Gender pay differentials in the low-paid labour ... - Flinders University

Gender pay differentials in the low-paid labour market Josh Healy, Michael Kidd and Sue Richardson National Institute of Labour Studies Flinders University, Adelaide

Abstract The Australian Fair Pay Commission would like to advance its knowledge of the gender pay gap. To further this understanding we investigate the size, trends in, and determinants of, the difference between male and female pay in Australia. We focus on the differences in hourly pay between adult men and women who are employed in occupations and industries with a high density of Pay-Scale reliant workers. This is where we expect to see the most immediate impact of Commission decisions on the pay gap. We assume that the essential interest of the Commission is the extent to which women get paid less than men because they are women. To answer this question we compare the wages of women with the wages of like men – who have the same observable attributes that might affect productivity/pay – to see if there is a systematic difference. We: • report the broad facts about women’s pay compared with men’s pay, focussing on hourly wage; • show how the share of women in employment and in hours worked has changed between 1978 and 2008 and differs between industries and low paid occupations; • compare the average hourly ordinary-time pay of adult non-managerial men and women, in aggregate and within the industries and occupations of greatest interest to the Commission, and show how these have changed between 2000 and 2006; • answer the question: what would the gender gap in hourly wages have been, if male and female employment in 2006 was distributed (by industries) in the same proportions as were present in 1998?; • identify the proportion of the gender pay gap that can be accounted for by the fact that women and men typically have different productive characteristics; • account for why some industries/occupations have large, while others have small, gender pay differences. Where possible, we examine the gender pay gap separately for those paid award rates, those paid under collective agreements and those paid under individual agreements. The size of the gender pay gap In 2006, female adult non-managerial employees were paid on average 11 per cent less than their male counterparts if they were permanent or contract workers, and 8 per cent less if they were casuals. There is virtually no gender pay gap for those paid award rates. This is not just because award rates leave no scope for paying men and women differently for doing the same job. It also requires that the pay for the sorts of award jobs done by men Author Affiliations: Josh Healy and Sue Richardson are respectively Research Assistant and Professor, National Institute of Labour Studies (NILS) at Flinders University. Professor Michael Kidd is Head of the School of Accounting, Economics and Finance at Deakin University and Research Associate at NILS. Acknowledgement: We express our great appreciation for the skilled and effective support provided by Lulu Sun, in the development and presentation of data.

page 237

Gender pay differentials in the low-paid labour market

is similar to the pay for the sorts of award jobs done by women. For both casual and other forms of employment, the more individual is the pay setting arrangement, the greater is the gender pay gap. In 2006 men employed on permanent/fixed terms were paid 15 per cent more than women on average if they have individual pay setting arrangements – a substantial rise from the 8 per cent in 2000. Over the same period, the female disadvantage more than doubled for casual workers. This provides grounds for the concerns that women do relatively badly out of the bargaining (as compared with the award) system. The occupations with the greatest concentrations of people paid award rates (comprising 68 per cent of all such employees) are ‘Intermediate clerical, sales and service workers’, ‘Elementary clerical, sales and service workers’ and ‘Labourers and related workers’. The gender hourly pay gap varied from 11 to 14 per cent among ‘permanent’ employees in these occupations and was much less among casuals – as little as 2 per cent for intermediate clerical and sales workers. Within industries and occupations, casual workers systematically have smaller pay gaps than do their permanent equivalents. But only in a few cases are women on average paid more than men, most notably casuals in retail. The dominant payment system for casual workers is the award safety net, setting pay directly for 42 per cent of casual employees in 2006. However, the award system has been shrinking in importance since 2000, when 46 per cent of casuals were safety net reliant. This decline is of concern for the future trajectory of the gender pay gap. Changes in the gender composition of the workforce Men are still more likely than women to be in the paid labour force, but the participation gap between the sexes has narrowed considerably between 1978 and 2008. In both the prime (25–54) and older (55–64) working age groups, women’s participation has risen substantially (by 23 and 28 percentage points respectively) while for men it fell (by 5 and 6 percentage points respectively). Much of the growth in women’s employment has been in the four ‘low pay’ industries of principal interest to the Commission: retail, accommodation, property and health services. Of the 1.4 million total new jobs created for women from 1994–2008, nearly two-thirds (64 per cent) were created in one of these industries. The comparable proportion for men was 46 per cent. These four industries thus represent especially important sites of recent employment growth for both sexes. The industry pattern of growth is similar if we examine hours worked, rather than number of people employed. The changes in employment composition over the 1998–2006 period, including the movement of women into low-paid sectors, have increased the gender pay gap. However, the overall effect is small. Do women with the same productive characteristics as men get paid the same? Women might be paid less than men because they are less productive – i.e. have less human capital. Or they might be paid less because employers reward the human capital they do have less generously than they reward men’s equivalent human capital. We find that women with the same productivity characteristics as men (such as tenure in current job, education, years of work experience, form of employment) on average earn around $3.70 or 13 per cent less than men per hour. In a more sophisticated analysis, we allow for the possibility that the returns to personal and human capital attributes might differ according to sex. We address the hypothetical question of what female wages would look like if women received the same returns to page 238


their characteristics as do men. The method decomposes the gender wage gap into two components: that which is ‘explained’ by gender differences in productive characteristics, and the ‘unexplained’ residual which measures the extent to which similar characteristics of females and males are valued differently by their employers. For the economy as a whole, differences in male–female characteristics explain only a small percentage of the overall wage gap (around 11 per cent of the gap). Interestingly, the proportion explained rises to 27 per cent – well over double – when we account for the industry of employment. This suggests that the industries in which women are employed tend to be lower paid. There is substantial variation in the gender wage gap across industries, ranging from 5 per cent in accommodation to more than 17 per cent in property. In property and health, our specifications account for upwards of 60 per cent of the observed gender gap in wages. In retail trade we explain only 30 per cent, and in accommodation only 20 per cent. The dominant pattern here is that the industries with smaller overall gender wage gaps (i.e. retail and accommodation) also have the smallest proportion explained by genderspecific differences in human capital. One interpretation of this result is that industries with a strong award structure successfully limit the size of the gender wage gap but also decrease the wage variance and the consequent impact of human capital. In property and health, where the gender wage gap is larger, the human capital characteristics explain a much larger proportion of the overall gap. These differences by industry and occupation highlight an important feature of the low-paid labour market, in that there are generally smaller differences between male and female wages in the sectors where award reliance is high. But the gender differential is only one of several important dimensions of earnings inequality. In the lowest-paid sectors, the problem of inequality manifests less in the specific form of gender disparities, and more in the form of a distribution which is highly-skewed towards low hourly wages. While employees remain within these industries their prospects of attaining better-paying jobs are curtailed by the very small number of such jobs on offer. Male and female wages may be more closely aligned in these sectors, but only because both sexes are disadvantaged in these sectors relative to most other Australian employees.

1. Introduction This report provides the results of a research project investigating the size, trends in, and determinants of, the difference between male and female earnings in Australia – commonly referred to as the ‘gender pay gap’. The research was commissioned by the Australian Fair Pay Commission, which is charged with the maintenance and adjustment of minimum wages for employees in the federal industrial system. As such, this report gives particular attention to the gender pay gap within lower-paid industries and occupations, where decisions of the Commission have their largest and most direct impacts on earnings and employment. These areas include industries such as retail trade and hospitality, and occupations such as elementary services, cleaning and manual labouring. Our first task is to understand the size of the contemporary gender pay gap and how it has been changing over time. Section 2 uses unpublished data from the ABS Survey of Employee Earnings and Hours (EEH) to document these patterns. The EEH data is based on a large survey of employers’ payroll records and thus provides hourly wage information of the highest quality. We focus on the gender difference in hourly wages rather than a weekly or annual earnings dimension. This allows us to directly control for hours of work page 239


and thus enables the inclusion of data from employees working part-time or on a casual basis, whose earnings are lower due to shorter working hours. We restrict attention in this section to adult, non-managerial employees, removing from the analysis employees on junior rates of pay (those under 21 years of age) and employees with significant responsibility for the supervision and management of other staff in their workplace. The exclusion of juniors/managerial employees will tend to raise/lower the average wage (for both sexes). The data we have requested for this project allow for further control over two aspects of employment potentially related to the gender pay gap: casual status and the method of setting pay. ‘Casual’ employees are those without paid leave entitlements, who typically receive loadings on their base hourly rates of pay, as a form of compensating wage differential. If these loadings are not controlled for, they will confuse the hourly wage comparison in a manner that potentially causes an under-statement of the gender pay gap, since women are more likely than men to hold casual jobs (a fact we document in Section 4). We also differentiate three groups of employees according to their method of setting pay: ‘awardonly’ employees, whose wages are set at exactly the minimum rates determined by decisions of the Australian Fair Pay Commission (AFPC); and employees covered by either collective or individual wage-setting agreements (including both registered and unregistered forms). This distinction between pay-setting methods, in addition to a disaggregation in some data tables by industry and occupation, allows us to look more closely at the relationship between female and male wages in parts of the labour market where AFPC decisions have their greatest effect. We also utilise the EEH data to explore changes in the gender pay gap over time, during the period 2000 to 2006. The scope of this analysis is constrained by the fact that the ABS has only collected the ‘methods of setting pay’ data since 2000 (and the results of the 2008 survey are not yet available). We can extend the analysis back a further two years, to 1998, by dropping the pay-setting variable. However, prior to this there are no separate estimates for casual employees, and the concerns we have raised about the distorting effects of casual loadings come into play. The period for which we can construct an accurate trend of the female–male hourly wage gap is therefore limited to the start and end years 1998 and 2006; or to 2000 as the start year where we include a disaggregation by pay-setting method. Male-female relative wages are determined by many factors other than the legal minimum conditions set by the AFPC. One potentially important determinant is the change over time in female labour force participation together with the rise in quality of their human capital (relative to men’s). Section 3 of this report is accordingly focussed on the historical course of female labour force participation and employment. We are able to document the trend in female participation (the number of women working or looking for work, as a share of the civilian population) for a longer period than that used to estimate female employment, because the participation trend is unaffected by changes in the classification of industries and occupations. In Section 3 we report the trend in female participation, by age, over the 30-year period 1978 to 2008. Our report on the trends in female employment is constrained by definitional changes. In examining female employment by industry sector, our reference period is 1994 to 2008. If we consider the trend by occupation, however, the analysis commences with 1997 data. A longer time trend is obviously useful for detecting more persistent and consequential changes, but we also need to relate any observed trends back to the groups for which we have reliable, recent wage estimates. We are interested particularly in estimating the share of increased female participation ‘absorbed’ by employment growth in the low-wage sectors of special interest to the AFPC. To improve our understanding of this growth, we consider both the number of women employed, and the number of hours worked.

page 240


Although we are interested in the expansion of female employment as a background to the trend in female earnings, we ultimately want to gain some perspective on how the changes in female participation have affected the gender pay gap. We utilise the ABS Labour Force Survey (LFS) to document female participation and employment trends. The major advantage of this source is the large sample size, the drawback is the absence of regular data on wages. To link the employment trend back to the wage trend, we need to return to the unpublished EEH data described earlier. Our requested dataset includes information about employment composition, in addition to wages, for the same population of adult, non-managerial employees. This allows us to undertake a basic form of ‘shift/share’ analysis for the period 1998 to 2006. We have observations of actual female employment and hourly wages disaggregated by industry for both 1998 and 2006. In the analysis we report in Section 3, we ask the counterfactual question: what would the gender gap in hourly wages be, if male and female employment in 2006 was distributed (by industry) in the same proportions as present in 1998? This analysis gives us a crude understanding of how far changes in employment composition can explain the changing gender pay gap. The approach is crude because we assume that employment and hourly wages are independent. We apply the actual hourly wage data for 2006 to a synthetic employee cohort, based on the actual composition (by industry) in 1998. The final stage of our analysis involves controlling for the different attributes of male and female employees at the individual level, using confidentialised unit record data from the 2005 ABS Survey of Education and Training (SET). In this analysis we are able to directly account for an array of human capital variables – including educational qualifications, labour force experience and current job tenure. This approach provides a better approximation of employee skill level than our earlier aforementioned analysis which implicitly uses occupation as a proxy for skill. The centrepiece of this analysis involves a decomposition of the overall gender wage gap into a component attributable to male–female differences in human capital skills and the remainder traditionally interpreted as indicative of the level of gender discrimination. A number of caveats associated with this interpretation are documented below.

2. The facts on gender pay gaps 2.1

A snapshot for 2006

We first explore whether men and women with the same employment status were paid differently under the three main pay-setting systems operating in 2006. Table 2-1 shows the relevant estimates from the May 2006 EEH survey. The first two columns show the raw estimates of average hourly ordinary-time earnings (AHOTE) for women and men. We then subtract the male estimate from the female estimate to obtain the dollar differences in the third column. The final column on the right of the table divides the female estimates by the male estimates, and multiples the result by 100, to obtain the percentage relativities. These represent our main measure of the ‘gender pay gap’. When the relativity is less than 100, women are paid less than men, and the difference from 100 provides an estimate of the percentage gap in average hourly pay. When the relativity exceeds 100, women enjoy a wage premium over men in similar employment. The major result shown in Table 2-1 is that there was little or no gender pay gap among employees reliant on award rates of pay in 2006. In the ‘non-casual’ (permanent and fixed term) labour market, the average hourly wage of women was 2.7 per cent higher than the male average (a relativity of 102.7). In the casual labour market, the relativity was 99 per cent, implying a mild (1 per cent) hourly wage disadvantage for women. The page 241


outcomes for women, relative to men, are less encouraging in sectors where pay is set as a result of bargaining above the award minima (or outside the safety net in sectors not covered by awards). On average, women’s wages were between 4 and 8 per cent less than men’s if their pay was set through collective agreement, and between 14 and 15 per cent less if their pay was set through individual arrangement. Differences of this size imply that concerns about inferior female relative wages in the bargaining system are well-founded. However, these concerns relate to the gap between female and male wages within different pay-setting systems, not to the absolute values of the wages paid. The first two columns of Table 2-1 show that although the gender pay gap is larger under both forms of bargaining, average female wages are also higher than in the award-only sector. Negotiation above the safety net therefore benefits both sexes in absolute terms, but delivers the greatest relative benefits to men. Table 2-1: Average hourly ordinary-time earnings (AHOTE), by sex, casual status and method of setting pay in 2006 Female AHOTE ($)

Male AHOTE ($)

Female/male $ difference

Female/male % relativity

Award only

19.30

18.80

0.50

102.7

Collective agreement

25.90

28.10

–2.20

92.2

Individual arrangement

23.60

27.80

–4.20

84.9

All pay-setting methods

24.10

27.00

–2.90

89.3

Award only

19.50

19.70

–0.20

99.0


24.20

25.30

–1.10

95.7


20.30

23.60

–3.30

86.0


21.00

22.70

–1.70

92.5

Permanent/fixed term

Casual

Source: Unpublished data, ABS Survey of Employee Earnings and Hours (EEH), May 2006.

Table 2-2 provides a similar arrangement of the data, but replaces the disaggregation by pay-setting method in Table 2-1 with a disaggregation by occupations, based on the first (‘major group’) level of the Australian Standard Classification of Occupations, 1996 (ASCO). We provide selected information for the three occupations with the highest concentrations of award-only workers in May 2006. These are: ‘Intermediate clerical, sales and service workers’, ‘Elementary clerical, sales and service workers’ and ‘Labourers and related workers’. The estimates we obtained in a separate part of our EEH data request indicate that these three occupations accounted for 68 per cent of adult, non-managerial employees paid by award-only in May 2006. This proportion is composed of 58 per cent of award-only workers in non-casual jobs, and 80 per cent of award-only workers in casual jobs. By focussing on these three key occupations, we therefore provide detailed descriptive evidence about the size of the gender pay gap in parts of the workforce principally affected by wage-setting decisions of the Australian Fair Pay Commission. The estimates in Table 2-2 document that women in non-casual jobs were paid between 11 and 14 per cent less than men in similar jobs, within the three key occupations in May 2006. Female relative wages were not noticeably better or worse (relative to men) in these occupations than on average across the non-casual workforce, where the gender pay gap was 11 per cent. The gap is generally narrower among casual workers (7 per cent) and is narrower still (2 to 3 per cent) among casual workers in the intermediate and elementary clerical, sales and service occupations where award-only employment is prevalent. The female disadvantage is comparatively large (in dollar and percentage terms) among the page 242


15 per cent of casual employees who were working as labourers and related workers in 2006. The wage differential in this occupation may be due to finer distinctions between male and female labouring jobs, which we cannot control for at this level (i.e., the 1-digit occupation level) of comparison. Table 2-2: Average hourly ordinary-time earnings (AHOTE), by sex, casual status and selected occupations (1-digit ASCO) in 2006 Female AHOTE ($)

Male AHOTE ($)



Intermediate clerical, sales and service workers

20.30

23.50

–3.20

86.4

Elementary clerical, sales and service workers

17.50

19.70

–2.20

88.8

Labourers and related workers

17.20

20.00

–2.80

86.0

All occupations

24.10

27.00

–2.90

89.3


19.90

20.40

–0.50

97.5


18.50

19.10

–0.60

96.9


18.30

20.10

–1.80

91.0

All occupations

21.00

22.70

–1.70

92.5

Permanent/fixed term

Casual


Table 2-3 and Table 2-4 extend the comparison further, by looking at the gap in hourly pay for men and women in similar occupations and in similar industries. We again restrict attention to those sectors with high levels of aggregate award reliance. The industries on which we concentrate are: ‘Retail trade’, ‘Accommodation, cafés and restaurants’, ‘Property and business services’ and ‘Health and community services’. Together these four industries contained 68 per cent of (adult non-managerial) award-only employees in May 2006, the proportion made up of 63 per cent of non-casual award workers and 74 per cent of casual award workers. Within these four industries, we compare female and male average hourly wages for employees in the three occupations represented in Table 2-2. By drilling down to key occupations within our four main industries, we are better able to control for skills differences, and provide an even tighter focus on those employees directly dependent on Commission decisions. Table 2-3 contains data for non-casual employees. Table 2-4 provides comparable data for casuals. Looking only at the industry wage comparisons (ignoring the occupational breakdown momentarily), we see a wide diversity of outcomes in the gender pay gap in 2006. In accommodation, the industry in which employees are most likely to be paid by award only, the gender gap is negligible (1 per cent for casuals and non-casuals alike). But in Property and business and Health and community services (which together include more than one-third of award workers), the gender pay gaps are close to 20 per cent among non-casuals, and 15 per cent among casuals. Part of the reason for these differences at the industry level is that the property and health sectors, despite containing significant proportions of all award-reliant workers, nonetheless have many higher-paid workers who draw out the intra-industry distribution of pay. Our EEH proportions data tell us that, for instance, half of the employees in the health industry were professionals or associate professionals in 2006. The comparable proportion for the accommodation industry is just page 243


12 per cent. We therefore need the occupational breakdown within the industry-level comparison, to ensure we are focussing on the workers likely to benefit from wage-setting decisions of the AFPC. When we include the occupational dimension, the gender pay gaps are in general narrower than those observed for the industries as a whole. In the property sector, the female–male wage relativity is consistently narrower in award-reliant occupations than in the sector as a whole, irrespective of casual status. In health, the improvements are also evident, with the exception of casuals in elementary clerical, sales and service jobs, for whom the female relative wage is below the industry average. In all other cases, the gender pay gap is narrower for the key occupation groups within these industries compared to the industries overall. Bearing in mind that the inclusion of occupations brings us closer to a focus on award-dependent employees, the results in Table 2-3 and Table 2-4 generally support the earlier conclusion – that male and female wages are more closely matched in sectors of the labour market where the Australian Fair Pay Commission has greatest influence over how and where actual wages are set. Table 2-3: Average hourly ordinary-time earnings (AHOTE) for permanent and fixedterm employees, by sex, for selected occupations within selected industries (1-digit ANZSIC) in 2006 Permanent/fixed-term employees only

Female AHOTE ($)

Male AHOTE ($)




18.40

20.30

–1.90

90.6


16.30

18.00

–1.70

90.6


16.60

16.40

0.20

101.2

All occupations

17.50

20.00

–2.50

87.5

Retail trade

Accommodation, cafes and restaurants Intermediate clerical, sales and service workers

18.60

18.50

0.10

100.5


16.20

17.30

–1.10

93.6


16.50

17.40

–0.90

94.8

All occupations

18.60

18.80

–0.20

98.9


19.30

22.80

–3.50

84.6


18.60

19.80

–1.20

93.9


15.70

17.70

–2.00

88.7

All occupations

23.60

28.90

–5.30

81.7


19.20

22.40

–3.20

85.7


18.70

21.30

–2.60

87.8


17.80

16.00

1.80

111.3

All occupations

24.90

31.10

–6.20

80.1

Property and business services

Health and community services


page 244


Table 2-4: Average hourly ordinary-time earnings (AHOTE) for casual employees, by sex, for selected occupations within selected industries (1-digit ANZSIC) in 2006 Casual employees only

Female AHOTE ($)

Male AHOTE ($)




18.00

18.00

0.00

100.0


18.40

18.40

0.00

100.0


17.30

16.10

1.20

107.5

All occupations

18.40

19.00

–0.60

96.8

Retail trade

Accommodation, cafes and restaurants Intermediate clerical, sales and service workers

18.90

19.60

–0.70

96.4


18.20

18.30

–0.10

99.5


18.40

18.70

–0.30

98.4

All occupations

18.90

19.10

–0.20

99.0


20.00

22.10

–2.10

90.5


19.70

20.10

–0.40

98.0


18.40

21.10

–2.70

87.2

All occupations

21.00

24.80

–3.80

84.7


20.10

22.30

–2.20

90.1


21.10

28.60

–7.50

73.8


19.10

19.50

–0.40

97.9

All occupations

24.50

28.80

–4.30

85.1




2.2

Changes over time

We now turn to the issue of whether the snapshot of data from 2006 is representative of the gender pay gap over the longer time period 2000 to 2006. We focus on the gap in average hourly wages between women and men with the same casual status and the same method of setting pay in 2000, 2002, 2004 and 2006. The underlying data are taken from the ABS EEH surveys conducted in May of each of these four years, and the estimates refer (as earlier) to adult, non-managerial employees only. Our interest here is whether – and, if so, for whom – the gender pay gap has been changing since 2000. The estimates we provide indicate the percentage relativity between female and male average hourly wages (the same estimates as reported in the right-hand columns of the four preceding data tables). These estimates summarise the gender pay gap at the four different points in time for which we have the necessary data. In interpreting these data, the reader should note that the EEH survey was not expressly designed to provide timeseries comparisons. There is a discontinuity between the 2000 and 2002 data that we have obtained, in that the amounts salary sacrificed by employees are not counted as part of ordinary-time earnings in 2000, but are in the 3 later years. We consider this break to

page 245


be of relatively minor importance for the current analysis, as we are comparing the ratios of female to male wages over time, rather than wage levels. The results of the analysis are shown in Figure 2-1. We see that in the non-casual labour market, the gender pay gap has widened slightly between 2000 and 2006, increasing from 8 to 11 per cent. However, this general pattern conceals substantial differences between the three main pay-setting groups. In 3 of the 4 years for which we have data, award-only women have maintained a pay advantage on average over men in the same group. The situation in the award-only sector has not changed noticeably over time. The gap among non-casual workers covered by collective agreements has also been stable: a female disadvantage of 8 or 9 per cent in all 4 years. In contrast, there is evidence of a significant widening in the gender pay gap among non-casual workers paid by individual arrangements. The female disadvantage increased from 8 to 15 per cent in this sector, between 2000 and 2006. The change for workers covered by individual arrangements has been the main driver of the fairly modest increase in the gender pay gap observed among all non-casual employees. The female–male relativity has also deteriorated in the casual labour market. Over all methods of setting pay, the change has been relatively small, a gap of 5 per cent in 2000 rising to 7 per cent in 2006. Again this overall trend masks considerable variation across the pay-setting groups. The female disadvantage has narrowed almost to zero in the award-reliant sector; has increased moderately (although with some volatility in the estimates) for workers with collective agreements; and has increased dramatically (more than doubled) for workers covered by individual pay-setting arrangements. The dominant payment system for casual workers is the award safety net, setting pay directly for 42 per cent of casual employees in 2006. However, the award system has been shrinking in importance since 2000, when 46 per cent of casuals were safety-net reliant. This decline is of concern for the future trajectory of the gender pay gap, given the evidence in Figure 2-1 of a persistently narrower gap in the award-only sector. Figure 2-1: Gender pay gap (%), by casual status and method of setting pay, various years (May 2000 to May 2006) Female/male relativity (%) 110

100

90

80

70 Permanent/ fixed term Award only

Permanent/ fixed term Collective agreement

2000

Permanent/ fixed term Individual arrangement

2002

Permanent/ fixed term All pay-setting methods

Casual

Casual

Casual

Casual

Award only




2004

2006

Note: Estimates for 2000 exclude amounts salary sacrificed, while subsequent years include these amounts. See discussion in text.

page 246


3. Trends in female labour supply 3.1

Participation and employment

We have commented briefly in the preceding discussion on the relationship between employment composition and the gender pay gap. In this section we provide a fuller discussion, beginning with the long-term trend in labour force participation, by gender. The participation rate represents the number of persons either currently working or looking for work, expressed as a proportion of the total civilian population (in this case by age). Figure 3-1 shows the participation rate in the period 1978 to 2008, for men and women in two age groups: 25–54 years and 55 years and over. We separate these groups so as to distinguish persons of ‘prime’ working age from others who are nearing or already in transition to retirement. There are remarkable differences between the sexes in both age groups. In 1978, 95 per cent of prime-age men were in the paid labour force. Men were almost twice as likely as women to be participating in 1978. In the intervening 30 years, the male participation rate has stagnated and declined, while the female rate has grown strongly, particularly in the 1980s. By 2008, three-quarters (75 per cent) of prime-age women were in the labour force, compared to just over half (52 per cent) in 1978. This constitutes a massive increase in the effective female labour supply. For older age groups, the pattern of declining male participation, and increasing female participation, is also observed. Participation rates for persons aged 55 years and older doubled between 1978 and 2008 for women (to 24 per cent), but fell by 5 percentage points for men (to 39 per cent). Men are still more likely than women to be in the paid labour force, but the participation gap between the sexes has narrowed considerably in the past three decades. Figure 3-1: Long-term trend in labour force participation rates, by sex and age group, 1978 to 2008 Participation rate (%) 100 Males 25–54 years

80 Females 25–54 years

60

40 Males 55+ years

20 Females 55+ years

0 1978

1983

1988

1993

1998

2003

2008

Source: ABS Labour Force Survey, cat. no. 6291.0.55.001 (LM1), Detailed - Electronic Delivery, May 2008. Original series.

How successfully has the substantial additional supply of women, shown in Figure 3-1, been met by expanding demand? To answer this question we explore the composition of female employment, by industry, for the period 1994 to 2008. This is as far back as the data allow us to go – in 1993 the ABS adopted a new Australian and New Zealand Standard Industry Classification (ANZSIC), causing a break in the industry estimates. Our primary focus is upon the proportion of total female employment absorbed by the four key industries (retail, accommodation, property and health services) detailed earlier. There are, of course, page 247


other ways of examining the matching of supply and demand, including the unemployment and under-employment rates of women. We concentrate on employment composition because it provides the strongest link to the wage rates established in low-paid industries by the AFPC. Table 3-1 summarises the change in industry employment composition over the 1994 to 2008 period. The first panel of the table shows, for men and women separately, the numbers employed in 1994 and in 2008, and the change (growth) between these 2 years, in the four industries of special interest to the Commission, and in total across all industries (including those not separately listed in the table).1 We also estimate the share of total employment that was located in the four key industries at each point in time, and the share of total employment growth that occurred in these industries. Employment in the four low-paid industries expanded by 597,000 jobs (or 46 per cent) for men, and by 894,000 jobs (53 per cent) for women, over the 1994 to 2008 period. Over the same interval, total employment in all industries grew by 29 per cent for men, and 41 per cent for women. Since jobs growth in the low-paid industries exceeded the total growth (for both sexes), their share of employment increased, from 29 to 32 per cent for men, and from 49 to 54 per cent for women. A much larger share of total female employment is found in the four low-paid industries listed in Table 3-1 than is the case for male employment. We also see that a larger proportion of the total employment growth for women in the 1994–2008 period was absorbed by these four industries. Of the 1.4 million total new jobs created for women in this period, nearly two-thirds (64 per cent) were created in one of the four industries of interest to the Commission. The comparable proportion for men was 46 per cent. These four industries thus represent especially important sites of recent employment growth for both sexes. The bottom panel of Table 3-1 repeats the analysis using total hours worked (rather than number working) as the measure of employment. This change makes little significant difference to the overall conclusion. The four low-paid industries still contain a larger share of female employment compared to male, and absorbed a greater proportion of the growth in female (relative to male) employment between 1994 and 2008. The changes reflected in Table 3-1 make it likely that the patterns of female employment growth are pertinent to the size of the gender pay gap, since so much of this growth has occurred in industries which are lower-paid, and disproportionately affected by minimum wage adjustments.

1

In Table 3-1, our measures of ‘employment’ are based on the broadest ABS definition, and are not restricted to adults or to employees only, as is the case in our estimates from the EEH survey.

page 248


Table 3-1: Share of total employment and total hours worked located in low-paid industries, by sex, 1994 to 2008 Male 1994

Male 2008

Male change

Female 1994

Female 2008

Female change

Retail trade

574

751

177

605

804

199

Accommodation, cafés and restaurants

161

228

67

207

289

82


401

683

283

323

590

266


165

237

71

542

889

347

Total in low-paid industries

1,301

1,899

597

1,677

2,571

894

Total in all industries

4,561

5,863

1,302

3,394

4,795

1,401

28.5

32.4

45.9

49.4

53.6

63.8

22,652

25,508

2,857

15,736

20,501

4,765

6,095

8,055

1,960

5,613

8,062

2,448


16,936

27,093

10,157

9,952

18,260

8,309


6,449

8,572

2,123

15,397

25,503

10,106

Employment ’000s

% Of total in low-paid industries Hours worked ’000s Retail trade Accommodation, cafés and restaurants

Total in low-paid industries Total in all industries

52,131

69,228

17,097

46,698

72,326

25,628

186,968

228,756

41,788

99,629

140,980

41,351

27.9

30.3

40.9

46.9

51.3

62.0

% of total in low-paid industries

Source: ABS Labour Force Survey, cat. no. 6291.0.55.003 (E03_Aug 94), Detailed, Quarterly, May 2008.

3.2

Shift/share analysis

We now test directly the proposition that changes in employment composition affect the measured gender pay gap, focussing on the period 1998 to 2006. We use these two points in time for the comparison because they represent the longest span for which we have reliable data on both employment composition and average hourly wages (split by casual status, to control for loadings) from the EEH survey. Our aim is to create a ‘synthetic’ malefemale wage structure in 2006 by applying the observed industry-level proportions of employees from 1998 to the average hourly wages for those industries in 2006. We then compare the synthetic male and female wages for 2006 to the actual wages observed by the ABS at that time, to determine the direct impact of changes in employment composition on the gender pay gap. Table 3-2 and Table 3-3 show the relevant data for adult non-managerial employees of each sex. The first two columns of these tables show estimates of the proportions of men and women in each of the major industries, split by casual status, in 1998 and 2006. In 1998, for instance, 19.8 per cent of all adult non-managerial male employees were in non-casual jobs in the manufacturing industry. This proportion fell to 14.5 per cent by 2006. The next two columns of Table 3-2 and Table 3-3 show average hourly wages for each combination of industry and casual status in both survey periods. The data in the column second from right are obtained by multiplying the 2006 proportions by the 2006 wage estimates. These indicate the actual wage structure in 2006. Summing all of these estimates, we obtain a male average wage in 2006 of $26.34 per hour, and a female average wage of $23.54 per hour. This translates to an actual gender pay gap of $2.80 per hour, or 11 per cent, in 2006.

page 249


We repeat this exercise in the final (right-hand side) column of Table 3-2 and Table 3-3 but replace the 2006 employment proportions with the 1998 proportions. From these, we derive the synthetic wage structure – what the average wage would have been in 2006 if employment composition had remained unchanged from 1998. Summing down the rows, we calculate an implied male average wage of $26.28 per hour, and an implied female average wage of $23.57 per hour. These estimates are not very different from the actual wage structure. The male wage is reduced marginally, and the female wage is slightly higher. In consequence, the gender pay gap falls from 11 to 10 per cent. We conclude that the changes in employment composition over the 1998–2006 period, including the movement of women into low-paid sectors as indicated in Table 3-1, have increased the gender pay gap, relative to what it would have been if the employment composition from 1998 had remained. However, the overall effect is small. Over a longer period, we would expect the differences to be more pronounced, but we do not have the data necessary to test this hypothesis formally because of changes in the survey instruments affecting how casual employment status is determined.

page 250


Table 3-2: Shift-share analysis of 2006 male wages, using 1998 composition

Male employees only

AHOTE in AHOTE in % in 1998 % in 2006 1998 2006

2006 % & 2006 AHOTE

1998 % & 2006 AHOTE

Permanent/fixed term B. Mining

2.2

2.3

28.60

38.40

0.88

0.84

19.8

14.5

17.30

24.80

3.60

4.91

D. Electricity, gas and water supply

1.5

1.3

21.80

33.40

0.43

0.50

E.

Construction

5.8

7.7

19.30

26.30

2.03

1.53

F.

Wholesale trade

6.5

6.7

17.60

24.90

1.67

1.62

G. Retail trade

7.0

9.2

14.40

20.00

1.84

1.40

H. Accommodation, cafés and restaurants

2.3

1.8

13.90

18.80

0.34

0.43

I.

Transport and storage

6.2

5.5

18.30

27.10

1.49

1.68

J.

Communication services

2.7

1.7

22.20

32.00

0.54

0.86

K.

Finance and insurance

2.6

3.2

21.70

33.90

1.08

0.88

L.


7.5

10.7

20.20

28.90

3.09

2.17

M. Government administration and defence

6.0

5.3

19.00

27.40

1.45

1.64

N. Education

5.6

4.6

23.20

32.80

1.51

1.84

O. Health and community services

3.9

4.5

21.00

31.10

1.40

1.21

P.

1.4

1.5

18.50

26.80

0.40

0.38

3.2

3.7

19.60

26.60

0.98

0.85

B. Mining

0.1

0.1

19.60

38.80

0.04

0.04

C. Manufacturing

1.5

1.4

15.00

20.60

0.29

0.31


0.0

0.0

15.30

22.80

0.00

0.00

E.

Construction

0.9

1.3

16.50

23.70

0.31

0.21

F.

Wholesale trade

C. Manufacturing

Cultural and recreational services

Q. Personal and other services Casual

0.5

0.5

14.50

19.30

0.10

0.10

G. Retail trade

2.3

2.1

14.30

19.00

0.40

0.44


1.9

1.9

14.80

19.10

0.36

0.36

I.


0.9

1.6

16.20

22.30

0.36

0.20

J.


0.1

0.0

14.90

18.40

0.00

0.02

K.


0.0

0.2

15.60

24.10

0.05

0.00

L.


3.4

3.4

18.20

24.80

0.84

0.84


0.2

0.2

16.50

23.80

0.05

0.05

N. Education

1.0

1.1

23.30

34.60

0.38

0.35


0.9

0.5

19.50

28.80

0.14

0.26

P.

0.9

0.9

17.90

20.60

0.19

0.19

0.9

0.5

15.80

19.90

0.10

0.18

100.0

100.0

18.70

26.50

26.34

26.28


Q. Personal and other services T.

All industries

Source: Unpublished data, ABS Survey of Employee Earnings and Hours (EEH), 1998; 2006.

page 251


Table 3-3: Shift-share analysis of 2006 female wages, using 1998 composition

Feale employees only

AHOTE in AHOTE in % in 1998 % in 2006 1998 2006

2006 % & 2006 AHOTE

1998 % & 2006 AHOTE

Permanent/fixed term B. Mining

0.3

21.40

30.10

0.09

0.09

C. Manufacturing

6.2

4.5

15.00

22.50

1.01

1.40


0.3

0.3

19.70

27.00

0.08

0.08

E.

Construction

1.0

1.4

15.40

21.40

0.30

0.21

F.

Wholesale trade

3.1

3.5

15.60

21.70

0.76

0.67

G. Retail trade

7.2

7.8

13.40

17.50

1.37

1.26


2.2

1.9

12.90

18.60

0.35

0.41

I.


2.1

2.0

17.20

22.80

0.46

0.48

J.


1.3

1.0

19.50

27.90

0.28

0.36

K.


4.8

4.5

16.60

25.40

1.14

1.22

L.


8.4

10.1

16.60

23.60

2.38

1.98


4.6

5.5

18.70

26.80

1.47

1.23

N. Education

11.5

11.9

20.30

29.30

3.49

3.37


17.7

17.5

17.70

24.90

4.36

4.41

1.2

1.3

17.50

23.10

0.30

0.28

2.4

2.8

15.30

22.40

0.63

0.54

B. Mining

0.0

0.0

19.50

26.20

0.00

0.00

C. Manufacturing

1.5

1.1

14.30

19.20

0.21

0.29


0.0

0.0

18.40

23.40

0.00

0.00

E.

Construction

0.2

0.3

15.00

18.70

0.06

0.04

F.

Wholesale trade

P.


Q. Personal and other services Casual

0.8

0.8

13.90

18.20

0.15

0.15

G. Retail trade

4.9

4.6

13.60

18.40

0.85

0.90


4.0

3.3

14.20

18.90

0.62

0.76

I.


0.3

0.6

14.90

20.10

0.12

0.06

J.


0.1

0.1

15.20

18.10

0.02

0.02

K.


0.3

0.2

16.60

19.40

0.04

0.06

L.


3.6

3.7

15.80

21.00

0.78

0.76


0.5

0.5

16.60

24.60

0.12

0.12

N. Education

2.9

2.5

23.40

30.90

0.77

0.90


4.6

3.8

17.10

24.50

0.93

1.13

P.

1.2

1.1

14.20

20.00

0.22

0.24

0.9

1.0

14.50

18.80

0.19

0.17

100.0

100.0

16.80

23.70

23.54

23.57


Q. Personal and other services T.

All industries

Source: Unpublished data, ABS Survey of Employee Earnings and Hours (EEH), 1998; 2006.

page 252


4. Econometric analysis The final section of our report explores how differences in individual attributes, such as education and tenure, contribute to explaining the average gender pay gap. We utilise the Expanded version of the ABS Survey of Education and Training 2005 CURF via the Remote Access Data Laboratory (RADL), a secure online data facility to which approved users submit program code. The individual unit record data enable estimation of multiple regression models based on the human capital framework. This enables us to estimate the size of the gender wage gap, traditionally summarised by the male–female difference in mean log-wage, while controlling for many productivity-related characteristics, such as education and labour market experience. The use of the SET CURF data provides a complementary analysis to that described above, allowing us to control for many characteristics not included in the EEH survey. The analysis is divided into three main components. First, an overview of the key variables included in the regression model is provided, together with summary descriptive statistics. The second component discusses estimation and interpretation of the human capital based wage specification. Results begin with a specification pooled across the male and female samples, followed by models estimated separately by gender. The final component involves a traditional Blinder/Oaxaca analysis, which decomposes the overall gender wage gap into a part attributable to gender differences in productivity as captured by human capital (the explained component) and the residual often interpreted as indicative of the level of discrimination (the unexplained component). This application sheds light on the hypothetical question: what hourly pay would a ‘typical’ female earn in the absence of discrimination? 4.1

Description of key variables and summary statistics

Table 4-1 provides summary statistics (means) for all variables used in the econometric analysis. The dependent variable is the logarithm of the hourly wage, where the hourly wage is calculated by dividing usual weekly earnings by usual weekly working hours for employees in their current main job. We delete from the sample those employees with zero earnings or hours, including those receiving payment in kind. The remaining variables listed in Table 4-1 constitute the independent variables in our econometric analysis. These variables attempt to control for a fairly standard set of productivity-related attributes. The means shown for the set of industry dummy variables at the bottom of Table 4-1 indicate the distribution of our male and female samples across the various industry sectors. Manufacturing is omitted from this list as it constitutes the base or reference category against which the other industry coefficients are compared in our econometric results. To maintain consistency with the EEH survey data, which are for non-farm employees, we have deleted from the SET sample employees in the 1-digit ANZSIC category ‘Agriculture, forestry and fishing’.

page 253


Table 4-1: Unweighted means for variables included in the econometric analysis, by sex

Hourly wage Natural logarithm of hourly wage (DV)

Male

Female

25.375

21.674

3.121

2.989

Demographics Potential labour market experience in years

21.982

21.876

Potential experience in years squared (/100)

6.243

6.220

Tenure in current job in years

6.387

5.745

Tenure in current job in years squared (/100)

0.846

0.693

Tenure in current occupation in years

8.305

7.643

Tenure in current occupation in years squared (/100)

1.199

1.062

Does not speak English at home

0.073

0.061

Health condition, mild or no core activity limitation

0.190

0.171

Health condition, moderate or severe core activity limitation

0.032

0.031

Lives in an inner regional area of Australia

0.190

0.194

Lives in an outer regional area of Australia

0.115

0.112

Bachelor degree or higher

0.253

0.306

Diploma or advanced diploma

0.096

0.116

Certificate III or IV

0.252

0.129

Year 12

0.155

0.168

Year 11 or equivalent (Certificate I or II)

0.069

0.082

Apprentice or trainee

0.021

0.012

Part-time worker

0.108

0.440

Casual worker (without paid leave entitlements)

0.158

0.234

Trade union member

0.286

0.273

Public sector worker

0.228

0.300

Workplace with less than 10 persons

0.216

0.232

Workplace with 10 to 19 persons

0.138

0.140


0.298

0.298

B. Mining

0.029

0.005


0.019

0.005

E.

Construction

0.086

0.015

F.

Wholesale trade

0.061

0.027

G. Retail trade

0.100

0.135


0.037

0.049

I.


0.067

0.025

J.


0.025

0.013

K.


0.033

0.053

L.


0.103

0.108


0.089

0.083

N. Education

0.060

0.151


0.051

0.206

P.

0.022

0.026

Q. Personal and other services

0.039

0.036

Number of observations

5,633

5,730

Qualification dummy variables (base is Year 10 or below)

Job attribute dummy variables

Industry dummy variables (base is manufacturing)


Source: Estimates from ABS Survey of Education and Training (SET) 2005 Expanded CURF. Note: Sample contains employees, other than owner managers of incorporated enterprises, who were aged 21 to 69 years, not employed in Agriculture, forestry or fishing, paid at least $1 per hour, and with valid observations for all variables listed.

page 254


The key control variables include potential labour market experience, measured in years. This represents the so-called Mincer proxy, and is defined as age minus the number of years of education minus 5 (school starting age). In the absence of observed actual labour market experience, comprising the entire labour market history, the Mincer proxy is the best available measure of labour market experience. In general, it is likely that the gender gap in potential labour market experience will under-estimate the true gap in actual experience. This is because the Mincer proxy assumes continuity of labour market attachment from the date of completing education. While this assumption may be valid for many men, given the high levels of male workforce participation we documented in Section 3, it is likely to over-estimate the true levels of participation among women, given the prevailing patterns of fertility and associated labour market discontinuities. It is traditional to include a quadratic in labour market experience within the human capital based wage specification (the raw measure of experience in years, plus the square of this variable), in order to reflect a positive but diminishing impact of each additional year of labour market experience. In addition to potential labour market experience the specification includes variables capturing length of firm and occupational tenure, i.e. number of months spent in the current firm and occupation. Once again quadratics are included for both variables in order to capture nonlinear effects. The other key skill-based variables in the model capture the highest level of educational qualification; i.e. bachelor degree or higher, diploma or advanced diploma, Certificate III or IV, completion of high school Year 12, completion of Year 11 and its equivalent (Certificate I or II) and finally the default (omitted) category of Year 10 or lower. The next set of controls in the wage specification reflect labour market attributes or characteristics of the individual employee. These include whether the employee is an apprentice, whether working in a casual post, whether working part-time, and whether the individual belongs to a union. We expect that both casual status and trade union membership will be associated with positive wage effects, given the prevalence of wage loadings as compensation for the loss of paid leave entitlements for casuals, and the prior econometric evidence of a union wage premium in Australia (see Waddoups, 2005). In contrast, our expectation is of a negative relationship between wages and both parttime and apprenticeship/trainee status. For part-time employees, this effect is due to the accumulation of less general and job-specific experience, relative to full-time workers with otherwise similar characteristics. For apprentices and trainees, the negative effect arises because of the deferral of earnings associated with the training process itself, and because of the existence of specific training wage rates set below the applicable standard adult minima. Finally, our model specification includes three dummy variables capturing workplace size (where the base is workplaces with 100 or more persons); two dummy variables indicating whether the individual lives in an inner or outer regional area of Australia (where the base is residents of major metropolitan areas); two dummy variables that measure the extent (if any) of core activity restriction arising from a long-term health condition or disability (where the base is employees without any such impairment); and one dummy indicating whether the employee normally speaks a language other than English at home. We anticipate that employees without English proficiency and with moderate to severe core activity limitations will, ceteris paribus, have lower hourly wages. For disabled workers, this effect is reinforced by the presence of a supported wage system which enables the payment of wages below the standard adult minima (with the wage being dependent on assessed level of impairment). Looking at the key differences between genders in Table 4-1, the first point to notice is that on average men earn around $3.70 more per hour. Although potential labour market experience is approximately equivalent by gender, women have slightly less tenure both page 255


in terms of the firm and the occupation. The discrepancy in tenure has the potential to explain part of the gender gap in wages. The much higher proportion of women working as casual and/or part-time also has the potential to explain why men earn higher average wages. However, this is an empirical question as casual work in Australia attracts a premium to compensate for the absence of paid leave. In terms of educational qualifications, women on average possess higher education levels. Almost 31 per cent of the female sample have a degree or higher, relative to only 25 per cent of men. Thus, the pattern of gender differences in education cannot directly contribute to the gender wage gap. Finally, Table 4-1 shows significant gender differences in industry affiliation. Females dominate health and community services (over 20 per cent), education (15 per cent) and retail trade (13 per cent). Men appear to be more evenly distributed throughout. 4.2

Econometric results – (1) pooled specification

The econometric analysis begins by estimating a human capital based wage equation, with specification as described above. In the initial analysis the regression pools over the sample of male and female employees and includes a dummy variable for female gender. This specification is limited in that implicitly we are assuming that all slope coefficients are equal across gender – i.e. that the rate of return to another year of experience or a higher educational qualification are the same for each gender. The female gender coefficient measures the return (premium or deficit) to being female, ceteris paribus. Given this definition it is not surprising that the empirical literature interprets this coefficient as being indicative of the extent of gender discrimination. In the basic specification described above, the estimated (and statistically significant) female coefficient is –0.136. Converting this to the wage difference as a percentage of the (geometric) mean male wage implies an estimated level of wage discrimination of 12.7 per cent. This should be interpreted within the context of the overall gender difference in mean log wages of 0.13 (or, as a percentage of the geometric mean male wage, 13.9 per cent). The implication is that after controlling for a variety of human capital characteristics, a significant gender wage differential persists. The size of the gender gap estimated from the basic econometric analysis is larger than that indicated in the earlier analysis of the EEH data. Recall from Table 2-1 that the female disadvantage was 11 per cent in the non-casual market and approximately 7 per cent in the casual market. The strongest candidate explanation for the difference between the two data sources is that the EEH dataset excludes employees whom the ABS defines as having ‘managerial’ responsibilities, while a similar restriction does not apply (and cannot be applied) to the SET data. If men are more strongly represented in managerial positions, this will tend to widen the estimate of the average gender pay gap in our econometric results relative to the estimates obtained from the EEH dataset where these employees are omitted. We can partially control for this effect by the inclusion of the stated human capital variables in our specification, but managerial status is not observable directly in the SET dataset. Two other substantive and related caveats apply to our basic econometric estimation. First, as with all empirical measurement of discrimination, we are implicitly assuming that we have correctly controlled for all gender differences in productivity characteristics, and hence the model is correctly specified. We have already suggested one attribute above (managerial status) which might compromise this assumption. Below we investigate further the sensitivity of our measurements, by extending the specification to include a vector of (1-digit) industry dummies. A second caveat is that the current specification assumes that it is legitimate to constrain the male and female coefficients to equality. This assumption is relaxed in the next section.

page 256


The extended model adds the set of industry dummies to the specification. This step leads to a reduction in the estimated level of discrimination to 10.7 per cent. The conclusion following from this is that after controlling for productivity differences as best we can given data limitations, there remains a significant gender wage gap, most of which is attributed to gender per se. 4.3

Econometric results – (2) separate specifications by gender

Estimating separate male and female models allows for the possibility that the returns to personal and human capital attributes – such as potential labour market experience and formal education – may differ by gender. For instance, men may receive a larger pay increment than women for every additional year of formal education or firm tenure completed. This approach relaxes the assumption underlying a single equation pooled over gender which implicitly assumes that the coefficients on each of the control variables are identical, regardless of sex. An F-test rejects the null hypothesis that it is legitimate to pool across gender. The econometric results in Table 4-2 show the specifications estimated separately by gender. The columns display the male and female results. The number of observations in the male sample is 5,633 and 5,730 for females. The specifications fit the data fairly well with an R-squared of 0.29 and 0.26 for men and women respectively. An F-test of overall fit rejects the null hypothesis that all slope coefficients are jointly equal to zero at the 1 per cent significance level for both male and female specifications.

page 257


Table 4-2: Full results from the basic male and female econometric regressions DV: Natural logarithm of hourly wage Potential experience in years Potential experience squared (/100)

Male

Female

Coeff.

t-stat

Coeff.

t-stat

0.024

12.94

0.018

11.11

–0.041

–10.45

–0.037

–10.49

Tenure in current job in years

0.002

0.51

0.015

5.01

Tenure in current job squared (/100)

0.020

1.27

–0.056

–3.53

Tenure in current occupation in years

0.015

4.51

0.005

1.78

Tenure in current occupation squared (/100)

–0.057

–3.72

–0.005

–0.39

Does not speak English at home

–0.176

–7.79

–0.116

–4.86

Mild or no core activity limitation

–0.057

–3.95

–0.034

–2.76

Moderate or severe core activity limitation

–0.202

–4.29

–0.113

–2.75

Lives in an inner regional area of Australia

–0.057

–4.28

–0.055

–4.69

Lives in an outer regional area of Australia

–0.021

–1.31

–0.071

–5.10

Bachelor degree or higher

0.439

22.32

0.357

21.60

Diploma or advanced diploma

0.290

12.94

0.158

8.44

Certificate III or IV

0.162

10.12

0.049

3.02

Year 12

0.128

6.54

0.096

5.81

Year 11 or equivalent (Certificate I or II)

0.065

2.69

–0.006

–0.32

Apprentice or trainee

–0.233

–7.22

–0.142

–4.75

Part–time worker

–0.051

–1.87

0.006

0.53

Casual (without paid leave entitlements)

–0.056

–2.69

–0.028

–1.88

0.008

0.68

–0.029

–2.75

Trade union member Public sector worker

0.003

0.20

0.086

7.61

Workplace with less than 10 persons

–0.217

–13.56

–0.105

–7.22


–0.184

–11.09

–0.109

–7.25


–0.115

–8.79

–0.079

–6.69

2.726

97.27

2.678

117.38

_Constant R-squared Number of observations

0.29

0.26

5,633

5,730

Source: Estimates from ABS Survey of Education and Training (SET) 2005 Expanded CURF. Note: This version of the model does not include the fifteen industry dummy variables listed in Table 4-1. We report in text the results of re-estimating the model with these variables included.

The results accord with some, but not all, of our a priori expectations. Focussing on the male results, labour market experience has a positive and significant impact on the wage. The negative quadratic coefficient confirms the traditional concave shape, suggesting a diminishing impact of marginal increases. For males, the quadratic pattern suggests that experience raises the log-wage up to 29.6 years of experience and then declines thereafter. Interestingly, firm tenure is not a significant determinant of male wages, whereas tenure within an occupation plays a significant role. In terms of educational qualifications, the results confirm that wages are lowest for employees with the least education. The male coefficients on all the education dummies are positive and statistically significant, and increase in size monotonically from lowest to highest education. This implies that, ceteris paribus, wages rise with academic qualification. The variable capturing apprenticeship is, as expected, negative and highly significant. An apprentice receives significant investment in general skills, transferable across the sector. As Becker points out, workers will pay for the acquisition of general skills and reap the returns post-training. The individual-specific variables capturing the role of language and page 258


disability are both significant and negative determinants of the wage, suggesting that these qualitative traits of employees are important independently of measurable experience and qualifications. Finally, the specification includes a set of characteristics linked to the nature of the job. Interestingly, only the variables related to workplace size and casual status affect wages significantly (at the 5 per cent confidence level) for men. The workplace size dummy variables all carry negative coefficients relative to the default category: establishments with 100 or more persons. This result supports the traditional finding that larger firms pay more. Contrary to expectations, working as a casual has a negative and significant impact on the wage. This result is surprising, as casuals are expected to earn a wage premium in lieu of benefits forgone. We note, however, that casual status is approximated in this study by the absence of paid leave entitlements (the standard proxy) rather than observed directly through the payment of a casual loading. It is plausible that compensatory payments to employees without paid leave vary in a range around the traditionally assumed loading of 20 per cent. Indeed, the ABS has estimated from its 2006 Survey of Working Time Arrangements that 39 per cent of employees without paid leave entitlements were not receiving loadings, and this was more likely to be true of men without paid entitlements (42 per cent) than of women in similar jobs (35 per cent) (ABS, 2007, Table 4, p. 14). If a similar pattern of casual loadings is found in the 2005 SET data, this may help explain why the coefficient on the casual proxy is negative for both sexes in our analysis, and larger in both size and statistical significance for men. Switching to the second column of results in Table 4-2 – those for the female wage specification – in general the estimates are similar to those for men. Qualitatively, the key differences are that unlike the male results, firm tenure rather than occupational tenure plays the significant role. This suggests an important difference between the characteristics of male–female jobs; it suggests that the skills accumulated by males are more easily transferable across firms within a given occupation. Also, the public sector status plays a positive and significant role for females. Finally, and somewhat surprisingly, union status plays a significant and negative role in wage determination for females. This result contrasts with evidence in the literature of a significant wage premium for male union members in Australia in the period 1993 to 2001 (Waddoups, 2005). In this study we fail to detect any direct wage benefit of union membership for men, and instead find a small (3 per cent) but statistically significant wage penalty for women. To examine the sensitivity of results we re-estimate the above specifications including a set of industry dummies. In a qualitative sense the results are largely unaffected, but industry does appear to play a central role in wage determination. The R-squared for males rises from 0.29 to 0.33 and from 0.26 to 0.29 in the female specification. A formal F-test of the null hypothesis that the vector of industry dummy coefficients is jointly equal to zero is resoundingly rejected for both male and female specifications. A puzzle which this analysis highlights, but which we do not directly attempt to resolve in this report, is why industry affiliation should have such a strong impact on the size of the gender pay gap. We have shown (Table 4-1) that women remain substantially more concentrated in key industries than men, and that, on average, women have higher levels of education. We might expect that the higher qualifications of women would counteract or eliminate any wage effect attributable to industry composition, but this does not appear to be the case. In fact, even after controlling for the human capital that men and women bring to the labour market, industry structure remains a significant determinant of wages. The greater problem of why female jobs are clustered in sectors which appear to offer women fewer opportunities to become highly-paid workers (at least by the measured standards of men) must be left to further research. For now, we address the more limited hypothetical question of what female wages would look like if women received the same returns to their (exogenously) given human capital characteristics as men. page 259


4.4

Decomposition analysis

The focus of this section is to use the above regression results in order to carry out a traditional Blinder-Oaxaca decomposition of the gender difference in mean log-wages into a component attributable to differences in productivity-related characteristics and a residual often interpreted as indicative of the extent of discrimination. Appendix A provides technical details of the decomposition, following Blinder (1973) and Oaxaca (1973). As discussed earlier, there are a number of caveats associated with the above interpretation of the residual as a measure of the extent of discrimination. The interpretation relies on the fact that the human capital based wage model is correctly specified. Failure to control for all appropriate measures of labour market skills will tend to lead to an over-estimate of the level of measured discrimination. Similarly, the analysis assumes that the explanatory variables are themselves free of the influence of prior discrimination. For example, the fact that industrial distributions differ by gender is taken as a legitimate potential explanation of why women earn less on average. However, if females face implicit or explicit barriers to entry to certain industries, the gender difference in industrial distributions may itself be a product of discrimination. In order to implement the decomposition, an assumption needs to be made about the hypothetical wage structure prevailing in the absence of gender discrimination. Our analysis assumes that the current male structure would prevail in the absence of discrimination. An argument in favour of using the male structure is that any redress of inequities will almost surely involve raising the salaries of females rather than lowering the salaries of males. The method decomposes the gender wage gap into two components: that which is ‘explained’ by gender differences in productive characteristics, and the ‘unexplained’ residual often interpreted, notwithstanding the aforementioned caveats, as a measure of discrimination. The residual measures the extent to which similar characteristics of females and males are valued differently by their employers. Table 4-3 summarises the decomposition results. The first two rows carry out the decomposition for the two wage specifications described above, i.e. the basic model, and the extended model with industry dummies. The first column displays the overall size of the gender wage gap as represented by the log-wage gap. The following two columns identify the two components within the decomposition, i.e. the explained and the unexplained components. The final column represents the percentage of the overall gap that is explained by gender differences in productivity-related characteristics. This column of results tells us how successfully the characteristics included in our various model specifications account for the gender pay gap, and how this varies across the different levels of the analysis we have undertaken.

page 260


Table 4-3: Summary of results from the decomposition analyses Specification

Size of log-wage gap

Explained component

Unexplained component

Percentage explained

Basic human capital

0.132

0.015

0.117

11

With industry dummies

0.132

0.035

0.097

27

Retail

0.090

0.027

0.063

30

Accommodation

0.051

0.010

0.040

20

Property

0.174

0.110

0.061

63

Health

0.168

0.101

0.067

60

Elementary clerical

0.099

0.020

0.080

20

Intermediate clerical

0.153

0.070

0.080

46

Labourers and related

0.129

0.055

0.074

43

Within industries

Within occupations

The basic human capital specification illustrates that differences in male–female characteristics explain only a small percentage of the overall wage gap (around 11 per cent). This is a relatively small percentage given the parsimonious specification of the basic human capital model. Interestingly, adding a vector of industry dummies increases the proportion explained to 27 per cent – well over double. This is once again suggestive of the important role played by male–female differences in industry of employment. The effect implies that the Commission may play an important role in determining the overall size of the gender wage gap, through its role in setting award wages. To explore this issue further we estimate separate gender-specific regressions for each of the key industry and occupation groups identified earlier in the report as of special interest to the Commission. This enables further examination of the role of gender differences in human capital in contributing to the within-occupation/industry gender wage gap. There is substantial variation in the gender wage gap across industries, ranging from 5 per cent in accommodation to more than 17 per cent in property. Retail and accommodation have smaller gender wage gaps than the average across all industry groups. Property and health on the other hand have higher than average gender wage gaps. In one sense this result is surprising given the substantial reliance on award minimum rates of pay in all four industries. The final column of Table 4-3 also indicates considerable variation in the degree to which human capital characteristics explain the gender wage gap for these four key industries. In property and health, our specifications account for upwards of 60 per cent of the observed gender gap in log-wages. In retail trade we explain only 30 per cent, and in accommodation only 20 per cent. The dominant pattern here is that the industries with smaller overall gender wage gaps (i.e. retail and accommodation) also have the smallest proportion explained by gender-specific differences in human capital. One interpretation of this result is that industries with a strong award structure successfully limit the size of the gender wage gap but also decrease the wage variance and the consequent impact of human capital. In property and health, where the gender wage gap is larger, the human capital characteristics appear to explain a much larger proportion of the overall gap. The explanatory power of our model is also related fundamentally to the distribution of hourly wages within any given workforce, in this case in specific industry sectors. One reason why we are able to explain a generally smaller proportion of the gender differential in hourly wages in retail and accommodation, and comparatively larger proportions in property and health, is that these industries exhibit strikingly different degrees of variation page 261


in the hourly wages employees receive. This fact is illustrated in Table 4-4, which reports the proportions of employees in each industry by their location in the overall distribution of hourly wages in 2005. All employees in our estimating sample were ranked according to their hourly wages, and then divided into ten groups of equal size (deciles). The 10 per cent of employees with the lowest-ranked hourly wages are in Decile 1, the next lowest 10 per cent in Decile 2, and so on. We observe, for instance, that a quarter of employees in the accommodation industry were in the lowest decile of all hourly wage-earners, and 44 per cent were in the first or second lowest decile. The median employee in both the retail and accommodation industries (the employee located at the mid-point of the distribution for that industry) was in the third decile of hourly wages. The retail and accommodation industries thus have very substantial proportions of their workforces situated near the bottom of the overall wage structure. Contrast this with the results for the property and health industries, where employees are much more evenly distributed. Around 54 per cent of employees in the property industry, and 53 per cent in health, earn more than the median hourly wage for the workforce as a whole. This is true of only 19 per cent of retail workers, and only 16 per cent of accommodation workers. These unequal distributions reveal much of why our model is less successful in explaining hourly wage differences in the retail and accommodation industries. In a very real sense, there is less to ‘explain’ – the distribution of pay is so heavily skewed towards low hourly wages that there is little work for our model to do. In property and health, however, the range of wages is much wider, and the human capital characteristics in our model successfully explain more of both the overall structure of wages, and the male–female differential. Table 4-4: (Cumulative) Proportion of employees in selected industries located below each decile cut-off in the distribution of hourly wages for all employees in 2005

Retail trade

Accommodation, cafés and restaurants

1 (Lowest)

19

25

8

10

2

41

44

19

19

3

58

57

27

28

4

72

71

37

38

5

81

84

46

47

6

88

90

56

59

7

93

93

66

70

8

95

97

76

80

Decile

9 10 (Highest)

Property and business

Health and community

97

98

86

92

100

100

100

100

Source: Estimates from ABS Survey of Education and Training (SET) 2005 Expanded CURF. Note: The Table reports cumulative estimates of the proportion of employees, adding from the first to tenth deciles. For instance, 19 per cent of retail trade employees are in the first decile. Another 22 per cent are in the second decile, raising the proportion to the 41 per cent shown in the table. Another 17 per cent are in the third decile, raising the cumulative proportion to 58 per cent, and so on for other cells and other industries.

This conclusion also applies to the results for specific occupations shown in Table 4-3. Elementary clerical, sales and service workers has a lower than average gender wage gap, and our model accounts for only 20 per cent of the gap. In intermediate clerical, sales and services, the gender gap is greater, and we account for a larger share of it with the observable variables available to us in the SET dataset. These differences by industry and occupation highlight an important feature of the low-paid labour market, in that there are generally smaller differences between male and female wages in the sectors where award reliance is high. But the gender differential is only one of several important dimensions of earnings inequality. In the lowest-paid sectors, the problem of inequality manifests less in the specific form of gender disparities, and more in the form of a distribution which is page 262


highly-skewed towards low hourly wages (as shown in Table 4-4). While employees remain within these industries their prospects of attaining better-paying jobs are curtailed by the very small number of such jobs on offer. Male and female wages may be more closely aligned in these sectors, but only because both genders are disadvantaged in these sectors relative to most other Australian employees.

Appendix A: The decomposition method Define the nominal wage gap as:

G=

Wm ‒ Wf . Wf

(1)

Then, taking the natural logarithm gives us

In(G + 1) = In(Wm) ‒ In(Wf ).

(2)

Write the human capital based semi-logarithmic earnings equation regression as:

In(Wi ) = Z'i β + ui .

(3)

The properties of ordinary least squares regressions imply that separate regressions on males and females would give us:

In(Wm ) = Z'm βˆm

(4)

and

In(Wf ) = Z'f βˆf .

(5)

Substituting (4) and (5) into (2) gives us:

In(G + 1) = Z'm βˆm – Z'f βˆf .

(6)

∆ Z' = Z'm – Z'f

(7)

and

∆ βˆ = βˆm – βˆf .

(8)

If we let

If we assume that the current male structure would prevail (there is a wage penalty for being female) in the absence of discrimination, substitutions can be made to arrive at a decomposition equation based on the current male wage structure:

In(G + 1) = ∆ Z'βˆm + ∆ Z'βˆf .

(9)

The first component illustrates that part of the gender wage gap attributable to male– female differences in productivity-related characteristics. The second component reflects the residual difference – often interpreted as indicative of the level of discrimination.

page 263


References ABS (2007). Working Time Arrangements, Australia, November 2006, catalogue no. 6342.0, Australian Bureau of Statistics: Canberra. Blinder, A.S. (1973). ‘Wage discrimination: reduced form and structural estimates’, Journal of Human Resources, 18 (4): 436–55. Oaxaca, R. (1973). ‘Male-female wage differentials in urban labor markets’, International Economic Review, 14 (3): 693–709. Waddoups, C.J. (2005). ‘Trade union decline and union wage effects in Australia’, Industrial Relations, 44 (4): 607–24.

page 264


Forum discussion General discussion following the presentation focussed on: • the finding that female-to-male wage ratios vary significantly by industry and occupational classification and by method of pay setting; • the extent to which women’s own attitudes to work-family balance might contribute to persistent pay gaps, especially at higher skill levels; and • whether gender pay gaps are due to persistent levels of gender discrimination in employment. In recent decades, changes in the structure of industry and occupation have coincided with a significant increase in the female participation rate. In particular, employment has grown very strongly in the (low-paid) service industries, with female employment growth outstripping male employment growth in these industries. By contrast, employment in traditionally maleintensive industries, such as Manufacturing, has been declining. Related to this, there was some discussion of whether a significant slowdown in employment growth would affect gender relativities. It was noted that industries with significant numbers of low-paid workers and low gender wage gaps (such as ‘Retail trade’ and ‘Accommodation, cafés and restaurants’) have a relatively compressed occupational pay structure. This is in contrast to the ‘Property and business services’ and ‘Health and community services’ sectors, which include high-paid professionals, such as realtors and doctors, as well as low-paid workers, such as cleaners and carers. Forum participants expressed interest in the finding that the female to male relativity of average hourly ordinary-time earnings varies by job contract status and method of setting pay. Workers on individual contracts showed a gender wage gap of around 15 per cent, compared with no pay gap for workers on awards and a relatively small gap for those on collective agreements. It was suggested that this may be due in part to differences in wage-setting practices between the private and public sectors. Women’s own attitudes to employment and their preferences to balance work and family commitments (including through substantial periods of part-time work) were also cited as possible contributors to a persistent gender pay gap, particularly among higher-skilled workers. The discussion concluded that, while it is possible that gender discrimination still exists in the workplace, it does not appear to manifest in a systematic fashion, with considerable differences apparent across major industry and occupational classifications. It was suggested that the gender pay gap tends to develop as people age, with little difference between male and female pay for a given job at a younger age.

Forum discussion summary prepared by the Australian Fair Pay Commission Secretariat.

page 265