Interpreting ABS productivity statistics

Key assumptions and limitations in interpreting productivity statistics.

Released
13/12/2023

Introduction

Productivity measures, broadly defined as the ratio of the volume of outputs to inputs, are key indicators of the effective use of economic resources. This is especially so in the long run, since investments in human and physical capital, technology and innovation take time to materialise into improved economic efficiencies and into output. In the short run, productivity measures may be subject to a range of distortions, including modelling assumptions, measurement challenges, and underlying data quality.

To illustrate, market sector labour productivity fell 2.9 per cent in 2022-23, the largest fall recorded since the time series commenced in 1994-95. However, caution should be exercised trying to infer the degree of productive efficiency or technology change from this result, as the economy has rebounded unevenly from recent natural disasters and the pandemic response. For example, customer facing industries saw a significant surge in employment and hours worked in 2022-23. The latest productivity estimates are also subject to revisions as firmer estimates become available.

This article outlines why it is important to measure productivity, how the Australian Bureau of Statistics (ABS) measures it, and some of the key assumptions and limitations when interpreting productivity statistics. The main drivers for the unprecedented fall in productivity in 2022-23 are examined, as well as some caveats interpreting this result in the context of recent macroeconomic developments.

What is productivity

Productivity is broadly defined as the ratio of a volume measure of output to a volume measure of input; that is, output per unit of input. Growth in productivity can occur from an increase in output, a decrease in inputs or a combination of both. Productivity growth is the part of output growth that is not accounted for by growth in inputs.

Labour productivity is defined as a ratio of output to labour input, that is, the amount of output produced for an hour of work. Changes in this ratio can also reflect changes in other inputs (such as capital). An increase in labour productivity means that more output is being produced per hour of work.

Capital productivity is defined as a ratio of output to capital input; that is, output per unit of capital. Changes in this ratio can also reflect technological changes, and changes in other inputs (such as labour).

Capital deepening refers to an increase over time in the capital to labour ratio. Increased capital deepening is generally beneficial to economic growth since, on average, each unit of labour has more capital to work with to produce output, and therefore is an indicator of an economy’s ability to reduce labour-intensive production practices. Labour saving practices such as automation of production will result in increased capital deepening.¹ The opposite phenomenon, capital shallowing, occurs when the labour inputs to production grow faster than capital inputs.

Multifactor productivity (MFP) is defined as a ratio of output to combined labour and capital inputs. It is often expressed in terms of a growth rate, that is, the growth rate of output minus the growth rate of inputs. At the aggregate and industry level, MFP is defined as the ratio of gross value added (GVA) to the combined inputs of capital and labour. It reflects the growth in GVA which is not explained by the combination of capital and labour inputs and is a measure of technological progress and enhanced efficiency.

Assumptions related to productivity measurement and analysis

GVA estimates outside the supply-use timeframe

Key assumption: For GVA estimates in time periods after the last supply-use benchmark year, gross output and intermediate consumption grow or decline at the same rate.  

Implication: If intermediate inputs are growing more quickly than gross output, productivity growth will be overstated.

Estimates of GVA for each industry are compiled through the ABS National Accounts supply-use framework.  The supply-use time series currently extends to 2021-22.² GVA estimates for 2022-23 are derived by extrapolating the annual benchmarks using the best available indicators at the time of compilation. A key assumption for estimates compiled for years after the supply-use time series is that gross output, intermediate consumption, and GVA grow or decline at the same rate. While indicator-based methods were considered relatively stable in the pre-COVID-19 period,³ the growth rates for some parts of the economy have moved more abruptly during and after the pandemic, resulting in some larger revisions, especially in customer-facing industries. For example, Accommodation and food services recorded a 2.4% revision to GVA in 2021-22, indicating the extrapolation was lower than the benchmark estimate. Revisions to indicator-based methods are expected which means productivity estimates will also be revised, resulting in an increased likelihood of a higher short-run distortion to productivity measures. These distortions are expected to recede as firmer source data become available through the supply-use benchmarks.

Capacity utilisation

Key assumption: Capital utilisation is constant over an assumed capital asset  life.

Implication: If capital utilisation is overestimated, productivity growth will be understated.

Businesses use labour and capital inputs to varying degrees of intensity depending on prevailing economic conditions. The intensity of use of inputs to production, measured against a possible maximum capacity, is defined as utilisation.  

Changes in the utilisation of capital inputs are typically unobservable. For example, a farm harvester may be used to full capacity in one year of operation due to a bumper crop, and then at half that intensity in another year, due to drought conditions. Data on the variation in the utilisation of the harvester is not available from the farmer. Instead, productive capital stock estimates are used.  These are derived using the perpetual inventory method (PIM).⁴ The PIM assumes that the utilisation of capital is constant over an assumed capital asset life.⁵ This approach is applied to all types of capital, regardless of industry.

In the short run, utilisation rates of labour and capital can vary due to a range of factors, including changes in the economic cycle or economic shocks. For example, in cyclical downturns, businesses may retain partially-underutilised skilled labour even though there is reduced economic activity. This helps businesses to reduce future costs for recruitment and training when economic activity recovers. Capital inputs are often fixed in the short term, and reducing utilisation may be the only option for many businesses in response to temporary economic shocks. 

Hours worked, on the other hand, is expected to reflect utilisation more accurately than capital because unutilised/underutilised labour is not counted. Further, hours worked can be adjusted for attributes of quality.

To the extent that capital utilisation is not constant, there will be an impact on capital deepening estimates. For example, in Accommodation and food services, modelled capital services⁶ growth was likely overstated (reflecting unused or underutilised capital) during the COVID-19 restrictions, whereas workers on Jobkeeper would have reported working less hours. Consequently, modelled capital services were likely to be understated in the recovery period (when the restrictions were lifted).⁷

In 2020, the ABS reviewed the potential impact of the capital utilisation assumption.⁸ Various methods were tested, but in general, they related to using observed estimates for labour underutilisation as proxies for the unobservable underutilisation of capital. On balance, the modelled utilisation changes were more influential on MFP in the short run. In the long run, the average growth of utilisation-adjusted capital converges on the unadjusted official index, so that long run MFP distortion is minimised. 

Extending the modelled utilisation-adjusted indicators to 2022-23 indicated that implied growth in utilisation adjusted capital services is sensitive to the choice of method. Two of the methods (actual/potential output and economy wide unemployment) resulted in lower capital services growth since 2019-20, which would result in higher MFP growth (Figure 1). The highest MFP growth was implied by the actual/potential output method. However, the two under employment methods showed a stronger dip in 2019-20 followed by a stronger acceleration in capital services, exceeding the official index in 2022-23. These would result in lower MFP growth. Moreover, capital shallowing in the latest year would not be as pronounced as the unemployment based method. This is because the aggregate under employment rates have been falling faster than the unemployment rate over the last two years.

Valuation of land

Key Assumption: The volume of agricultural land is held constant.

Implication: If the volume of agricultural land is decreasing, Agriculture productivity growth will be understated.

Capital services from land are difficult to estimate because, in practice, most real estate transactions contain land combined with other types of assets and are difficult to separate.

For example, a volume estimate for agricultural land is derived from the National Balance Sheet in the Australian System of National Accounts.⁹ Because there is no suitable price index for agricultural land, its volume is assumed to be constant over time.¹⁰ The current price value in the reference year (that is, T-1) is used for the volume measure. Variations over time in the number of hectares of agricultural land, for example, a reduction in hectares due to rezoning to residential land, are therefore not captured.¹¹ Further, the reference year value may change with each annual update due to revaluations over time. This may influence the share of the remainder of capital services in Agriculture for the entire time span, resulting in revisions to the capital services index.

In addition, capital services from land are not adjusted for changes in utilisation or changes in quality such as degradation or erosion, which can make them a source of bias that is difficult to quantify.

Capital rental prices

Key assumption: The internal rate of return is CPI + 4 per cent, or the rate that exhausts capital income.

Implication: Any missing capital are assumed to have the same growth properties as the capital that was included.

Key assumption: The internal rate of return is the same of all assets within an industry.

Implication: If the true (unobserved) compensation for each type of asset within an industry varies, industry measured productivity growth will be distorted.

The capital rental price, which can be thought of as the ‘wage’ of capital, requires an estimated rate of return to capital.

The rate of return¹² to capital can be estimated by either the endogenous or exogenous approach. The ABS uses both. First, the total value of capital services in each industry is assumed to be equal to the compensation for all assets in that industry.¹³ The resulting internal rate of return exhausts capital income and is consistent with perfect competition and constant returns to scale. The rate of return is the same for all assets in an industry but may vary across industries.

However, this endogenous approach does not guarantee that rental prices are positive, which is a requirement of productivity estimation. To prevent negative rental prices, the ABS imposes an exogenous floor limit on the internal rate of return equal to CPI plus 4 per cent.¹⁴ This method is less strict on efficient markets since it allows the value of capital income to deviate from property compensation, assuming imperfect competition and non-constant returns to scale.¹⁵

Measuring non-market productivity

Key assumption: A significant portion of non-market output uses cost of service to estimate output.

Implication: Where the non-market sector has incorporated technological innovation, productivity growth will be understated.

The ABS does not currently publish MFP estimates for three industries considered to be predominantly non-market in nature. These industries are Public administration and safety, Education and training, and Health care and social assistance.  

These industries are considered to be predominantly non-market because a significant proportion of their output is provided free of charge or at prices which are not economically significant (in that there is only a weak relationship between the price and the cost of providing that service). Examples of non-market services include public schools and public hospitals. Assumptions of perfect competition and constant returns to scale that underpin the calculation of capital rental prices don’t hold for non-market activity. Therefore, while labour productivity can be estimated for non-market activity, it is difficult to measure capital productivity or MFP.

The lack of meaningful prices for the non-market industries means it is impossible to calculate volume measures of output via price deflation. Therefore, the approach to estimating non-market output for individually consumed non-market services is cost-weighted directly observed output (such as number of full-time equivalent student enrolments and number of hospital separations). While directly observed output measures provide superior output indicators to the cost-of-service delivery approach, the latter approach is still used extensively for collectively consumed services such as public administration and safety. Therefore, to the extent that there has been technological innovation in the non-market sector, productivity growth measures would be expected to be understated relative to corresponding market sector growth measures.¹⁶

The ABS has produced a set of experimental MFP estimates for the following components of the Education and Training and Health Care and Social Assistance divisions using output measures that are independent of the cost-of-service delivery:

Experimental capital services indexes for the health and education industries were calculated using the exogenous method described above (i.e., a floor limit on the internal rate of return equal to CPI plus 4 per cent).¹⁷

Further detail on the conceptual and practical difficulties in measuring output and productivity growth for the non-market sector can be found in this paper: Non-market output measures in the Australian National Accounts.

Quantity and quality of labour

Key assumption: The growth in labour quality is held constant after the latest Census benchmark.

Implication: If labour quality growth falls in the short term, productivity growth will be understated.

An important component in measuring productivity is labour input, typically measured by hours worked. Adding up the total number of hours worked is a simplified measure of the services provided by labour. It does not recognise variation in labour’s contribution to output from a variety of factors. For example, there is no adjustment for the fall in the marginal utility of each additional hour worked that may occur due to working longer hours or working in multiple jobs¹⁸. Each hour worked by each worker is assumed to be of equal quality. For example, the marginal utility of each hour worked in multiple jobs is assumed to be equal to the marginal utility of each hour worked in their main job.

Some quality metrics are separately captured via labour composition. This measure of labour input captures changes in quality of the workforce. That is, it quantifies the evolving abilities of workers from varying educational achievements and experience as contributing factors to labour input. Worker’s characteristics are determined by cross classifying the workforce by industry of employment, sex, educational attainment, and age group. These variables are sourced from the Census of Population and Housing. Adjusting hours worked for changes in labour composition results in a Quality adjusted hours worked measure (or Quality adjusted labour input (QALI)). 

Labour composition changes for the years falling between the census years are interpolated while changes for years past the most recent 2021 census are extrapolated. The extrapolation of the labour composition component beyond the 2021 census benchmark has the effect of holding the growth in labour quality constant.¹⁹

Since labour composition captures improvements in labour quality of the workforce, the rate of growth of labour input measured on a Quality adjusted hours worked basis is generally higher than the growth measured on an hours worked basis. Consequently, MFP will grow at a slower rate when measured on a QALI basis compared to an hour worked basis.

More recently, historically low unemployment rates indicate excess labour demand. This is supported in 2022-23 when workers worked longer than usual hours and more workers worked in multiple jobs.²⁰ In these circumstances, firms may be incentivised to retain more labour than required in the short-run. While this helps firms to retain essential skills, reduce training costs, and meet increased future demand for goods and services in the long-run, it can act as a drag on labour productivity growth in the short-run.

Factors outside the production boundary and the timing of recording on productivity

Key assumption: National accounts inputs do not include the weather.

Implication: To the extent that the weather impacts output growth, it is also captured in MFP growth.

Key assumption: Progressive building of large infrastructure projects result in capital services being recorded before the assets are productive.

Implication: If capital services are overstated, productivity growth will be understated.

Inputs do not necessarily line up with outputs, which can create distortions in productivity statistics. Some noteworthy examples include:

  • A key input to the agriculture industry is the weather. Rainfall and sunshine are not recorded as inputs in the production function.²¹ To the extent that the weather influences agricultural outputs, but not measured inputs, this influence spills into the MFP estimates. For example, Figure 2 shows that MFP growth for Agriculture, forestry and fishing is broadly associated with output growth. This is due to favourable weather conditions, which are associated with large harvests. Conversely, declines in MFP growth were associated with the 2002-03 and 2017-19 drought.
  • In the mining industry, large infrastructure projects are needed to facilitate extraction and shipment of mineral and energy resources. Capital flows (gross fixed capital formation) are recorded progressively, which means unproductive componentry of projects can enter the capital stock, and are therefore assumed to be providing capital services, well before the project or facility is delivered and becomes operational. Figure 3 shows an MFP slump from 2001-02 to 2012-13. This was associated with strong combined inputs growth, reflecting capital services from some large, but incomplete, infrastructure projects. From 2013-14, MFP growth turned positive, reflecting output growth associated with newly operationalised projects exceeding combined inputs.²²

Household assets in production

Key assumption: National accounts exclude household assets from the productive capital stock.

Implication: To the extent that household assets are used in production, productivity growth will be overstated.

Household assets such as cars, laptops and appliances owned by households are classified as final consumption by households. These items are not included in the productive capital stock, and therefore do not generate capital services. 

When a household uses their car, their bicycle, their phone or other belongings to produce services such as ride sharing or food delivery, the capital used in this productive activity is excluded from productivity statistics. This has the effect of increasing productivity growth.

Impact of COVID-19 on labour productivity

The pandemic meant that many businesses operated under stringent economic conditions or were forced to close, which meant capital was underutilised or left completely idle. Some workforce participants were paid under Jobkeeper to retain their connection to their employer. Hours worked data showed that by June quarter 2020, 8.8 per cent less hours were worked than the same quarter of the year before. This translated into a positive temporary shock to labour productivity, implying that even though the economy had slowed, those who were still able to work were more productive. 

Once workers returned to work, the shock to labour productivity flipped sign, as can be seen in Figure 4. Current trends suggest that the negative shock to labour productivity is unwinding as the temporary surge in hours worked following the easing of restrictions is beginning to taper off.

Labour productivity and real unit labour costs

Key assumption: In measuring growth in real unit labour costs, average labour compensation is deflated by the GDP deflator.

Implication: There is no direct link between labour productivity and real unit labour costs.

Unit labour costs (ULC) are an indicator of the cost of labour per unit of output produced in the economy, providing a link between average labour costs and productivity. ULCs are calculated as average unit labour costs, divided by average labour productivity. For example, there will be no change in the ULC if an increase in labour productivity exactly offsets an increase in average labour costs.

Labour productivity growth reflects growth in two areas. The first is from an increasing capital-labour ratio (capital deepening), indicating more capital per unit of labour input. The second is from increasing MFP, which can occur through the introduction of new technologies, organisational improvements, economies of scale, or the implementation of research and development. 

Real unit labour costs (RULC) are calculated by deflating the ULC’s numerator with the GDP deflator. This methodology is widely used internationally, but it presents limitations for small open economies such as Australia that are reliant on commodity exports. Swings in the terms of trade impact the GDP deflator which in turn flow through to RULC. This means the link between labour productivity and the cost of labour in real terms is no longer direct. In general, assuming no change in underlying labour productivity, a fall in the terms of trade will drag the GDP deflator down, which will push RULC up.

Recent quarterly data shows a strong decline in the terms of trade, which contributes to a flattening of the GDP deflator. In the September quarter 2023, the GDP deflator had increased 2.4 per cent on a through the year basis, well below CPI. This is largely driven by a fall in the terms of trade, which had fallen 9.0 per cent through the year. The combination of these influences caused RULC to increase strongly, to 3.9 per cent through-the-year. 

As the terms of trade component of the GDP deflator is a volatile influence on the RULC, caution needs to be exercised when attributing the RULC changes directly to domestic influences such as labour productivity.

Figure 5 shows that labour productivity rose by 6.5 percentage points (ppts) from 2012-13 to 2022-23. ULC rose by 23.2 percentage points over the same period. However, the GDP deflator (GDP IPD), over the same period, grew more than 31.0 percentage points. As the figure shows, the influence of the terms of trade on the GDP deflator was significant, especially since 2016-17. The rise in the terms of trade was a factor in pushing RULC about 6.0 percentage points lower since 2012-13.

Recent productivity estimates

Market sector labour productivity fell 2.9 per cent in 2022-23, the largest fall recorded since the time series commenced in 1994-95. The strong fall was largely due to the strength in hours worked, which grew 6.9 per cent, the largest market sector growth recorded. The strong growth in hours worked outpaced GVA growth of 3.8 per cent. Caution should be exercised to infer the degree of productive efficiency or technology change from the year-to-year productivity growth results.  This is especially so in 2022-23, as the economy has rebounded unevenly from recent natural disasters and the pandemic response. 

During the COVID-19 period, productivity indicators, particularly at industry level, saw exacerbated swings, data revisions, and downward shocks followed by significant rebound effects in subsequent periods. The experimental labour productivity industry indicators showed significant reallocations of labour between low productivity and high productivity industries and some distortions in labour allocation due to the Jobkeeper programme.

Interpreting the 2022-23 results against the last five years provides some useful context (Table 1).

Table 1: Long-term and short-term MFP estimates comparisons, percentage change
 2022-232018-19 to 2022-23**1994-95 to 2022-23
GVA3.82.03.2
Hours worked6.91.41.4
Capital services1.61.53.8
Labour productivity-2.90.61.8
MFP-0.50.60.8

**2018-19 to 2022-23 is meant to consider COVID-19 as a time frame. However, it is one year longer than the latest growth cycle. This is so that the latest year can be compared both in isolation as well as in the context of the post COVID-19 period. 

There are likely a multitude of reasons why the strong hours worked growth of 6.9 per cent has not translated into stronger GVA growth in 2022-23. One explanation may be that there was less capital available for each hour worked (that is, production was more labour intensive) in 2022-23. Averaging 3.8 per cent annual growth per year from 1994-95 to 2022-23, capital services has been growing at less than half the average rate (1.5 per cent) over the last five years, contributing less to GVA growth than labour over that period.

On balance, MFP averaging 0.6 per cent per year from 2018-19 to 2022-23 was just below the long run average per year of 0.8 per cent. That is, the large difference between the five-year average and long run labour productivity did not translate into a large difference between five-year average and long run MFP. Moreover, the gap between the five-year average and long run labour productivity and GVA can be explained mainly by the gap in capital services growth, since there was no hours worked gap. Lower GVA growth has translated into lower labour productivity growth.

Key drivers of labour productivity decline in 2022-23

To shed light on the recent significant reversal in labour productivity growth estimated for 2022-23, labour productivity for the market sector can be decomposed as in Figure 6. Figure 6 shows that total market sector labour productivity can be decomposed into a direct effect, also known as the within-industry effect, and reallocation effects between industries.²³ The latter effect occurs when workers move to different industries. Over the last two years, market sector labour productivity growth has been mainly due to the direct effect. That is, the reallocation effect has been less important in driving the market sector labour productivity change since 2021-22 than in previous years.

Turning to the direct effect, Figure 7 shows each of the direct effect contributions. For 2021-22, the strength in labour productivity was mainly due to MFP that recorded a 2.2 per cent growth. In 2022-23, the main contributor to the fall in labour productivity was non-IT capital deepening, which recorded a 2.1 per cent fall. Noting that since these estimates are after the supply-use benchmarks, they need to be treated with caution.

The main industry contributors to the fall in non-IT capital deepening were Mining, Financial and insurance services, Transport, postal and warehousing, Accommodation and food services, and Construction (Figure 8). These industries accounted for more than 80 per cent of the fall in market sector capital deepening. A fall in capital deepening is also known as capital shallowing, since there is less measured capital services available for each hour worked. 

Capital deepening is based on the growth in capital services less the growth in hours worked. Figure 9 shows the unweighted percentage changes for the main industry contributors. Note that positive increases in hours worked are plotted as a negative because they subtract from capital deepening. The growth in hours worked outpaced the growth in capital services, resulting in capital shallowing. Growth in hours worked in 2022-23 exceeded 10 per cent in Transport, postal and warehousing, Wholesale trade and exceeded 20 per cent in Accommodation and food services. 

** A positive growth in hours worked is recorded as a deduction when estimating capital deepening.

Productivity growth cycles

While averaging over the last five years helps to abstract from a variety of temporary distortions, MFP estimates are perhaps most useful when viewed as average growth rates between growth-cycle peaks. Growth cycles are chosen with reference to peak deviations which are determined by comparing MFP estimates with their corresponding long-term trend estimates. The peak deviation between these two series is the primary indicator of a growth-cycle peak. General economic conditions at the time are also considered. In this way, most of the effects of variations in capacity utilisation and much of the random variation is removed. However, average growth rates may still reflect any systematic bias resulting from the methodology and data used.

Currently, the ABS publishes market sector MFP growth cycles, both in the Australian System of National Accounts and in the Estimates of Industry Multifactor Productivity. These releases identify a new productivity growth cycle, from 2017-18 to 2021-22. In this growth cycle:

Output (real GVA) grew 1.5 per cent per year on average.

The main contributors to output growth were:

  • Capital services (0.7 percentage points);
  • Hours worked (0.1 percentage points); and
  • MFP (0.8 percentage points).

The small contribution from hours worked of 0.1 percentage points reflects that through this cycle, a strong trough, influenced by pandemic impacts such as workers ability to work, almost offset hours worked growth in neighbouring years. The new growth cycle also indicates the continuing downward trend in GVA growth. Hours worked and capital services contributions are also at historic lows in this cycle.

Conclusion

Estimating productivity growth over time is difficult, and necessarily involves making a range of assumptions that must be taken into consideration when analysing productivity statistics. This article has outlined some of the major assumptions made, and the analytical limitations they produce, but this list is not exhaustive or complete. 

Quarterly and annual productivity statistics are sensitive to assumptions and are calculated from data subject to revisions, sometimes making it challenging to separate signal and noise in productivity statistics, particularly over the short run. Therefore, it is recommended to consider productivity trends over the long run, or over multi-year growth cycles, to reveal and understand how the way we work, the way we invest, and technological progress impact the productivity of Australia.  

Footnotes

Show all

Back to top of the page