Underemployed women: an analyss of voluntary and nvoluntary parttme wage employment n South Afrca Colette Muller Mullerc2@ukzn.ac.za School of Economcs and Fnance Unversy of KwaZulu-Natal Abstract Usng natonally representatve household survey data from 1995 to 2006, ths paper explores heterogeney among female part-tme wage (salared) workers n post-aparthed South Afrca, specfcally dstngushng between ndvduals who choose to work parttme and part-tme workers who report wantng to work longer hours. As n studes of voluntary and nvoluntary part-tme employment n other countres, the fndngs show that nvoluntary part-tme workers n South Afrca are outnumbered by voluntary parttme workers. In contrast to other countres, however, nvoluntary underemployment n South Afrca has not rsen substantally over tme, nor s there consstent evdence to suggest a posve correlaton between nvoluntary underemployment and broad unemployment. Sgnfcant dfferences are found among part-tme workers, wh occupatonal characterstcs specfcally beng dentfed as key correlates of nvoluntary part-tme employment. The wage premum to female part-tme employment n South Afrca, dentfed n an earler study, s shown to be robust also to a dstncton among part-tme workers, and nvoluntary part-tme workers are found to have a stronger labour force attachment than women who choose to work part-tme. JEL Classfcaton: J21; J31; J60. Keywords: part-tme; underemployment; South Afrca 1. Introducton Studes of part-tme and full-tme employment among women often assume mplcly that women choose part-tme work, even f ths s a constraned choce n the face of chldcare and other home responsbles, and that women would not work more f addonal employment was made avalable. But n developng countres, and partcularly countres lke South Afrca that face hgh and rsng unemployment rates and wdespread poverty, women who work part-tme may be nvoluntarly underemployed. Although these women may prefer full-tme employment they may be forced to take on part-tme jobs because there s no other, or no more, employment avalable. Usng data from selected natonal household surveys, ths paper ams to nvestgate nvoluntary and voluntary part-tme wage employment among women n South Afrca. Frst, I explore trends n part-tme employment n South Afrca from 1995 to 2006, dstngushng nvoluntary from voluntary part-tme workers. I also examne whether changes n underemployment among women n post-aparthed South Afrca tracks trends n female unemployment over the perod. Second, I use multvarate analyss to examne how dfferent female voluntary and nvoluntary part-tme workers are from each other n terms of ther ndvdual attrbutes as well as n ther household and occupatonal
characterstcs. Thrd, I consder the returns to voluntary and nvoluntary part-tme work and examne specfcally whether the premum to women s part-tme work n South Afrca, dentfed by Posel and Muller (2008), s robust to a dstncton among the parttme employed. Fnally, I dentfy whether there s evdence of dfferences n labour market attachment among South Afrca s voluntary and nvoluntary part-tme workers. The next secton dscusses the data sources analysed and the defnons of voluntary and nvoluntary part-tme employment used n the study. Secton 3 descrbes trends n voluntary and nvoluntary part-tme wage employment, whle secton 4 uses multvarate analyss to dentfy the correlates of voluntary and nvoluntary part-tme employment. Earnngs dfferences among part-tme workers are explored n secton 5, and n secton 6 dfferences n labour market attachment among voluntary and nvoluntary part-tme workers are dentfed. 2. Data sources and defnons Ths study uses both cross-sectonal and panel household survey data collected by South Afrca s offcal data collecton agency, Statstcs South Afrca (StatsSA), to explore voluntary and nvoluntary part-tme employment among women n South Afrca. Trends n voluntary and nvoluntary part-tme work and n unemployment are descrbed usng data from the 1995 and 1999 October Household Surveys (OHSs), together wh data from selected September rounds of the Labour Force Survey (LFS) from 2000 to 2006. Both of these survey nstruments comprse a sample of approxmately 30 000 households n 3 000 clusters, and collect comprehensve nformaton on ndvdual s labour market partcpaton and wages. To dentfy the correlates of voluntary and nvoluntary part-tme employment, and for the estmatons of wage equatons that control for dfferences n observable characterstcs between workers, a dataset constructed by poolng the bannual rounds of the crosssectonal LFSs from September 2001 to March 2006 s used. Fnally, I use StatsSA s release of the LFS Panel, whch also comprses data from September 2001 to March 2006, to control for ndvdual fxed effects n the earnngs estmatons and to dentfy the frequency and percentage of women changng labour market status over tme. Part-tme workers are dentfed as wage (salared) employees who usually work fewer than 35 hours a week. The 35-hour threshold, whch was also adopted n the South Afrcan study of part-tme/full-tme earnngs dfferentals among women by Posel and Muller (2008), s often used to defne part-tme employment by surveys n the Uned States (see Hrsch 2005 and Hardoy and Schøne 2006, for example). A dstncton between voluntary and nvoluntary part-tme workers s made based on the recommendatons of the Sxteenth Internatonal Conference of Labour Statstcans (ICLS) regardng the defnon of tme-related underemployment (ILO 1998). In partcular, the ICLS recommends that the underemployed be dentfed as ndvduals who are a) wllng to work addonal hours; b) avalable to work addonal hours and c) work less than an hourly threshold durng the reference perod. Ths study adopts a defnon of nvoluntary underemployment that s broadly consstent wh the ICLS recommendatons, dentfyng nvoluntary part-tme workers (the nvoluntarly 2
underemployed) as part-tme workers who are wllng to work longer hours (satsfyng parts a) and c) of the ICLS crera.) 1 Voluntary part-tme workers are those part-tme workers who do not want to work longer hours. 3. Trends n nvoluntary and voluntary part-tme employment, and n unemployment n post-aparthed South Afrca Studes that dstngush between voluntary and nvoluntary part-tme employment have been prmarly concerned wh the underemployed as an underutlsed labour resource, focusng on the ncdence of nvoluntary part-tme employment n relaton to the level of economc actvy. Researchers have shown that although voluntary part-tme workers often outnumber nvoluntary part-tme workers, nvoluntary underemployment has typcally become more prevalent over tme wh frms turnng to part-tme rather than full-tme employment as a means of reducng labour costs. There s also evdence of a strong posve relatonshp between nvoluntary part-tme work and unemployment. Faced wh a recesson, frms may decrease the hours worked by some of ther employees n addon to layng-off workers (Tlly 1991; Stratton 1996, Görg and Strobl 2003). Furthermore, ndvduals may be more wllng to consder part-tme employment as an alternatve to a full-tme job when faced wh an envronment of economc declne (Buddelmeyer et al 2008). Table 1 descrbes trends n women s wage employment n South Afrca from 1995 to 2006. The results show that women s work has expanded substantally, rsng by more than twenty percent from 1995 to 2006. Part-tme work has also been a key component of the ncrease n women s wage employment, growng by more than 150 000 jobs over the perod. In addon, women s share of part-tme employment has ncreased consderably, rsng from about 60 percent to more than 67 percent. These fndngs suggest that the expanson n part-tme employment n South Afrca has been an mportant component of the documented femnsaton of the country s labour force over the post-aparthed perod (see also Posel and Muller 2008). [Table 1 about here] In addon to presentng estmates of compose part-tme wage employment among women, the table shows dsaggregated estmates that dstngush between voluntary and nvoluntary part-tme wage workers. The results reveal that, lke n other countres, the number of ndvduals workng part-tme voluntarly typcally exceeds those part-tme workers who desre longer workng hours. In contrast to other countres, however, nvoluntary part-tme employment has not become more prevalent over the years. Followng an ncrease of approxmately forty percent n the number of nvoluntary parttme workers from 1995 to 2000, nvoluntary part-tme employment has remaned que stable, averagng at around 164 000 women. The share of nvoluntary part-tme employment n total part-tme work has also remaned relatvely constant from 2001 onwards, at approxmately one thrd on average, whle the share of nvoluntary part-tme 1 See Muller (2009) for further dscusson on the dstncton between voluntary and nvoluntary part-tme wage workers n South Afrca. 3
work n total wage employment has typcally declned. Ths s because of an expanson n total employment that has contnued snce 2001. To dentfy whether nvoluntary underemployment n South Afrca follows changes n unemployment, graphcal representatons of trends n nvoluntary and voluntary part-tme wage employment, together wh trends n broad unemployment 2 are shown n Fgure 1. [Fgure 1 about here] For the 1995 to 1999 perod, the fgure suggests that the change n nvoluntary part-tme employment tracks the change n broad unemployment. In partcular, both nvoluntary part-tme work and broad unemployment ncreased sgnfcantly over these years, suggestng a posve relatonshp between unemployment and underemployment. From 1999 onwards, however, changes n broad unemployment and n nvoluntary part-tme work have moved mostly n oppose drectons, and are ndcatve of a negatve correlaton between unemployment and underemployment among women n South Afrca. Broad unemployment ncreased by about one mllon ndvduals from 1999 up untl 2002. Snce 2003, broad unemployment has typcally fallen (although the magnudes of the reductons n unemployment n each year have not been that large). In contrast, nvoluntary part-tme employment declned rapdly from 1999 to 2002. Followng a small ncrease of about 20 000 ndvduals from 2002 to 2003, nvoluntary underemployment remaned relatvely stable from 2003 to 2006. 4. Dfferences n voluntary and nvoluntary part-tme employment Although studes do not usually make a drect comparson between voluntary and nvoluntary part-tme workers, a few researchers have recognsed that there may be dfferences among ndvduals who work part-tme when comparng part-tme workers to those who work full-tme and to other labour market groups (Leppel and Clan 1993; Barret and Doron 2001; Görg and Strobl 2003). Although dfferng methodologes are adopted n these studes 3, the results suggest that dfferences n the preferences of parttme workers for addonal hours are lkely to stem both from dfferences n personal characterstcs among part-tme workers as well as dfferences n the types and/or qualy of part-tme jobs. To test the correlates of nvoluntary versus voluntary underemployment n South Afrca a smple prob model s estmated wh data from the pooled September 2001 to March 2004 LFS cross-sectons: Pr( y = 1 X, T ) = Φ( η, X, T ) (1) 2 The broadly unemployed nclude ndvduals who are wllng to accept employment but who may not be actvely seekng work. Estmates of broad unemployment have been dvded by ten to allow them to be compared on the same scale as those for nvoluntary part-tme work. 3 An overvew s provded n Muller (2009). 4
The dependent varable, y, s a bnary categorcal varable whch takes the value of 1 f the ndvdual s nvoluntarly underemployed and 0 f the ndvdual works part-tme voluntarly. X s a vector of observed characterstcs for ndvdual, T contans fve tme-dummy varables, each representng one of the cross-sectonal waves (the frst wave s used as the reference category), η s a vector of parameters, and Φ s the standard cumulatve normal dstrbuton. Because the sample of nvoluntary part-tme workers n each of the LFS cross-sectons s que small, usng data from the pooled cross-sectons of the LFS allows for a larger sample sze whch ncreases the relably of the estmated coeffcents and test statstcs. The ssue of sample selecton bas, whch typcally arses n estmatons wh contnuous dependent varables, also poses a problem n models such as (1) whch use bnary dependent varables. In eher case, not accountng for unobservable dfferences between voluntary and nvoluntary part-tme wage workers may result n an endogeney problem that can bas the estmated coeffcents. The results presented n Table 2 have, however, not been corrected for sample selecton bas and should be nterpreted as condonal on the selecton nto voluntary and nvoluntary part-tme work. 4 Four sets of regresson results are shown, wh the number of varables constutng X ncreasng n each specfcaton. In the frst specfcaton (I) varables controllng for ndvdual characterstcs (populaton group and maral status) and locaton (provnce of resdence and urban/rural area) are ncluded, together wh varables affectng the ndvdual s potental productvy (age, educaton and job duraton). The second specfcaton (II) controls further for household composon, ncludng controls for the number of chldren and for the number of unemployed adults n the household. The number of employed men and the number of other employed women lvng n the household reflect the ndvdual s access to earned ncome whn the household. In specfcaton III characterstcs related to occupatons are ntroduced namely occupatonal and ndustry categores, whether the ndvdual works n a large frm (n excess of ffty employees), unon membershp and sector of employment. Fnally, controls for condons of work are ncluded n specfcaton IV. The results reported are the margnal effects, estmated at the mean for contnuous varables and for a dscrete change from zero to one for the dummy varables. The effects of dfferent subsets of controls across the estmatons are dscussed n detal below. 5 [Table 2 about here] a) Experence and job duraton Across all four specfcatons, the results suggest that the probably of workng part-tme nvoluntarly rather than voluntarly nally ncreases n age whch then tapers off. The 4 For further dscusson see Muller (2009). 5 Lkelhood rato tests confrmed that the addonal varables ncluded n each specfcaton were jontly sgnfcant. Full sets of estmates for all the econometrc results presented are shown n Muller 2009 (Appendx B). 5
posve effect of the contnuous varable for age s sgnfcant only n the frst and fourth specfcatons, however, whle the effect of the negatve quadratc age varable s sgnfcant (albe very small) across all specfcatons. Longer job duraton s negatvely assocated wh nvoluntary part-tme work and may reflect the precarous and unstable nature of the jobs occuped by nvoluntary part-tme workers. b) Educaton In all four specfcatons, educatonal attanment s an mportant correlate of nvoluntary part-tme employment. In comparson to women n part-tme employment who are smlar n other observed characterstcs, the probably of beng n nvoluntary part-tme employment ncreases by between 6.1 and 7.1 percentage ponts when the woman has completed prmary but not secondary school. Ths suggests that among nvoluntary parttme workers, educaton may act as barrer to full-tme employment. Analogously, havng a completed tertary educaton decreases the probably that female wage workers are workng nvoluntarly n part-tme employment. Ths effect s sgnfcant n specfcatons I and II and nsgnfcant n specfcatons III and IV; the magnude of the margnal effect also declnes substantally n the latter two specfcatons (from more than nne percentage ponts n I and II to less than three percentage ponts n IV). Multcollneary between tertary educaton and some of the occupaton and ndustry varables ntroduced n specfcaton III may account for ths result. c) Populaton group In all specfcatons the results show that the probably of nvoluntary part-tme employment s sgnfcantly lower among the other populaton groups n comparson to Afrcans, the reference group. In partcular, among part-tme workers who are smlar n other observable characterstcs, the probably of wantng to work longer hours s the lowest among Whes (between 15 and 21 percentage ponts lower than among Afrcans). Indans also have a smaller probably of nvoluntary part-tme employment n comparson to Afrcans (between eght and 13 percentage ponts lower) as do Coloureds (sx to 7.5 percentage ponts lower). These fndngs may reflect the effect of dfferences n ncome between ndvduals, for whch populaton group may serve as a proxy. d) Maral status and household characterstcs Across all specfcatons beng prevously marred (wdowed or dvorced) rather than unmarred sgnfcantly rases the probably of workng part-tme nvoluntarly by about four percentage ponts. In specfcatons I, II and III the probably of nvoluntary underemployment s lower for ndvduals who are marred or cohabng 6, although ths effect s sgnfcant only n specfcaton I. It s possble that the declne n the magnude (and sgnfcance) of the marrage/cohabaton dummy varable from specfcaton II 6 The LFS questonnares only dfferentated between marrage and cohabaton n surveys conducted from September 2004 onwards. In the LFS data used here s therefore mpossble to dstngush ndvduals who are marred and who lve wh ther spouses from ndvduals who are not marred but who resde wh ther partners. 6
onwards s the result of multcollneary: marrage/cohabaton s posvely correlated wh the number of employed men n the household whch was ntroduced as a control varable n specfcaton II. Havng access to earned ncome (through lvng n a household wh other employed men or women) sgnfcantly reduces the probably of workng part-tme nvoluntarly by about sx percentage ponts n specfcatons II, III and IV. These fndngs suggest that fnancal support from household members may be a crcal factor enablng women to voluntarly work part-tme. In contrast, as the number of unemployed men and women n the household rses, the probably of a part-tme worker wantng to work longer hours ncreases ndcatve of the worker s need to work more hours to n order to earn more to support members of her household. There s also an nverse relatonshp between nvoluntary part-tme employment and non-market actves such as chldcare, whch s consstent wh women choosng part-tme employment as a way of combnng market work wh the care of chldren. e) Locaton In all specfcatons nvoluntary part-tme workers are shown to be sgnfcantly more lkely to lve n urban areas than voluntary part-tme workers. One explanaton for ths fndng s that women may face greater fnancal pressure to work longer hours f lvng expenses are hgher than n urban areas. It s also possble that the estmated relatonshp between nvoluntary part-tme employment status and resdng n an urban area s overstated as a result of a selecton bas. Ths could occur f, for example, women who want to work longer hours mgrate to urban areas where there are more employment opportunes. f) Occupaton and ndustry The results reported n specfcaton III suggest that nvoluntary part-tme workers are sgnfcantly less lkely than voluntary part-tme workers to work n occupatons that offer unon protecton. The margnal effect of unon membershp on the probably of nvoluntary part-tme employment declnes n specfcaton IV, although remans negatve. Ths s probably accounted for by multcollneary between unon membershp and condons of work, for whch controls were ntroduced n specfcaton IV n addon to hgher wages, the benefs of unonsed employment may also nclude preferental workng and job condons. Involuntary part-tme work s also posvely assocated wh workng n a large frm. Large frms may be more wllng than smaller frms to employ part-tme workers to meet demand durng peak perods, and may also be more lkely to shorten the workng hours of ther full-tme staff complement durng economc slow downs. In addon, the probably of workng part-tme nvoluntarly s sgnfcantly lower n the agrcultural sector (the reference ndustry) than n other ndustres. The types of jobs offered part-tme n South Afrca s agrcultural sector (fru and vegetable pckng, for example) are lkely to be seasonal n nature, attractng ndvduals specfcally seekng nterm employment. 7
g) Condons of work Involuntary part-tme employment among women s assocated both wh sgnfcantly fewer benefs (medcal ad contrbutons and pad leave, n partcular) and wh more nsecure employment workng n an occupaton whch s permanent sgnfcantly decreases the probably of nvoluntary part-tme work by 16.4 percentage ponts, ceters parbus. Ths result s among the largest of the margnal effects, and would be consstent wh nvoluntary part-tme workers seekng ways to maxmse ther current ncome streams n the face of uncertan future employment prospects. The results of ths analyss suggest that sgnfcant dfferences exst between women who work part-tme voluntarly and those who are reported to desre longer workng hours. In addon to ndvdual characterstcs lke age and educaton, household characterstcs, such as lvng n households where employed men and unemployed adults also resde, appear to be crcal correlates of nvoluntary part-tme work. The probably of nvoluntary part-tme employment ncreases sgnfcantly wh an ncrease n the number of unemployed adults resdng n the household, for example, whle the probably of nvoluntary part-tme employment s sgnfcantly lowered by an ncrease n the number of employed men n the household. These fndngs suggest that fnancal support from household members (or a lack thereof) s a key factor nfluencng whether part-tme workers desre longer workng hours. Job characterstcs and condons of employment n partcular, are also mportant correlates of nvoluntary part-tme employment. In comparson to the jobs of women who voluntarly work part-tme, the work performed by nvoluntary part-tme workers s sgnfcantly less lkely to be permanent and less lkely to offer unon protecton or benefs n comparson. The poor qualy of the jobs occuped by nvoluntary part-tme workers can also be expected to reflect n ther remuneraton. Wage dfferences between voluntary and nvoluntary part-tme workers and the full-tme employed are nvestgated n the followng secton. 5. Voluntary and nvoluntary part-tme employment and wages Posel and Muller (2008) show that despe female part-tme workers n South Afrca earnng sgnfcantly less per month, on average, than ther full-tme counterparts, as well as less per hour, once observable and unobservable dfferences between part-tme and full-tme workers are accounted for, there s evdence of a wage premum to female parttme wage employment. Gven sgnfcant dfferences n the characterstcs of voluntary and nvoluntary part-tme workers hghlghted above, ths secton nvestgates whether ths wage premum perssts once the heterogeney n observed and unobserved characterstcs between these two groups are accounted for. Estmates of average hourly and monthly wages, calculated usng data from the September 2003 LFS, together wh estmates of average workng hours are shown for female voluntary and nvoluntary part-tme workers and female full-tme workers n Table 3. The dstrbutons of hourly wages for voluntary and nvoluntary part-tme 8
workers and the full-tme employed are shown by the kernel densy plots n Fgure 2, and the dstrbutons of workng hours for voluntary and nvoluntary part-tme workers are shown n Fgure 3. [Fgure 2 about here] [Fgure 3 about here] [Table 3 about here] Fgure 2 shows that the hourly wage dstrbuton for nvoluntary part-tme workers s more compressed than that for voluntary part-tme workers and s skewed to the rght. As a result, average hourly wages are sgnfcantly hgher among part-tme workers who do not want more hours. In contrast, the dstrbuton of workng hours for voluntary parttme workers, shown n Fgure 3, s more compressed than for the nvoluntarly underemployed and s skewed to the left. Mean workng hours are therefore lower among nvoluntary part-tme workers than among voluntary part-tme workers. Workng fewer hours, on average, than voluntary part-tme workers, and at a lower mean hourly wage, translates nto monthly wages that are sgnfcantly lower among nvoluntary part-tme workers. On average, nvoluntary part-tme workers earn less than half the monthly wage of ndvduals who voluntarly work part-tme. The statstcs presented n Table 3 also reveal sgnfcant dfferences n both monthly and hourly wages between nvoluntary part-tme workers and the full-tme employed, and between voluntary part-tme workers and the full-tme employed. Involuntary part-tme workers earn sgnfcantly less per hour, on average, than full-tme workers, whle women who voluntary work part-tme earn sgnfcantly more. Because they work fewer hours, however, the monthly wages of both voluntary and nvoluntary part-tme workers are sgnfcantly lower than for the full-tme employed. Per month, the average wage for a voluntary part-tme worker s about forty percent lower than for a full-tme worker, whle the average monthly wage of an nvoluntary part-tme worker s less than one-quarter of that receved by a full-tme worker. To explore wage dspares between voluntary and nvoluntary part-tme workers and those who work full-tme further, data from the pooled LFS cross-sectons, along wh data from the LFS Panel from September 2001 to March 2004 are used. Of partcular nterest here s establshng whether the premum to female part-tme employment n South Afrca, dentfed by Posel and Muller (2008), s robust to a dstncton among parttme workers. The analyss begns by usng data from the pooled LFS cross-sectons to estmate: ln( W ) = α + φv + ϑi + βx + τt + ε (2) t The dependent varable s the natural logarhm of ndvdual hourly earnngs ( W ) and ε s the error term. Indvdual, household and job characterstcs are ncluded n the vector X, and fve dummy varables, each representng one of the cross-sectonal data- 9
sets, are ncluded n the vector T t (the frst cross-sectonal data set serves as the reference category). The dummy varable V takes on a value of 1 f the ndvdual works part-tme voluntarly, whle the dummy varable I equals 1 f the ndvdual s an nvoluntary parttme worker. Full-tme workers are ncluded n the comparson category. If the premum to female part-tme wage employment n South Afrca s robust to a dstncton between voluntary and nvoluntary part-tme wage workers, then both φˆ and ϑˆ wll be posve. One concern wh usng a model such as (2) to estmate and compare the returns to voluntary and nvoluntary part-tme employment s that does not account for the possbly that there are also non-random unobservable dfferences between the two groups of workers. Falure to account for dfferences n selecton between the two groups could bas the coeffcent estmates. To address problem of selecton bas, data from the LFS Panel s used. The cross-sectonal waves of the LFS Panel are pooled, and OLS s used to estmate: ln( W ) = α + φv + ϑi + βx + τt + + v (3) t The key dfference between (2) and (3) s n the specfcaton of the error term. In (3) the compose error term has been dsaggregated nto a tme varant and a tme nvarant component. The tme nvarant component of the error term,, s presumed to capture the effects of unobservable characterstcs that reman constant over tme. One of the problems wh usng panel data s that non-random attron may cause the resultng sample to be unrepresentatve of the populaton. To assess how representatve the cross-sectonal waves of the panel are gven the dstncton between voluntary and nvoluntary part-tme workers, results from the estmaton of equaton (3) are benchmarked aganst those obtaned by estmatng equaton (2) usng the pooled data from the full cross-sectonal waves of the LFS (data whch should be unbased by the problem of attron). The fxed-effects transformaton s then estmated, where, through tme-demeanng, the tme nvarant component of the error term s removed. ln( W ) ln( W ) = φ FE ( V V ) + ϑ FE ( I I ) + β FE ( X X ) + τ FE ( T t T ) + v v (4) In the study by Posel and Muller (2008), controllng for ndvdual fxed effects n the wage estmatons for part-tme and full-tme employment resulted n an ncrease n the estmated premum to female part-tme employment, suggestng that workers were negatvely selected nto part-tme employment. It s possble, though, that the selecton effects nto part-tme employment may dffer for voluntary and nvoluntary part-tme workers. Negatve selecton may be expected among voluntarly part-tme workers f these ndvduals have less commment to the labour force or are less motvated, whle the converse would be expected among the nvoluntarly underemployed f ther desre to 10
work longer hours sgnals greater motvaton or a stronger commment to employment. FE If there s negatve selecton nto voluntary part-tme work then φˆ from equaton (4) wll exceed φˆ from equaton (3). Smlarly, FE ϑˆ from equaton (4) should be lower than ϑˆ from equaton (3) f there s posve selecton nto nvoluntary part-tme work. The dentfcaton of a posve selecton effect may, however, be complcated by attenuaton bas. If measurement error n the change n voluntary/nvoluntary part-tme status causes the fxed effects estmates to be understated, then may be dffcult to determne whether any declne n the fxed effects estmate of the wage premum relatve to the OLS estmate s the result of posve selecton or the consequence of attenuaton bas. In addon, the effects of negatve selecton may be understated n the presence of measurement error. 7 [Table 4 about here] The results of the wage regressons from the pooled cross-sectonal data, estmated for three sets of covarates usng OLS, are presented n Table 4. In the frst specfcaton, controls for ndvdual characterstcs (age and job duraton, educaton, maral status and locaton) are ncluded, and n the second specfcaton addonal controls for occupaton type and ndustry, along wh sector of employment, whether the frm s large (a large frm s one wh more than ffty employees), and whether the ndvdual belongs to a unon are added. In the thrd specfcaton, varables controllng also for condons of work are ncluded whether employment s permanent rather than casual or temporary, whether the ndvdual receves penson fund and/or medcal ad and/or Unemployment Insurance Fund contrbutons from ther employer, and whether the employer provdes pad leave. The fndngs suggest that the wage premum to female part-tme employment n South Afrca s robust to a dstncton among part-tme workers, wh an estmated wage premum to nvoluntary part-tme employment of between 28 percent and 67 percent, and a premum to voluntary part-tme employment of between thrty and 58 percent, dependng on the controls utlsed. The results reflect that not only are there sgnfcant dfferences n observable characterstcs between part-tme workers and the full-tme employed, but that substantal dfferences exst also among part-tme workers. The results of F-tests show that the dfference n the premum to voluntary and nvoluntary part-tme employment s sgnfcant only n specfcaton III, however. It s therefore as a result of dfferences n ther condons of work that sgnfcant dfferences n the wage premums to voluntary and nvoluntary part-tme employment are observed, despe ther beng substantal dfferences also n the ndvdual and occupatonal characterstcs of these groups. 7 Although there are a number of correctve procedures avalable to address the problem of errors n varables (such as weghted regresson and nstrumental varables, for example) data lmatons prevented these from beng mplemented here. 11
Although these results are consstent wh the cross-sectonal estmates of the part-tme employment premum documented n Posel and Muller (2008), as n ther study, falure to account also for dfferences n unobservable characterstcs between voluntary and nvoluntary part-tme workers and those who work full-tme could bas the estmated coeffcents. Possble dfferences also n the drecton of selecton nto voluntary and nvoluntary part-tme employment could further complcate the nterpretaton of the results: negatve selecton nto voluntary part-tme employment and posve selecton nto nvoluntary part-tme employment, for example, would reduce the dfference n the wage premums between each group. To address the problem of selecton bas, data from the LFS Panel s used to estmate a fxed effects regresson, whch dfferences out the unobserved effects. The results, estmated usng the full set of covarates, are shown n Table 5. The frst column presents estmates from the pooled LFS data from the full cross-sectons, and results from the pooled waves of the LFS Panel are shown n the second column. By comparng the estmates obtaned from the pooled waves of the panel wh those from the pooled crosssectons of the orgnal sample, s possble to dentfy whether the panel sample has been affected by the problem of attron. The results presented n Table 5 suggest that the dfferences between the panel sample and the orgnal cross-sectonal sample are not that large. The thrd column reports the fxed-effects estmates, where the effect of nonrandom unobservable dfferences between voluntary and nvoluntary part-tme workers and those who work full-tme have been accounted for. [Table 5 about here] The estmates from all three specfcatons confrm the earler cross-sectonal fndngs, and show that the estmated wage premum to part-tme work n South Afrca s not sensve to a dstncton among part-tme workers. A substantal and sgnfcant premum to both voluntary and nvoluntary part-tme wage employment among women perssts even when unobservable dfferences between workers have been accounted for. The dfference between the premums to voluntary and nvoluntary part-tme employment narrows consderably n the fxed effects estmaton, however, and although controllng also for unobservable dfferences between workers causes the estmated premum to nvoluntary part-tme work to exceed that for voluntary part-tme employment, F-tests show that the dfference n the magnude of these estmated wage premums s not sgnfcant. Ths narrowng of the gap n the wage premums between voluntary and nvoluntary part-tme workers appears to be a consequence of dfferences n the drecton of the selecton effect between voluntary and nvoluntary part-tme workers. When comparng the results from column II and column III, can be seen that the sze of the coeffcent on voluntary part-tme employment ncreases when estmatng the whntransformaton, whle the there s a (small) decrease n the coeffcent on nvoluntary part- 12
tme employment. These results are consstent wh negatve selecton nto voluntary parttme employment, and wh posve selecton nto nvoluntary part-tme employment. 8 Even though the effects of endogeney bas on the parameter estmates, ntroduced by the problem of sample selecton, has been addressed n the fxed-effects estmaton a further source of bas (n addon to that resultng from errors n varables) remans n the results presented above. In partcular, smultaney bas may occur f changes n employment status are a functon of changes n the wage rate. Hgher wage growth could see women workng full-tme choosng to work fewer hours, resultng n them changng ther status to voluntary part-tme. Alternatvely, hgher wage growth may nduce employers to reduce workng hours, causng women workng full-tme to become nvoluntarly underemployed. Classfcaton as an nvoluntary/voluntary part-tme worker may also be dependent on earnngs. 9 Hgher wage growth could cause the nvoluntarly underemployed to become voluntary part-tme workers (condonal on workng hours) whle low wage growth could result n converse. The mplcaton of smultaney bas for the results presented here s that the estmated wage premums to both voluntary and nvoluntary part-tme employment may be overstated. But because s not possble to dentfy any nstrumental varables n the LFS Panel that dstngush between voluntary and nvoluntary part-tme workers and the full-tme employed, any potental overestmaton of these wage premums cannot be addressed. 6. Labour force attachment among voluntary and nvoluntary part-tme workers Although the premum to women s part-tme employment n South Afrca appears robust to a dstncton between voluntary and nvoluntary part-tme employment, evdence pontng to possble dfferences n the drecton of selecton nto these employment categores would suggest that voluntary and nvoluntary part-tme workers may exhb dfferng degrees of labour market attachment. By usng panel data to track the movements of ndvduals nto and out of varous labour market states over tme s possble to examne labour force attachment among the employed, and among part-tme workers. Research undertaken n the Uned States (US) suggests that part-tme workers may be more lkely to change labour market status than other groups (Blank 1989; Stratton 1996). Among part-tme workers n the US, dfferences n transon probables have also been dentfed, wh voluntary part-tme workers beng less lkely to move nto fulltme employment than the nvoluntarly underemployed (Stratton 1996). To nvestgate the labour force attachment of voluntary and nvoluntary part-tme workers n South 8 Note that f nvoluntary part-tme employment status, as well as changes n nvoluntary part-tme employment status over tme has been measured wh error, then the effects of posve selecton may be overstated and the effects of negatve selecton understated. 9 From the questons asked of respondents n the LFS questonnares s not possble to dentfy whether a part-tme wage employee who s reported to want longer workng hours would work these addonal hours at the exstng wage rate, or whether they would be content wh ther current hours gven an ncrease n ther wage. Smlarly, for those who do not want longer workng hours, s not possble to determne whether ther preferences would reman unchanged f they were faced wh a hgher or lower wage. 13
Afrca the frequency and percentage of women changng labour market status between adjacent perods n the LFS Panel are presented n Table 6. [Table 6 about here] The results on the leadng dagonal of Table 6 show the frequency and percentage of workers who stayed n ther respectve labour market statuses. The transon probables depct consderable churnng n the South Afrcan labour market, partcularly among those who work part-tme. Less than one quarter of voluntary part-tme workers, and less than one-ffth of nvoluntary part-tme workers remaned n these respectve employment states over the adjacent panel waves. Involuntary part-tme workers have only lmed success n achevng ther desre for longer workng hours: approxmately one-thrd of part-tme workers who ndcated that they would lke to work more hours transoned nto full-tme jobs. An even larger porton (almost forty percent) of voluntary part-tme workers reported full-tme employment n the followng perod, however. These fndngs suggest that voluntary part-tme workers fnd easer to access full-tme employment than the nvoluntarly underemployed. One possbly s that voluntary part-tme employment s transory. Women may revert to full-tme employment followng perods of reconclng market work and household responsbles such as chldcare, for example. It s also possble that the knds of occupatons held by voluntary part-tme workers offer greater opportunes for longer workng hours. The precarous and unstable nature of the jobs occuped by nvoluntarly part-tme workers can be seen when consderng the movements of workers out of employment over the waves of the panel. In comparson to voluntary part-tme workers, of whom less than one-thrd reported leavng employment, a greater percentage of nvoluntary parttme workers (almost 35 percent) exed employment. However, nvoluntary part-tme workers who left employment were more lkely to mantan an attachment to the labour market (becomng unemployed) than voluntary part-tme workers, who were more lkely to leave the labour force. Almost one-quarter of nvoluntary part-tme workers were reported as unemployed n the followng perod, as compared to only 14 percent of voluntary part-tme workers, and approxmately 17 percent of voluntary part-tme workers exed the labour market as compared to just ten percent of the nvoluntarly underemployed. These fndngs on the transon out of employment are ndcatve of dfferences n commment to employment between voluntary and nvoluntary part-tme workers, and would be consstent wh the results presented earler whch ponted to possble dfferences also n the drecton of selecton nto these types of employment. Fnally, there s only lmed evdence that part-tme employment n South Afrca provdes a successful route out of unemployment, wh unemployed ndvduals beng more lkely to transon nto full-tme than part-tme wage employment. Whle almost nne percent of the unemployed found full-tme jobs, only about 2.5 percent of ndvduals who started off unemployed were able to obtan part-tme employment by the next perod, and nearly half of these ndvduals reported workng n part-tme jobs that offered nsuffcent workng hours. Overall, appears to be que dffcult for ndvduals whout jobs to obtan work n South Afrca. Across adjacent panel waves, less than 12 14
percent of the unemployed were reported to fnd employment, and nearly seventy percent remaned unemployed but wllng to accept employment. A further twenty percent of broadly unemployed workers were reported as economcally nactve n the next perod. 7. Concludng comments Usng nformaton on the workng hour s preferences of female part-tme workers, ths paper explores emprcally the dfferences between voluntary part-tme worker and the nvoluntarly underemployed (part-tme workers who are reported to want longer workng hours). Lke n other countres, n South Afrca the proporton of part-tme workers who desre longer workng hours s less than the proporton workng part-tme voluntarly. However, n contrast to other countres, where nvoluntary part-tme employment has rsen over tme, n South Afrca the number of nvoluntary part-tme workers has remaned relatvely stable. There s also no consstent evdence of a posve relatonshp between nvoluntary underemployment and unemployment n South Afrca. Although both broad unemployment and nvoluntary part-tme work ncreased from 1995 to 1999, n subsequent years broad unemployment and nvoluntary underemployment have typcally dverged. A multvarate analyss, whch tested the correlates of voluntary and nvoluntary parttme employment, suggested that occupatonal characterstcs n partcular, are key correlates of nvoluntary underemployment. Women who work part-tme and who desre longer workng hours are sgnfcantly more lkely than voluntary part-tme workers to work n occupatons that are nsecure and unprotected by unons, and are sgnfcantly less lkely have permanent jobs. The analyss of earnngs dfferences revealed sgnfcant dfferences also n wages between voluntary and nvoluntary part-tme workers. The mean monthly wage of nvoluntary part-tme workers s sgnfcantly lower than that for voluntary part-tme workers - the result of workng sgnfcantly fewer hours, on average, at a lower mean hourly wage. When dfferences n both ndvdual and job characterstcs are controlled for usng multvarate analyses a premum to both voluntary and nvoluntary part-tme employment s found. Ths result shows that the premum to female part-tme employment n South Afrca s robust to a dstncton n workng hour preferences among part-tme workers. The premum to nvoluntary part-tme employment s also found to be sgnfcantly larger than for voluntary part-tme work when controllng for dfferences n condons of work. However, when fxed effects estmaton s used to address the possbly that non-random unobservable dfferences exst between voluntary and nvoluntary part-tme workers and the full-tme employed, the dfference n the estmated wage premums to voluntary and nvoluntary part-tme employment decreases and s no longer sgnfcant. Dfferences n the drecton of selecton nto voluntary and nvoluntary part-tme employment could account for ths result, whch would be consstent also wh dfferences n labour market attachment among these workers. 15
The labour market attachment of voluntary and nvoluntary part-tme workers was nvestgated n the fnal part of the study. The fndngs correspond, n part, wh those from studes of the Uned States: female part-tme workers n South Afrca are more lkely than other groups to change ther labour market status. Unlke n the Uned States, however, nvoluntary part-tme workers n South Afrca are less lkely to transon nto full-tme employment than voluntary part-tme workers. Although ths result could suggest that voluntary part-tme workers behave n a manner whch s nconsstent wh ther preferences, s also possble that the occupatons of voluntary part-tme workers offer greater opportunes for advancement nto full-tme employment than those of nvoluntary part-tme workers. The analyss of labour market transons also shows that nvoluntary part-tme workers may have a stronger attachment to the labour market than voluntary part-tme workers. A hgher percentage of the nvoluntarly underemployed who left the labour market were reported as unemployed and wllng to accept work n the next perod n comparson to voluntary part-tme workers, of whom a greater percentage were reported as economcally nactve. Part-tme jobs provde a valuable source of employment to many women n South Afrca, partcularly to those wh household responsbles. Although part-tme jobs also have the potental to offer ndvduals who lack the sklls and/or qualfcatons to obtan fulltme employment the opportuny to enter nto the labour market and acqure labour market experence, ths study presents only lmed evdence to suggest that part-tme jobs provde a steppng stone nto employment n the South Afrcan labour market. Gven the already hgh, and contnually ncreasng rates of unemployment n the country, more research s needed to explore whether there s scope to expand the opportunes for parttme employment n South Afrca and to dentfy the role that both the government and the prvate sector can play n ncreasng both the number and the qualy of part-tme jobs avalable. References Barret, G.F.. and Doron, D.J. (2001). Workng part-tme: by choce or by constrant? Canadan Journal of Economcs, 34(4), 1042-4065. Blank, R. (1989). The Role of Part-Tme Work n Women s Labor Market Choces Over Tme. The Amercan Economc Revew. 79(2), 295-299. Buddelmeyer, H., Mourre, G. and Ward, M. (2008). Why Do Europeans Work Part- Tme? A Cross-Country Panel Analyss. European Central Bank Workng Paper No 872. Görg, H. and Strobl, E. (2003). The Incdence of Vsble Underemployment: Evdence for Trndad and Tobago. The Journal of Development Studes. 39(3). 81-100. Hardoy, I. and Schøne, P. (2006). The Part-Tme Wage Gap n Norway: How Large s It Really? Brsh Journal of Industral Relatons, 44(2), 263-282. Hrsch, B.T. (2005). Why Do Part-Tme Workers Earn Less? The Role of Worker and Job Sklls. Industral and Labor Relatons Revew, 44(2), 263-282. Internatonal Labour Organsaton. (1998). Resoluton concernng the measurement of underemployment and nadequate employment suatons, adopted by the 16
Sxteenth Internatonal Conference of Labour Statstcans. Avalable from: http://www.lo.org/publc/englsh/bureau/stat/res/underemp.htm. Leppel, K. and Clan, S.H. (1993). Determnants of voluntary and nvoluntary part-tme employment. Eastern Economc Journal, 19(1), 59-70. Muller, C. (2009). An analyss of the extent, nature and consequences of female parttme employment n post-aparthed South Afrca. Unpublshed PhD thess. Unversy of KwaZulu-Natal. Posel, D. and Muller, C. (2008). Is there evdence of a wage penalty to female part-tme employment n South Afrca? The South Afrcan Journal of Economcs, 76(3), 466-479. Stratton, L.S. (1996). Are Involuntary Part-Tme Workers Indeed Involuntary? Industral and Labour Relatons Revew, 49(3), 522-536. Tlly, C. (1991). Reasons for the contnung growth of part-tme employment. Monthly Labor Revew. March. 10-18. 17
Table1. Wage employment among women n South Afrca, 1995-2006. 1995 1999 2000 2001 2002 2003 2004 2005 2006 Total female wage employment 3 508 (30) 3 662 (37) 3 855 (48) 3 830 (48) 3 758 (44) 3 914 (49) 3 947 (56) 4 129 (56) 4 320 (63) Female part-tme wage employment 405 (12) 503 (16) 612 (22) 506 (20) 456 (16) 520 (19) 479 (20) 553 (22) 557 (25) Proporton of parttme wage employed who are women 60.5 (1.2) 62.5 (1.2) 63.0 (1.3) 64.0 (1.5) 64.2 (1.5) 65.9 (1.5) 64.0 (1.7) 64.9 (1.6) 67.3 (1.7) Involuntary female part-tme wage 138 (7) 256 (10) 193 (9) 152 (8) 142 (8) 166 (10) 164 (12) 166 (12) 163 (12) employment Voluntary female part-tme wage 267 (9) 227 (10) 414 (17) 349 (16) 312 (12) 353 (13) 313 (14) 387 (16) 393 (20) employment Proporton of female part-tme wage employment that s nvoluntary 34.1 (1.5) 53.0 (1.6) 31.8 (1.5) 30.3 (1.7) 31.4 (1.6) 31.9 (1.7) 34.4 (2.0) 30.0 (1.9) 29.4 (2.0) Source: OHS 1995 and 1999; September LFSs: 2000 to 2006. Notes: The data are weghted and counts are n thousands. Standard errors are n parentheses. All employment estmates (total and part-tme) are for ndvduals older than 15 years of age wh wage employment, who reported non-zero workng hours of less than 113 hours a week and for whom earnngs nformaton s not mssng. In 1995 only actual hours worked are avalable. Voluntary and nvoluntary part-tme categores may not sum to total part-tme due to mssng nformaton on the desre to work longer hours. 18
Table 2. Margnal effects estmates from a bnomal prob comparson between nvoluntary and voluntary part-tme wage workers. I II III IV Age 0.006* 0.003 0.005 0.008** (0.003) (0.003) (0.003) (0.003) Age 2 /1000-0.120*** -0.097** -0.113*** -0.150*** (0.038) (0.039) (0.040) (0.041) Job duraton -0.021*** -0.021*** -0.018*** -0.010*** (0.002) (0.002) (0.002) (0.003) (Job duraton) 2 0.367*** 0.363*** 0.316*** 0.148* (0.083) (0.083) (0.083) (0.090) Coloured -0.075*** -0.060*** -0.071*** -0.073*** (0.019) (0.020) (0.020) (0.020) Indan -0.128*** -0.107*** -0.107** -0.083* (0.038) (0.041) (0.042) (0.045) Whe -0.204*** -0.185*** -0.195*** -0.156*** (0.016) (0.017) (0.018) (0.021) Urban 0.114*** 0.105*** 0.087*** 0.083*** (0.013) (0.013) (0.014) (0.014) Prmary educaton 0.018 0.012 0.015 0.024 (0.020) (0.020) (0.020) (0.021) Incomplete secondary educaton 0.071*** 0.064*** 0.061*** 0.070*** (0.022) (0.022) (0.023) (0.023) Matrc or equvalent 0.011 0.007 0.004 0.017 (0.025) (0.025) (0.028) (0.030) Tertary educaton -0.094*** -0.098*** -0.040-0.025 (0.023) (0.023) (0.035) (0.037) Marred/cohabng -0.035** -0.009-0.001 0.008 (0.014) (0.015) (0.015) (0.016) Prevously marred 0.038* 0.037* 0.036* 0.041** (0.020) (0.020) (0.020) (0.021) Number of employed men n the household (ages 16 to 64) Number of employed women n the household (aged 16 to 59 years) Number of unemployed adults n the household Number of chldren younger than 7 years n the household Number of chldren aged 7 to 14 years n the household - - - - - -0.060*** -0.061*** 0.054*** (0.011) (0.011) (0.011) -0.017-0.014-0.018 (0.013) (0.013) (0.013) 0.026*** 0.025*** 0.024*** (0.007) (0.007) (0.007) -0.020*** -0.017** -0.016** (0.007) (0.008) (0.008) 0.003 0.002 0.001 (0.006) (0.006) (0.006) Professonal - - -0.058-0.086 (0.089) (0.086) Techncal and assocated professonal - - -0.003-0.032 (0.090) (0.090) Clerks - - 0.008-0.036 (0.091) (0.088) Sales and servce - - -0.040-0.077 (0.086) (0.083) Fshery - - 0.395*** 0.328** (0.124) (0.142) Craft and related trades - - 0.085 0.037 (0.112) (0.110) Plant and machne operators - - -0.002-0.017 (0.113) (0.114) Elementary occupatons - - 0.051 0.009 (0.096) (0.095) Domestc Servces - - 0.187* 0.160 (0.112) (0.115) 19