc Sector *** Year 11 (Omted) *** Other (Omted) Occupaton Frm Sze Manager Small ** Professonal *** Medum (Omted) Trade ** Large Communy *** Other Job Characterstcs Clercal Part-tme Worker *** Sales Casual Worker *** Machnery * Unon member *** Labourer (Omted) *** Maral Status Industry Partnered *** Prmary ** Separated *** Utles/Mnng *** Never Marred *** Manufacturng (Omted) Source: HILDA, Waves 1-7 Notes: *,**,** means mothers and non-mothers are statstcally dfferent at the 5%, 1% and 0.1% levels of sgnfcance. Means are weghted by cross-sectonal probably weghts V. EMPIRICAL RESULTS The pooled cross-sectonal model reveals no sgnfcant raw wage dfferental between mothers and non-mothers (Table 3). 20 However, as mothers n the sample are older, and consequently have more years of experence than non-mothers, a sgnfcant eght percent penalty arses when controllng for experence. The penalty becomes small (around two percent) and nsgnfcant when controllng for educaton and tenure. There contnues to be no sgnfcant dfference between mothers and non-mothers wages untl controls are 20 All models presented are un-weghted. There are no approprate probably weghts avalable for an unbalanced panel (usng fxed effects and frst dfference estmaton). Applyng cross-sectonal respondng person weghts to the pooled cross-sectonal model has ltle effect on the estmates (results avalable on request).

15 ncluded for maral status. In the fnal model, there remans a sgnfcant resdual wage penalty of 2.7 percent for one chld, and no ncremental penalty for addonal chldren. A Heckman-corrected wage model (Table A.1, Appendx B) shows sgnfcant evdence of selecton bas, wh the Heckman-corrected motherhood penalty estmate for one chld larger (by around 2.6 percentage ponts) than the pooled OLS results. Ths suggests that mothers most lkely to suffer a wage penalty are less lkely to be employed, and therefore falure to account for selecton nto employment wll understate the true motherhood wage penalty. The fxed effects estmates show no sgnfcant motherhood penalty controllng for human capal varables. However when part-tme and casual status are added, a sgnfcant fve percent penalty for one chld, and four percent penalty for two or more chldren (sgnfcant at the ten percent level) appears, reflectng the large premum to parttme employment for Australan women (Table A.1, Appendx B). 21 Ths contrasts results for Bran and the Uned States whch found part-tme work to be a source of the motherhood penalty (Josh, Pac and Waldfogel, 1999, Waldfogel, 1997). Includng other job characterstcs and maral status has an mmateral effect on the motherhood coeffcent. Wh all controls ncluded, the fxed effects results show mothers wh one chld receve a fve percent penalty whle mothers of two or more chldren earn around nne percent less than non-mothers on average, even after controllng for observable and tme-nvarant unobservable dfferences. 22 These penaltes are smlar to the unweghted Heckman-corrected pooled estmates provdng some evdence that selecton nto employment may be accounted for by fxed effects. 21 A part-tme wage premum s n lne wh the fndngs of Booth and Wood (2008) 22 Fxed effects models were estmated allowng for nteracton between motherhood and professonal occupaton, partnered and part-tme status (results not shown). None of the nteracton terms however were sgnfcant suggestng that the motherhood penalty s not statstcally dfferent across maral, part-tme work and professonal occupaton status.

16 N = ( 4242) No controls Experence Only TABLE 3 Pooled Cross-Secton OLS +Educaton, tenure +Part-tme, Casual Status +Industry, Occupaton +Sector, Frm Sze, Unon Status +Maral Status (All Controls) Chld *** * (0.016) (0.016) (0.014) (0.015) (0.013) (0.013) (0.013) Chld (0.016) (0.016) (0.014) (0.014) (0.012) (0.012) (0.012) R F *** *** *** *** *** *** *** N = ( 4242) Fxed Effects Chld * * * * (0.020) (0.020) (0.020) (0.021) (0.020) (0.020) (0.020) Chld (0.024) (0.024) (0.024) (0.024) (0.024) (0.024) (0.024) R F *** *** *** *** *** *** *** Source: HILDA, Waves 1-7 Notes: Robust clustered standard errors are used; standard errors are n parentheses, * p < 0.05, ** p < 0.01, *** p < The dependant varable s the natural log of the hourly wage n constant 2007 dollars. All regressons correspond to Equaton 1. Chld1 s the estmated effect of motherhood on log wages of the frst chld;chld2 gves the ncremental effect of a second chld. Regressons are not weghted. Full results for fnal model reported n Table A.1, Appendx B.

17 Ths mples that mothers do not dffer from non-mothers on unobservable productvy tras, other than those whch affect selecton nto employment. The resdual wage penalty s consstent wh a tred mother effect whereby mothers have less energy to devote to work. Alternatvely, motherhood may have no causal effect on productvy, yet a wage penalty arses through taste dscrmnaton or an (naccurate) percepton by employers that mothers are less productve. The penalty estmates are smlar to some observed for the Uned States (Lundberg and Rose, 2000; Waldfogel, 1997; Waldfogel, 1998b) however are larger than estmates for Denmark (Gupta and Smh, 2002) and other Amercan results (Anderson, Bnder and Krause, 2002; 2003; Budg and England, 2001), and smaller than results for Bran (Waldfogel, 1995; 1998b). 23 To determne whether the wage dfferental arses mmedately after the brth of a chld, or reflects the effect of motherhood on wage growth, Equatons 3 and 4 have been estmated. The frst-dfference results reveal that havng a frst or second chld has no mmedate effect on the wage level (Table 4). Across all sets of control varables, the average effect of havng a chld n the frst year of return to work s close to zero and not statstcally sgnfcant. Contrary to pror expectatons of a negatve effect, the coeffcent of Gap s not statstcally dfferent from zero. In contrast, the coeffcent of YChld1 shows that havng a frst chld sgnfcantly reduces wage growth. After the brth of a frst chld, annual wage growth s reduced by one percentage pont (p<0.05). The coeffcent of YChld2 shows an offsettng posve effect of a second chld on wage growth, however s not sgnfcant at conventonal 23 Ths result stands n contrast to pror Australan studes whch found no sgnfcant resdual motherhood wage penalty (Krepp, 2007; Whehouse, 2002). The dfference does not appear to be due to unobserved heterogeney as a sgnfcant motherhood wage penalty was also found n the pooled cross-sectonal model. As such, the contrastng results are lkely to be due to dfferences n the way mothers are defned, sample composon, the tme perods covered and the controls ncluded n the models. 15

18 levels. Altogether, the frst-dfference model suggests that the wage dfferental between mothers and non-mothers has come about through a reducton n wage growth, rather than an mmedate fall n the wage level after brth. Loughran and Zssmopoulos (2009) found the oppose result for the Uned States fndng motherhood reduces wages by two percent n the year of brth, and has no sgnfcant effect on subsequent wage growth The frst dfference wh fxed effects model was estmated to account for the effect of unobserved heterogeney on wage growth (Table 5). The mmedate effects of motherhood on the wage level reman generally the same. In contrast, allowng for ndvdual slopes n experence has a large mpact on the effect of motherhood on wage growth. The pont estmates show havng a frst chld reduces wages by 3.5 percent per year and havng a second chld ncreases annual wage growth by around 4.5 percentage ponts. However, the standard errors are much larger n Equaton 4 renderng the effect of a frst chld nsgnfcant, and the effect of a second chld only sgnfcant at the ten percent level. The dssmlary of the Equaton 3 and 4 pont estmates may be due to the effect of unobserved heterogeney on wage growth, whch s only controlled for n the latter model. A comparson of the coeffcents suggests that those women wh wage-growth enhancng unobserved tras are more lkely to have a frst chld, but less lkely to have a second. However, snce these effects are only margnally sgnfcant at best, there may be no systematc effect on wage growth when unobserved factors are taken nto account. Further analyss reveals that whle the frst-dfference model (Equaton 3) s relatvely nsensve to the ncluson of large wage-growth observatons, the frstdfference wh fxed effects model (Equaton 4) shows very large and opposng mpacts 16

19 N = 7989 (2247) No controls Experence Only TABLE 4 Frst Dfference +Gap 17 +Educaton, tenure +Part-tme, Casual Status +Industry, Occupaton +Sector, Frm Sze, Unon Status +Maral Status (All Controls) Chld1 ( b 1 ) * (0.018) (0.018) (0.019) (0.019) (0.018) (0.018) (0.018) (0.018) Chld2 ( b 2 ) (0.019) (0.019) (0.020) (0.020) (0.020) (0.020) (0.020) (0.020) Gap ( 5 ) (0.023) (0.023) (0.023) (0.023) (0.023) (0.023) Effect of Motherhood on Wage Growth YChld1 ( b 3 ) ** * * * * * (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) YChld2 ( b 4 ) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) Immedate Effect of Motherhood: 1st Frst Chld: b1 b3 5 Gap (0.017) (0.018) (0.019) (0.019) (0.018) (0.018) (0.018) (0.020) 2nd Second Chld: b2 b4 5 Gap (0.019) (0.019) (0.019) (0.020) (0.020) (0.020) (0.020) (0.020) R F * ** ** ** *** *** *** *** Source: HILDA, Waves 1-7 Notes: Robust clustered standard errors are used; standard errors are n parentheses, * p < 0.05, ** p < 0.01, *** p < The dependant varable, the natural log of the hourly wage, and all regressors, are n frst-dfference form as specfed n Equaton 3. Full results for fnal model are reported n Table A.1, Appendx B. 1st b b Gap gves the estmated average effect of motherhood on log wages n the year of returnng to work after the brth of the frst chld where st Gap s equal to 0.63; b3 gves the estmated effect of a frst chld on subsequent annual wage growth. effect of a second chld on wage levels and wage growth where for a frst and second chld are b1 b3 and b2 b4 respectvely. 2nd Gap 2nd b b Gap and b 4 gve the ncremental s equal to In regressons where Gap s not controlled for, the mmedate effects

20 N = 7989 (2247) No controls TABLE 5 Frst Dfference wh Fxed Effects Experence Only +Gap 18 +Educaton, tenure +Part-tme, Casual Status +Industry, Occupaton +Sector, Frm Sze, Unon Status +Maral Status (All Controls) Chld1 ( b 1 ) * * (0.026) (0.026) (0.028) (0.028) (0.026) (0.026) (0.026) (0.026) Chld2 ( b 2 ) (0.026) (0.026) (0.027) (0.027) (0.027) (0.027) (0.027) (0.027) Gap ( 5 ) (0.026) (0.026) (0.026) (0.026) (0.026) (0.026) Effect of Motherhood on Wage Growth YChld1 ( b 3 ) (0.022) (0.023) (0.023) (0.023) (0.023) (0.023) (0.023) (0.023) YChld2 ( b 4 ) (0.026) (0.027) (0.026) (0.027) (0.026) (0.027) (0.027) (0.027) Immedate Effect of Motherhood 1st Frst Chld: b1 b3 5 Gap (0.015) (0.023) (0.025) (0.025) (0.025) (0.025) (0.025) (0.025) Second Chld: b b Gap 2nd (0.027) (0.027) (0.028) (0.028) (0.028) (0.028) (0.030) (0.028) R F *** *** *** *** Source: HILDA, Waves 1-7 Notes: Robust clustered standard errors are used; standard errors are n parentheses, * p < 0.05, ** p < 0.01, *** p < The dependant varable, the natural log of the hourly wage, and all regressors, are n frst-dfference form as specfed n Equaton 4. Full results for fnal model are reported n Table A.1, Appendx B. 1st b b Gap gves the estmated average effect of motherhood on log wages n the year of returnng to work after the brth of the frst chld where st Gap s equal to 0.63; b3 gves the estmated effect of a frst chld on subsequent annual wage growth. effect of a second chld on wage levels and wage growth where for a frst and second chld are b1 b3 and b2 b4 respectvely. 2nd Gap 2nd b b Gap and b 4 gve the ncremental s equal to In regressons where Gap s not controlled for, the mmedate effects

21 of a frst and second chld on wage growth, whch are nconsstent wh the motherhood penaltes presented n Table 2 (Table A.2, Appendx B). In estmatng the latter model for the Uned States wh the Natonal Longudnal Survey (NLS) data sets, Loughran and Zssmopoulos (2009) reported no such sensvy. It appears that the frstdfference wh fxed effects specfcaton s too demandng for seven waves of HILDA data used and requres a longer panel to obtan precse results. The sample szes reported n Table A.2 show that a large porton of women have large year-to-year wage changes. The source of these large changes s unclear, however gven the sensvy of the results, they are presumably not explaned by the varables ncluded n the model. VI. SUMMARY AND CONCLUSIONS Holdng observable and unobservable dfferences constant, fxed effects estmates reveal mothers wh one chld earn around fve percent less than non-mothers on average, wh a further four percent penalty for a second chld. Ths resdual wage penalty may be due to actual productvy dfferences f responsbly for chldren leaves mothers wh less energy to exert at work. Alternatvely, mothers may be pad less than non-mothers because employers perceve mothers to be less productve or have a taste for dscrmnatng aganst them. Contrary to studes overseas, part-tme employment was not found to be a source of the motherhood wage penalty. Smlarly, the propensy of mothers to work n certan occupatons, ndustres and sectors explans ltle of the gap. The frst-dfference results (Equaton 3) show a frst or second chld has no mmedate effect on wages n the frst year of returnng to work, but a frst chld reduces subsequent annual wage growth by one percentage pont. Sensvy tests showed these 19

22 results are relatvely robust to large wage changes. The frst dfference wh fxedeffects (Equaton 4) results however proved much more sensve to outlers. Overall, the wage growth analyss suggests that the motherhood penalty emerges through reduced wage growth, rather than an mmedate wage declne after brth. Ths s consstent wh Australan materny leave legslaton whch entles most women to return to ther pror poson wh ther pre-brth employer and thus wages may be unaffected n the short-term. However subsequent wage growth may be reduced f the presence of chldren reduces mothers actual or perceved productvy, or f employers have a taste for dscrmnaton. Moreover, a reducton n wage growth s also consstent wh flatter wage profles for part-tme workers. As such, polces whch make combnng work and motherhood easer are lkely to mprove Australan mothers wages and wage growth relatve to non-mothers. Greater access to chldcare servces, partcularly n the workplace, may mprove women s ably, and employer s percepton of ther ably, to balance famly and work. Moreover, practces whch mprove career advancement of part-tme workers are lkely to ncrease mothers wage growth. Future research may estmate the motherhood penalty usng nstrumental varables technques; as more waves of data become avalable n HILDA, nstruments used overseas such as sblng sex-mx may be feasble. A longer panel may also render the frst-dfference wh fxed effects specfcaton more robust. Gven the sensvy of the wage growth results and the large proporton of women wh large year-to-year wage changes, further nvestgaton nto the source of outlyng wage trajectores s warranted. 20

23 REFERENCES Abhayaratna, J., Andrews, L., Nuch, H. and Podbury, T. (2008). Part Tme Employment: The Australan Experence. Staff Workng Paper, Productvy Commsson, Canberra. Albrecht, J. W., Edn, P., Sundstrom, M. and Vroman, S. (1999). Career Interuptons and Subsequent Earnngs: A Reexamnaton Usng Swedsh Data. Journal of Human Resources. 34(2): Amuedo-Dorantes, C. and Kmmel, J. (2005). The Motherhood Wage Gap for Women n the Uned States: The Importance of College and Fertly Delay. Revew of Economcs of the Household. 3: Anderson, D. J., Bnder, M. and Krause, K. (2002). The Motherhood Wage Penalty: Whch Mothers Pay It and Why? Amercan Economc Revew. 92(2): Anderson, D. J., Bnder, M. and Krause, K. (2003). The Motherhood Wage Penalty Revsed: Experence, Heterogeney, Work Effort, and Work-Schedule Flexbly. Industral and Labor Relatons Revew. 56(2): Angrst, J. D. and Evans, W. N. (1998). Chldren and Ther Parents' Labour Supply: Evdence from Exogenous Varaton n Famly Sze. The Amercan Economc Revew. 88(3): Arza, A. and Ugdos, A. (2007). Entry nto Motherhood: The Effect of Wages. Unversy of Basque Country, DFAE-II Wokrng Paper Seres. Australan Treasury Department (2007). Intergeneratonal Report. Canberra. Baum, C. L. (2002). The Effect of Work Interruptons on Women's Wages. Labour. 16(1): Becker, G. S. (1957). The Economcs of Descrmnaton. Chcago, Unversy of Chcago Press. Ben-Porath, Y. (1967). The Producton of Human Capal and the Lfe Cycle of Earnngs. Journal of Polcal Economy. 75(4): Booth, A. L. and Wood, M. (2008). Back-to-Front Down Under? Part-Tme/Full-Tme Wage Dfferentals n Australa. Industral Relatons. 47(1): Budg and England (2001). The Wage Penalty for Motherhood. Amercan Socologcal Revew. 66(Aprl): Bulgescu, B., de Crombrugghe, D., Mentesoglu, G. and Montzaan, R. (2009). Panel Estmates of the Wage Penalty for Maternal Leave. Oxford Ecnomc Papers. 61(Supplement 1 (Aprl)): Cameron, A. C. and Trved, P. K. (2009). Mcroeconometrcs Usng Stata. College Staton, Stata Press. Drolet, M. (2002). Wves, Mothers and Wages: Does Tmng Matter? Statstcs Canada, Workng Paper, No.186. Gupta, N. D. and Smh, N. (2002). Chldren and Career Interruptons: The Famly Gap n Denmark. Economca. 69(276): Hll, M. S. (1979). The Wage Effects of Maral Status and Chldren. Journal of Human Resources. 14(4): Hotz, V. J., Wllams McElroy, S. and Sanders, S. G. (2005). Teenage Chldbearng and s Lfe Cycle Consequences. The Journal of Human Resources. 40(3): Hyclak, T., Johnes, G. and Thornton, R. (2005). Fundementals of Labor Economcs. Boston, Houghton Mffln Company. Jacobsen, J. P., Wshart Pearce, J. and Rosenbloom, J. L. (1999). The Effects of Chldbearng on Marred Women's Labor Supply and Earnngs: Usng Twn 21

24 Brths as a Natural Experment. The Journal of Human Resources. 34(3): Josh, H., Pac, P. and Waldfogel, J. (1999). The Wages of Motherhood: Better or Worse? Cambrdge Journal of Economcs. 23(5): Klepnger, D. H., Lundberg, S. and Plotnck, R. (1999). How Does Adolescent Fertly Affect the Human Capal and Wages of Young Women? Journal of Human Resources. 34(3): Korenman, S. and Neumark, D. (1992). Marrage, Motherhood, and Wages. Journal of Human Resources. 27(2): Krepp, L. (2007). The Effect of Chldren on Women's Earnngs: Evdence from Australan Data. Department of Economcs, Unversy of Melbourne, Research Paper Number Loughran, D. S. and Zssmopoulos, J. M. (2009). Why Wa? The Effect of Marrage and Chldbearng on the Wages of Men and Women. The Journal of Human Resources. 44(2): Lundberg, S. and Rose, E. (2000). Parenthood and the Earnngs of Marred Men and Women. Labour Economcs. 7(6): Neumark, D. and Korenman, S. (1994). Sources of Bas n Women's Wage Equatons. Journal of Human Resources. 29(2): Rodgers, J. R. (2004). Hourly Wages of Full-Tme and Part-Tme Employees n Australa. Australan Journal of Labour Economcs. 7(2): Rosen, C. J. (1986). The Theory of Equalzng Dfferences. Handbook of Labor Economcs. Amsterdam, Elsever Scence. 1: Smonsen, M. and Skpper, L. (2006). The Costs of Motherhood: An Analyss Usng Matchng Estmators. Journal of appled econometrcs. 21(7): Tanguch, H. (1999). The Tmng of Chldbearng and Women's Wages. Journal of Marrage and Famly. 61(4): Waldfogel, J. (1995). The Prce of Motherhood: Famly Status and Women's Pay n a Young Brsh Cohort. Oxford Economc Papers. 47(4): Waldfogel, J. (1997). The Effect of Chldren on Women's Wages. Amercan Socologcal Revew. 62(2): Waldfogel, J. (1998a). Understandng the `Famly Gap' n Pay for Women wh Chldren. Journal of Economc Perspectves. 12(1): 137. Waldfogel, J. (1998b). The Famly Gap for Young Women n the Uned States and Bran: Can Materny Leave Make a Dfference? Journal of Labor Economcs. 16(3): 505. Whehouse, G. (2002). Parenthood and Pay n Australa and the Uk: Evdence from Workplace Surveys. Journal of Socology. 38(4): Wooden, M. and Watson, N. (2007). The Hlda Survey and It's Contrbuton to Economc and Socal Research (So Far). The Economc Record. 83(261):

25 APPENDIX A Descrpton of Varables Hourly Wage. Usual weekly gross wages and salary n the man job dvded by the hours per week usually worked n the man job. Nomnal wages are converted to 2007 dollars. Chld Varables. A mother s defned as a woman wh one or more natural or adopted, resdent or non-resdent lvng chldren of any age. Chldren who have snce ded, foster chldren and step chldren are not counted when constructng the Chld 1 and Chld 2. Experence (Equatons 1 and 2). Equatons 1 and 2 use the HILDA varable for work experence whch measures tme (expressed n years) spent n pad work n all jobs, part-tme or full-tme, snce fnshng full-tme educaton for the frst tme. Experence (Equatons 3 and 4). In the way the sample s constructed, actual experence ncreases by approxmately one year between wage observatons. However due to dfferences n the ntervew dates, and ncomplete work years, most changes n experence are slghtly more or less than one. Snce the econometrc specfcaton requres that experence ncreases by exactly one each year, experence has been constructed whch equals actual experence n the frst year, and ncreases by one each workng year thereafter. The constructed and HILDA experence varables are strongly correlated ( ). Educaton. The hghest educaton level acheved: Post Grad = Doctorate, Masters, Graduated Dploma or Graduate Certfcate Bachelor = Bachelor Degree Dploma = Dploma or Advanced Dploma Certfcate = Certfcate I, II, III, IV or undefned certfcate level Yr 12 = Year 12 Yr 11 = Year 11 or below Tenure. Tme (expressed n years) spent workng wh the current employer. Part-tme. Part-tme employees are those workng fewer than 35 hours per week. Casual. Casual workers are those recevng no pad holday or pay sck leave. Industry: The Australan and New Zealand Standard Industral Classfcaton (ANZSIC) classfcaton s used: Prmary = Agrculture, Forestry and Fshng Utl/Mnng = Electrcy, Gas, Water, Waste Servces and Mnng Manufacture = Manufacturng Constructon = Constructon and Wholesale Trade Ret/Hosp = Retal Trade, Accommodaton, Food Servces and Admnstraton Transport = Transport, Postal and Warehousng Culture = Informaton, Meda, Telecommuncatons, Arts and Recreaton Fn/Scence = Fnancal, Insurance, Rental, Hrng, Real Estate, Professonal, Scentfc and Techncal Servces Educ/Health = Publc Admnstraton and Safety, Educaton and Tranng, Health Care and Socal Assstance Other = Other Servces Occupaton: The Australa and New Zealand Standard Classfcaton of Occupatons (ANZSCO) s used: Managers = Managers Professonals = Professonals Trade = Techncans and Trade Workers 23