RDP 1999-12: Unemployment and Skills in Australia 3. The Australian Labour Market Disaggregated by Skill

We now summarise the labour market position of skilled and unskilled workers in Australia.[9] Given available data, we generally measure skill either by educational attainment or occupation. Clearly other dimensions of skills are important as well since, as documented in Borland (1998), increases in earnings inequality in Australia can be attributed not at all to increasing returns to education or other observable skill characteristics (such as experience), but to increased dispersion within education and skill groups. This suggests that changes in demand and supply for unobservable skill characteristics have been important, and that correspondingly our aggregate measures of skill provide only a partial picture.

3.1 Employment

Whether measured by educational attainment or occupation skill level, employment of skilled labour has increased substantially over the past two decades (Figure 4). In particular, employment of tertiary educated workers has increased rapidly since 1979, and accounts for virtually all the employment growth since that time.

Figure 4: Employment by Educational Attainment and Occupation Group
Figure 4: Employment by Educational Attainment and Occupation Group

Sources: Transition from Education to Work, ABS Cat. No. 6227.0, various issues. Labour Force Status and Educational Attainment, ABS Cat. No. 6235.0, various issues. Labour Force, Australia, ABS Cat. No. 6203.0, various issues. There is a structural break in the educational attainment data in 1993 due to the change in survey from ABS Cat. No. 6235 to ABS Cat No. 6227.0 and to a change in the classification of some courses by the ABS. There are structural breaks in the occupation data in 1986 and 1996 due to reclassification of occupation groups. See Appendix A for detailed definitions of skill by education and occupation.

Data on employment by occupation are available from 1966 onwards, a much longer time series than employment by educational attainment. However, several problems arise when classifying skill according to occupation. First, workers are classified by occupation based on the last full-time position held in the previous three years. Thus, part-time workers, and unemployed persons who have either never held a job, or have not recently worked full-time, are excluded. Second, occupation groupings were completely reclassified in 1986 and in 1996, making comparisons over time difficult. Third, prior to 1986 a single occupation group covered both tradespersons (generally classified as skilled labour) and labourers and production workers (usually classified as unskilled). As a result, this occupation group (representing a substantial 27 per cent of total employment in 1986) has been excluded from calculations of skilled and unskilled employment – which explains the upward jump in both skilled and unskilled employment after 1986 in the lower panel of Figure 4.

Notwithstanding these problems, Figure 4 illustrates that employment grew faster in high-skilled occupations (currently defined as managers and administrators, professionals, associate professionals, tradespersons and advanced clerical workers).[10] However, the difference between skilled and unskilled employment growth is not nearly as large as when skill is indexed by educational attainment. Comparing growth rates for consecutive years when occupation classifications did not change, growth in skilled employment averaged 2.8 per cent, compared with 1.9 per cent for unskilled employment.

Is this observed increase in skilled employment driven by demand or supply factors? Several authors have tested whether the behaviour of wages, employment and labour supply in Australia is consistent with a stable set of labour demand equations across skill groups. These tests have generally found a consistent increase in the demand for educated labour (Borland and Wilkins 1997) and high-skill occupations (Gregory 1993) since the 1970s. However, the supply of skilled labour also increased substantially over this period, helping to contain wage relativities between skill groups.

3.2 Unemployment

The unemployment rate for individuals with tertiary qualifications is nearly six percentage points below the rate for those with no post-school qualifications. That less-educated individuals have a high unemployment rate is not, however, a recent phenomenon; it has been a feature of the data since the beginning of the sample period. Also, unemployment rates across education groups have fluctuated in a relatively synchronous fashion.

Our education data unfortunately do not cover the mid 1970s, the period generally associated with the increase in the estimated natural rate of unemployment in Australia.[11] However, data by occupation group, which does include this period, present a picture consistent with the education data. Both skilled and unskilled unemployment exhibit a strong upward trend since the 1960s, but in the short run movements in both unemployment rates are dominated by changes in the business cycle. Also, skilled unemployment has been consistently much lower than unskilled unemployment, although the percentage point difference between the two rates has increased over time. Since many of those excluded from the occupation data (part-time workers and individuals with no recent employment history) have a high propensity to be unemployed, both the skilled and unskilled unemployment rates shown in Figure 5 are lower than the actual aggregate unemployment rate.

Figure 5: Unemployment by Occupation and Educational Attainment
Figure 5: Unemployment by Occupation and Educational Attainment

Sources: See Figure 4.

Using the Nickell and Bell (1995) framework outlined in Section 2, we can make a rough estimate of the relative importance of aggregate and relative shifts in labour demand on unemployment. We ask the following question: How much would the unskilled unemployment rate have increased following an aggregate shock big enough to cause the observed increase in the skilled unemployment rate? By an aggregate shock, we mean a proportional upward shift in the wage-setting curve for both skilled and unskilled workers.

From the previous section, we have showed that an aggregate shock of this nature will tend to increase unskilled unemployment more than skilled unemployment. Assuming a CES production function and a competitive product market, Nickell and Bell show that the elasticity of unskilled unemployment with respect to skilled unemployment following an aggregate shock is given by:

where us and uu are skilled and unskilled unemployment rates, η(us) and η(uu) are the unemployment elasticities of wages for skilled and unskilled workers and σ is the elasticity of substitution between skilled and unskilled workers.

We assume there is a common upward shift in the wage-setting curve sufficient to cause the increase in unemployment that was actually observed for degree qualified individuals. Equation 2 allows us to calculate how this increase in unemployment would be expected to affect the other three education groups.

Given that the occupation skill measure excludes a large proportion of the workforce, we focus on education measures for this exercise. We disaggregate the labour force by sex, and into four educational groups: (1) bachelor's degree or higher qualification, (2) non-degree post-secondary qualification, such as a trade certificate or undergraduate diploma, (3) completed high school, but no further qualifications, and (4) did not complete high school. Individuals still at school are excluded. There is a structural break in the education data due to the change in survey from ABS Cat. No. 6235.0 to ABS Cat. No. 6227.0, and thus our data period finishes in 1994.

To apply Equation (2) to the data, we need estimates of η(us), η(uu) and σ. Clearly, estimating these parameters is a substantial task, and we make no independent attempt to do so here. Instead, we use estimates from other studies to calibrate our model. To our knowledge, there are no Australian estimates of the elasticity of substitution between different education groups. Hamermesh (1993) provides a summary of estimates of this elasticity from a range of overseas studies. Nickell and Bell's reading of this evidence leads them to choose an elasticity of 3, although this is substantially higher than some other estimates.[12] We might also expect a greater degree of substitutability between similar labour market groups (e.g. between university-educated and non-degree-tertiary-educated individuals).[13] In the results presented here, we set σ = 3. However, the results barely change when a lower estimate is used, or when different estimates are used for different groups.

Blanchflower and Oswald (1994) estimate the elasticity of earnings with respect to the unemployment rate for Australia. Using data from the 1986 Income Distribution Survey, they find an elasticity of −0.19, based on cross-sectional differences in unemployment rates across Australian states. Kennedy and Borland (1997) re-estimate these results using pooled data from four Income Distribution Surveys and including dummy variables for state of residence. They find an elasticity of −0.073, somewhat smaller than Blanchflower and Oswald. We use this latter estimate as the wage elasticity for each of the education groups.

Based on these estimates, we are now in a position to calculate the predicted effect of an aggregate wage-setting shock on unemployment for the other three education groups. We compare our predictions to actual experience. Results are presented in Figure 6.

Figure 6: Unemployment and Aggregate Wage Pressure
University degree
Figure 6: Unemployment and Aggregate Wage Pressure

Notes: Black line is actual unemployment; grey line is predicted unemployment

Over this period, the evolution of unemployment across different education groups can be explained almost entirely by changes in the aggregate unemployment rate. The actual and predicted unemployment rates track each other quite closely for every education group. This is consistent with a series of aggregate labour market shifts that increased the overall unemployment rate, but left the structure of unemployment rates across education groups basically intact.

As mentioned above, our findings are not sensitive to changing the elasticity of substitution between skill groups. However, the results are quite sensitive to our estimate of the wage elasticity (η(u)). Borland and Kennedy's preferred estimate of −0.073 is estimated with a considerable degree of uncertainty (the reported standard error on this estimate is 0.04). We tested the robustness of our results to changes in this estimate. We also experimented with different elasticities for different skill groups. Increasing the wage elasticity to a large number (e.g. −0.2 or −0.3) made some differences to our results, but did not alter our substantive conclusions. But lowering the elasticity to a very small value does make a substantial difference; the predicted rise in less-educated unemployment is much smaller. For example, if we set the elasticity at −0.01, the predicted rate for males who did not complete high school is only 12.2 per cent at the end of the sample period, compared with an actual rate of 16.3 per cent.

One further limitation of this exercise is that our data only begin in 1979. Most studies suggest that the increase in the natural rate of unemployment in Australia occurred prior to this, in the early to mid 1970s (Debelle and Vickery 1998). However, although we have no education data for the 1970s, a visual inspection of the data on unemployment by occupation in Figure 5 does not suggest that low-skilled unemployment increased disproportionately over this period.

How can these findings about the stability of relative unemployment rates be reconciled with the large increase in the demand for skilled labour that occurred over this period? The answer is that during this period, the supply of educated labour increased rapidly, broadly keeping pace with the increase in demand. Consistent with this explanation, wage relativities across skill groups remained fairly constant between the late 1970s and early 1990s (as shown below).

Our results are consistent with Nickell and Bell (1995) and Jackman et al (1997), who find that the increase in unskilled unemployment in a range of OECD countries can be explained mainly by aggregate factors which also caused an increase in skilled unemployment. Murphy (1995) and Katz (1998) have offered a possible explanation for the substantial rise in measured skilled unemployment in these countries, even in the presence of strong SBTC. This explanation relies on the difficulty of measuring skill, and the likelihood that education provides only a noisy indicator of it. Then, high unemployment among a small group of highly educated but low skilled workers could explain why the unemployment rate of highly educated workers has increased. Furthermore, labour market policies which artificially maintain wage relativities within education and occupation groups could lead to an increase in unemployment in all groups, especially if there are generous unemployment benefits which discourage the unemployed skilled workers from competing for the less skilled jobs.

3.3 Wages

Borland (1998) presents evidence on pre-tax earnings according to educational attainment using data on individuals from the Income Distribution Survey. These data suggest that the earnings premium attached to a university degree actually declined substantially during the 1970s, but has remained fairly constant since that time. In 1968/69, average earnings for a university-educated male worker were 2.4 times those of a worker who had not completed secondary school. By 1978/79 this ratio had dropped to 1.9, and by 1989/90 had stabilised at around 1.8 (based on annual earnings). A similar reduction in educational premium can be observed for women (a summary of Borland's data is presented in Table 1).

Table 1: Average Relative Earnings by Level of Educational Attainment
Full-time workers: 1968/69 to 1989/90, relative to those who did not complete high school
  University Degree Trade qualification/
Completed high
Not completed
high school
1968/69 235.2 131.2 113.9 100.0
1973/74 207.8 124.9 111.9 100.0
1978/79 187.1 121.1 108.4 100.0
1981/82 178.9 117.1 99.1 100.0
1985/86 171.2 122.1 105.2 100.0
1989/90 180.4 120.4 107.4 100.0
1973/74 208.1 135.8 109.7 100.0
1978/79 169.8 124.3 109.2 100.0
1981/82 174.3 121.6 109.5 100.0
1985/86 167.9 124.8 109.0 100.0
1989/90 170.4 125.2 105.4 100.0

Notes: Reproduced from data in Borland (1998), Table 8.

Borland also decomposes changes in inequality of weekly earnings between 1982 and 1994–95 into changes in observable factors (education and years of experience) and unobservable factors (the remainder). He finds that the returns to education and experience fell by a small amount over the period. Earnings inequality increased somewhat for both males and females; however, this was due entirely to changes in unobservable factors.

Borland does not control for occupational skills as part of his set of observable characteristics. Table 2 presents data on the evolution of relative wages across occupational groups since 1975.

Table 2: Wages for Full-time Workers by Occupation
Relative to the wage of professionals
  Males   Females
1975 1980 1985 1975 1980 1985
Professional and technical 100 100 100   100 100 100
Administrative, executive, managerial 104 102 99   97 96 92
Clerical 74 75 76   75 74 74
Sales 73 72 72   66 62 62
Transport and communications 73 78 81   75 79 77
Tradesmen, production workers 68 70 72   66 65 64
Service, sport, recreation 69 76 75   68 68 66
Farmers, fishermen, timbergetters 57 58 56   59 51 54
  Males   Females
1986 1990 1995 1986 1990 1995
Professionals 100 100 100   100 100 100
Managers and administrators 100 98 99   91 98 102
Para-professionals 86 82 86   80 87 88
Clerks 71 72 72   68 72 72
Salespersons and personal service workers 69 72 72   58 63 66
Plant and machine operators 72 73 71   58 61 57
Tradespersons 68 67 64   57 60 58
Labourers and related workers 61 62 58   59 58 56

Notes: Wages for full-time workers in main job. Earnings between 1975 and 1985 are indexed to the ‘Professional and technical’ group. The occupational classifications were re-defined in 1986, after which wages are indexed to the ‘Professionals’ occupational group. Occupational groups are not directly comparable between these two periods. Occupational groupings were re-classified again in 1996. The new groupings are excluded from the above table, since only one set of results using the new classifications is publicly available.

Source: Weekly Earnings of Employees (Distribution), Australia. ABS Cat. No. 6310.0

More-skilled occupational groups (managers and administrators, professionals and para-professionals) clearly enjoy a substantial wage premium over less-skilled workers. However, there is little evidence that this premium has increased over time. Our analysis is made more difficult by the change in classification of occupations in 1986. From 1975 to 1985 the earnings of several low-paid male occupations (service, sport and recreation, and tradepersons and production workers) actually improved relative to professional and managerial salaries. Between 1986 and 1995 there was no clear trend; wages for some low-paid occupations (such as salespersons) improved relative to professionals, while others declined. For women, there was some increase in the wage premium for managers/administrators and para-professionals compared with other groups. However, once again, there is no clear pattern suggesting an increasing premium for skill.

Thus, data on wages by educational attainment, experience and occupation suggest little evidence of a widening in earnings across skill groups. The increase in income inequality that did occur over the last two decades, is attributable to a widening distribution of incomes within skill groups due to unobservable factors.

3.4 Summary

Our reading of the evidence suggests the following:

  1. Labour demand for educated labour in Australia has increased rapidly since the 1970s, consistent with trends in other countries. Employment has also increased in skilled occupation groups, although this increase has not been as pronounced as the increase in demand for the highly educated. Although overseas evidence suggests the increased demand for skill has been driven largely by skill-biased technological progress, not enough evidence exists for Australia to make definitive statements about the causes of the demand shift.
  2. Despite the shift in demand towards skilled labour, the structure of unemployment rates across skill groups has remained substantially unchanged. Thus, changes in labour supply have basically kept pace with demand shifts. The increase in unemployment rates across all education and occupation groups is consistent with an aggregate shift in wage-setting. This is consistent with the experience of a number of other countries (for example, Canada and Britain), although the United States did experience a disproportionate rise in unskilled unemployment over the past two decades.
  3. Wage relativities between education, experience and occupation groups became more compressed over the 1970s, but remained fairly constant over the 1980s and early 1990s. Once again, this is consistent with our hypothesis that movements in unemployment rates reflect aggregate, rather than group-specific factors. Earnings inequality did increase over this period; however, this was not due to an increase in returns to observable skill factors.


Our discussion is drawn in part from the evidence presented in Debelle and Swann (1998). [9]

The Australian Standard Classification of Occupations (ASCO) ranks occupation grouping from 1 to 5 according to skill level (where 1 is the most skilled). We deem occupation groups with a ranking of 1, 2 or 3 to be skilled. [10]

See Debelle and Vickery (1998) for a summary of natural rate estimates in Australia. [11]

By way of contrast, Jackman et al (1997) find the elasticity of substitution between educated and uneducated labour to be consistent with a Cobb-Douglas production function, implying an elasticity of unity. So clearly there is a large degree of uncertainty regarding the appropriate elasticity – as illustrated by the wide range of estimates provided in Hamermesh's Table 3.8. [12]

Alternatively, there may be asymmetry in substitution between different groups – an educated individual may be a good substitute (in most cases) for someone who has not completed high school, although the converse would not generally be true. [13]