RDP 9505: Labour-Productivity Growth and Relative Wages: 1978-1994 2. Trends in Labour-productivity Growth

In analysing trends in labour-productivity growth, the traditional measure of labour input is hours worked. While this is the appropriate measure for assessing the average output produced per hour worked, it ignores the fact that unemployed workers are producing no measured output. If these unemployed workers find jobs, this is likely to slow productivity growth as, on average, the new workers will be producing less output than the existing workers. In contrast, output per potential worker may be growing quite strongly. This appears to have been the case in the second half of the 1980s. Figure 2 shows an index of output per potential worker (where the potential workforce is equal to the labour supply). This labour-productivity series is more volatile than the one presented in Figure 1 as it does not adjust for the decline in labour input in recessions. More importantly, this series also shows a much better productivity performance between 1982 and 1989. While declining unemployment contributed to the slowdown in growth in the standard measure of labour productivity, the slowdown does not necessarily imply a decline in the rate at which average living standards were improving.

While swings in the business cycle have a pronounced effect on output per potential worker, they also affect the standard measure of output per hour worked. In recessions, labour productivity tends to decline as firms are reluctant to lose workers who have firm-specific knowledge. In the recovery, this ‘labour hoarding’ means that labour productivity can increase quite quickly. The existence of increasing returns to scale may accentuate this cyclical influence. While taking account of these cyclical influences is important, there is no standard method by which this is done. The approach adopted here is to use the recession troughs as the break dates for splitting the sample into three subperiods. At best, this is only a partial solution as neither the starting nor end points of the entire sample are recession troughs. Even if they were, differences in the nature of recessions may lead to differences in productivity performance over different business cycles.

Table 1 presents labour-productivity growth rates for each industry over the various sub-periods and Figure 3 shows the level of labour productivity in the March quarter of 1978 and the June quarter of 1994. The ASIC industry classifications are used.^[3]

Data on hours worked by industry are obtained from the Labour Force Survey.^[4] This is a survey of individuals and provides industry-level hours worked data from 1978 onwards. Data from a similar survey of firms (the Survey of Employment and Earnings) were first published in September 1983. At least for data on total employment, the Labour Force Survey is preferred by the ABS.^[5] While the general trends in the two surveys are similar, the Survey of Employment and Earnings has shown weaker employment growth over recent years. At the industry level, differences between the two surveys can help explain some of the anomalies in the industry labour-productivity data (see below).

There are clearly large differences across industries in both growth rates and levels of labour productivity. In the communications sector, labour productivity increased at nearly 8 per cent per year between 1978 and 1994, while labour productivity fell at almost 1 per cent per year in the finance, property and business services and the recreation, personal and other services sectors. In terms of levels, in the June quarter of 1994, labour productivity in the mining industry (the sector with the highest level of labour productivity) was almost 5.5 times that in the recreation, personal and other services industry (the sector with the lowest level of labour productivity).

In a number of industries there is an important issue concerning the measurement of output and thus productivity. The most frequently cited measurement problems are in the non-market sectors – public administration and defence, finance, property and business services and community services – where it is difficult to obtain the direct market value of output. In public administration and defence, annual estimates are derived by extrapolating base year output by the sum of deflated estimates of wages, salaries and supplements and constant price estimates of consumption of fixed capital.^[6] With public service average wages rising more quickly than public service pay rates (due to an increase in the average classification level of public servants) measured labour-productivity growth has been positive. To the extent that a higher average classification represents a higher average skill level, it is consistent with rising labour productivity. More problematic is the notion that an increase in the pay rate of a particular classification represents an increase in the ‘price’ of the output, rather than an increase in output produced per hour worked. Similar issues arise in the community services industry.

In the finance, property and business services sector, output in the base year is extrapolated using data on hours worked.^[7] This is based on the assumption that there is zero labour-productivity growth. Despite this, on the measure presented above, labour productivity does change in these industries. While compositional effects play some role in explaining this change, differences in the Labour Force Survey (LFS) and the Survey of Employment and Earnings (SEE) are also important.

Since September 1983, the ABS have used the SEE to extrapolate output in this sector. Figure 4 shows the ratio of finance, property and business services sector employment, as measured by the SEE, to employment as measured by the LFS.^[8] It shows that changes in labour productivity, as measured in this paper, are almost entirely explained by changes in this ratio. Thus if the SEE measure of labour input is used in the productivity calculations, there is no decline in productivity. While in the medium term, the two measures of labour input might be expected to behave similarly, substantial deviations appear to persist for some time. In analysing the differences in the two measures, the ABS states that in the upswing of the employment cycle the (aggregate) SEE series will lag the LFS series due to delays in updating the business register. The increasing use of contract labour in the finance, property and business services sector may also be contributing to the differences.

Other measurement problems, particularly in service industries, arise due to the difficulty in measuring the quality of output. While the statistician makes adjustments to price indices for the improvement in the quality of goods, such adjustments are typically not made for improvements in the quality of services. Whether or not such adjustments should be made is a matter for debate. For example, if shops are open longer hours, providing an increased level of convenience, should some adjustment be made to the price deflator for the retail industry? Should adjustments be made for improvements in quality of service in hotels and restaurants? There are no clear, easily implementable answers to these questions. However, these issues do suggest a deal of caution in concluding that a slowdown in measured labour-productivity growth necessarily implies a slowdown in the march forward of living standards.

Notwithstanding these measurement problems, we now turn to examining the role that particular industries play in driving the aggregate productivity numbers. The aggregate level of labour productivity can be expressed as a weighted average of productivity in each industry, where the weight for a particular industry is that industry's share of total employment. That is:

where Y is aggregate output, L is aggregate labour input and Y_i and L_i are output and labour input in industry i.

Given equation (1), the change in aggregate labour productivity can be expressed as:

The first component measures the contribution to the change in aggregate labour productivity made by productivity improvements within each industry. The second component measures the contribution made by workers shifting between industries. One weakness of this accounting approach is that the two components may not be independent. A sector that has a higher than average level of labour productivity may further increase labour productivity by firing any workers with low marginal product. This would be recorded as an increase in within-sector productivity, as well as a decrease caused by the movement of labour out of an industry with a high average level of productivity. The problem arises because the marginal product of the average and marginal workers may be quite different. This hints at a second and related problem – while average labour-productivity levels differ considerably across sectors, the differences in labour productivity of the marginal worker are likely to be much smaller. Thus, the relatively small compositional effects identified below are likely to be upper estimates of the true size of compositional effects at least at this level of aggregation.

We now turn to the decomposition suggested by equation (2), and examine the contributions to aggregate labour-productivity growth made by various industries in each of the three periods defined in Table 1. For each period, Figure 5 shows the annual change in labour productivity in each sector, multiplied by the sector's share in total employment – the first component of equation (2). Figure 6 shows the second component of equation (2) – the contribution to productivity growth from the movement of labour between sectors. Figure 7 shows the share of total hours worked in each of the industries in March 1978 and June 1994. The biggest changes have been an increase in the employment shares of finance, property and business services and community services and a decline in manufacturing's share.

2.1 The Slowdown in Labour-Productivity Growth From 1983

After growing at an annual rate of 1.34 per cent between March 1978 and March 1983, labour productivity in the non-farm sector grew at just 0.68 per cent per annum between March 1983 and June 1991. Figure 5 and Table 1 show that this slowdown did not occur in all industries, with a number of industries increasing their contribution to aggregate labour-productivity growth. Most notable amongst these was the mining sector, with labour-productivity growth increasing from −3 per cent per annum, to around 6½ per cent per annum; this added 0.16 of a percentage point to annual aggregate productivity growth.

Of the industries contributing to the slowdown, the construction industry made the largest contribution. Had productivity growth in this industry been maintained at the rate that was achieved between March 1978 and March 1983, annual aggregate labour-productivity growth would have been 0.35 of a percentage point faster in the 1983–91 period. The slowdown in labour-productivity growth in manufacturing contributed a further 0.18 of a percentage point to the aggregate slowdown. In total, the slowdown in measured productivity growth in the five service sectors contributed 0.45 of a percentage point.

Within the service sector, wholesale and retail trade made the largest contribution. The slowdown in this sector, from an already slow growth rate, took around one-quarter of a percentage point off aggregate labour-productivity growth, relative to the 1978–83 period After experiencing average labour-productivity growth of around 1 per cent per annum between March 1978 and March 1983, the wholesale and retail trade sector actually recorded negative productivity growth over the following eight years. By March 1991, the level of labour productivity in the sector had fallen nearly 11 per cent from its peak reached in March 1984.

The finance, property and business services sector contributed around 0.13 of a percentage point to the aggregate slowdown. As discussed above, the negative productivity growth in this sector reflects differences in the Labour Force Survey and the Survey of Employment and Earnings. Between 1983 and 1991, the level of labour productivity also fell in the recreation, personal and other services sector. In fact, this sector experienced the largest fall of any industry; the level of labour productivity falling over 14 per cent between March 1983 and June 1991. The deterioration in the productivity performance in this sector contributed almost 0.1 of a percentage point per annum to the overall slowdown.

As noted above, the slowdown in productivity growth in the mid 1980s did not occur in all industries – in addition to the mining sector, the utilities, transport and storage and communications industries all made larger contributions to aggregate labour-productivity growth in the 1983–91 period than they had done in the 1978–83 period. Labour-productivity growth in utilities was 7.7 per cent per year between March 1983 and June 1991 (1.0 per cent in the earlier period), while labour-productivity growth in communications was 7.0 per cent (6.1 per cent), and in transport and storage it was 1.8 per cent (−0.6 per cent).

In part, the faster productivity growth in these industries reflected the microeconomic reforms that were taking place.^[9] To some degree, the low productivity growth in the manufacturing, construction and wholesale and retail trade industries has hidden the macroeconomic benefits of these microeconomic reforms. Not only has the improved performance of sectors that provide important inputs made an indirect impact on the competitiveness of Australian business, but it has also made a significant direct contribution to aggregate labour-productivity growth. Had the productivity growth of these sectors (utilities, communications and transport and storage) been unchanged from the rate that was achieved between March 1978 and March 1983, aggregate labour-productivity growth between 1983 and 1991 would have been at least 0.3 of a percentage point slower than the already low level that was actually experienced.

Finally, an examination of Figure 6 suggests that at the level of aggregation used in this section of the paper, the movement of labour between sectors played only a very small role in explaining the productivity growth slowdown. In the period between March 1978 and March 1983, the movement of labour between sectors added just over 0.1 of a percentage point to annual labour-productivity growth. This compares with a zero contribution over the period from March 1983 to June 1991. In both periods, the growth of employment in the relatively low-productivity community services industry acted to retard aggregate labour-productivity growth. This was offset by the employment growth in the relatively high-productivity finance, property and business service sector. The impact of these effects was, however, quite small. The major difference between the two periods is the decline in the latter period in the employment shares of the mining and utilities industries which have relatively high labour productivity.

2.2 The Pick-up in Labour-Productivity Growth Since 1991

Since mid 1991, the growth rate of labour productivity has increased considerably – averaging 2.5 per cent per annum between June 1991 and June 1994. This improvement has been widespread, with most industries experiencing a pick-up in productivity growth. While in the construction and the finance, property and business services sectors, labour productivity has continued to decline, it has done so at a slower pace. In contrast, the levels of labour productivity in wholesale and retail trade and recreation, personal and other services industries have risen over recent years.

The sector that has made the largest contribution to the turn-around in aggregate productivity performance is the wholesale and retail trade sector – it accounts for around half of a percentage point of the increase in the growth rate. Together, the transport and storage and communications industries have contributed a further 0.3 of a percentage point to the pick-up in aggregate labour-productivity growth. A considerable contribution has also been made by the community services industry (0.45 of a percentage point).

While the movement of labour between sectors has had a larger impact on productivity growth than was previously the case, the role remains relatively small (see Figure 6). The declines in employment shares of the high-productivity utilities and communications industries have contributed negatively to aggregate labour-productivity growth. In total, the effect of the movement of labour between sectors has been to subtract nearly 0.2 of a percentage point per annum from aggregate labour-productivity growth (compared with zero, in the previous period). However, as mentioned earlier, the within-sector productivity contributions may not be independent of the contribution made by movements of labour between sectors. In particular, while in an accounting sense, the fall in relative employment in the high-productivity utilities industry has resulted in a direct decline in overall productivity, the fall in employment was probably important in generating the rapid increases in labour productivity within the sector.

Does the increase in labour-productivity growth represent a fundamental change from the 1980s experience or is it just a one-time adjustment in productivity levels and/or the result of a cyclical upswing?

The top panel of Figure 8 presents indices of labour productivity for the non-farm sector around the time of the recessions of the early 1980s and early 1990s.^[10] The zero point on the time line represents the trough of the recession. The bottom panel shows analogous indices for real non-farm GDP. For the first two years of recovery, labour-productivity growth is similar in the two episodes – increasing at nearly 3 per cent per year. This is despite the fact that in the most recent episode, output growth was considerably slower. In terms of productivity growth, the real difference appears in the third year. In the episode of the early 1980s, labour productivity reached its peak ten quarters after the trough of the recession and then was basically flat for seven years. In the current recovery, labour productivity has continued to increase and, in relative terms, is now considerably above where it was at the same time in the previous cycle. In contrast, non-farm output has not yet increased to the same extent as was the case in the early 1980s.

It is tempting to cite this recent difference in productivity growth as the first sign of a structural change in the underlying rate of productivity growth. However, it is clearly too early to make a definitive judgment on whether such a change has occurred. Explaining, let alone predicting, changes in productivity trends is notoriously difficult. There are, however, a number of factors that give cause for optimism. First, ongoing microeconomic reform and the competitive pressures induced by the increased internationalisation of the economy should help deliver continuing productivity improvements.^[11] The implementation of the Hilmer reforms, which the Industry Commission estimate will add around 5½ per cent per annum to real GDP, is also a positive factor for future productivity growth.

Second, real wages should not need to fall as they did after the wages push of the early 1980s. To some extent, the early-1980s recession was the result of an unsustainably large increase in real wages – this was not the case in the early-1990s recession. An implication of this is that as demand increases, real wages, productivity and employment should all increase. As the cycle progresses, declining unemployment rates may well be associated with declining productivity growth, but with real wages growing, a repeat of the stagnant productivity of the second half of the 1980s experience is unlikely.

Third, as the following discussion argues, the measured performance of the wholesale and retail trade industry should be considerably better than in the 1980s. Most of the adjustment to the deregulation of shopping hours has been completed, so that the adoption of new technologies and more efficient forms of retailing should contribute to aggregate productivity growth.

Fourth, as Baily (1993) argues, the electronics revolution of the past decade has soaked up a lot of resources, as companies have computerised their workplaces and come to grips with the new technology. We may now be on the verge of reaping more fully the gains from this investment. David (1990) argues that the small productivity gains that western countries have so far seen from computerisation are analogous to the small initial gains from the invention of the electric dynamo at the end of the previous century. In that case, large productivity gains were not realised until new industrial facilities, organisational structures and complementary technologies were developed. In the current context, it is difficult to predict exactly when, and to what extent, the benefits of the recent investment in computers and information technology will show up in the productivity numbers. Nevertheless, it is probable that in the next five years substantial benefits will accrue.

While the magnitude and timing of these various factors is uncertain, collectively they provide a basis for believing that labour productivity growth will continue at a reasonable pace for the remainder of the current decade. As growth in GDP slows, some slowing of productivity growth is inevitable, but a return to the stagnant levels of labour-productivity experienced in the second half of the 1980s seems unlikely.

Footnotes

In August 1994, the Australian Bureau of Statistics (ABS) switched from the Australian Standard Industrial Classification (ASIC) to the Australian and New Zealand Standard Industrial Classification (ANZSIC) system. This change has led to some alterations in the definition of industries. Since at the time of writing, historical data under the ANZSIC definitions are only available from December 1984, this paper predominantly uses data based on the ASIC industry definitions. In some cases in the industry studies, ASIC data are used in conjunction with ANZSIC data. Attempts have been made to ensure that the data are as comparable as possible. [3]

Hours-worked data are calculated by multiplying employment by average hours worked. The employment data are seasonally adjusted by the ABS. The hours-worked data are seasonally adjusted using the X11 procedure and then a centred (Henderson) moving average is applied to the seasonally-adjusted data. [4]

See ABS Cat. No. 6248.0. [5]

Quarterly estimates are derived by interpolating the annual estimates with estimates of hours worked (see ABS Cat. No. 5243.0). [6]

This procedure was also used for the 1984/85 base-year estimate for public administration and defence. [7]

The ratio has been rebased to 100 in March 1985. The ABS state that ‘if the incidence of multiple job-holding, labour turnover and part-time working remains fixed over time, the estimates of movement in employment provided by the two series are, in concept, the same.’ (ABS Cat. No. 5211.0, 1989/90, p. 18:33). [8]

For a discussion of the impact of recent microeconomic reforms on productivity see Filmer and Dao (1994). [9]

The GDP(A) series is used here. The data period has been extended to include the December quarter 1994. [10]

See Ergas and Wright (1994) for evidence that internationalisation has affected the performance of firms in the manufacturing industry. [11]