Appendix 2

Modelling the labour market impact

Our modelling framework is designed to explore the implications of technological change on the shape of the Australian labour market. It is based on a long-term assumption of equilibrium employment and therefore does not attempt to forecast fluctuations in the rate of unemployment. Our model makes use of granular industry-occupation employment matrices sourced from the Australian Bureau of Statistics, which represent today’s labour market equilibrium. This is combined with data from O*NET on the nature of occupations, providing a comprehensive account of the skills (35 categories) and workplaces tasks11 (41 categories) for each job, as well as other characteristics12. Our scenario projections represent a shift in the structure of the labour market compared to today’s equilibrium. Our approach involves a sequence of two modelling exercises, which we set out below.

Modelling the static displacement effect

The displacement effect was derived from consensus expert judgements in an interactive workshop, as described in Box 1. The workshop produced quantitative assumptions about the extent to which technology would make us more productive in the future in producing the equivalent of today’s level of output. In doing so, it would “displace” workers from performing specific tasks that are more amenable to automation. These “task changes” were mapped to occupations based on the task-profiles of 433 occupations13 in the Australian labour market, which were produced using O*NET data.

The task profile of a given occupation14 ( ) provides an estimate of the share of working time spent completing each workplace task ( ). It is calculated using the relative importance ( ) of each workplace task, only including tasks that are deemed important for that occupation. The importance score is derived from O*NET data, which gives each task for each occupation a score from 1-5. We normalise this score on a 0-100 scale and label all tasks with a score greater than or equal to 50 as “important” to a given occupation. By way of context, the 433 occupations involve an average of 23 “important” tasks. Cleaners, for example, have six important tasks to complete, whilst conference and event planners have 35. The estimated share of working time spent on task 1 for occupation is therefore as follows:

Our expert panel produced task-specific assumptions about the change in FTE employment required to perform each task. This resulted in a new occupational task profile, where is the static displacement effect for task.

The implied impact on FTE employment from the displacement effect was calculated for each occupation ( ) based on the gross reduction in FTE hours required to complete each task under the technology scenario, compared to baseline (3). The labour market-wide impact was calculated as the sum of the occupational effects (4).

¹Referred to as Generalised Work Activities in O*NET
¹² All O*NET data is mapped from US Standard Occupation Classifications to ISCO 08 in order to be incorporated with the industry-occupation employment matrices
for Australia.
¹³4-digit occupations in the International Standard Classification of Occupations 2008
¹⁴ O*NET data is not industry specific, therefore the displacement effect is the same across industries for a particular occupation

Whilst the expert panel dictated that most tasks would be subject to productivity gains over the next decade, i.e. a negative displacement effect, some tasks would see an increase in demand for workers as a necessary corollary of achieving those productivity gains. An example is that workers would have to ‘interact more with computers’, in order to achieve technology-based productivity gains in other tasks, such as ‘inspecting equipment’. These “positive displacements” are interpreted differently to negative displacements. They are interpreted solely as a change in the task-composition of an occupation, which results in a rise in aggregate hours the worker spends on that task at the expense of other tasks, such that the net FTE employment for the occupation is unchanged. Therefore, the total displacement effect for each individual occupation is always non-positive. For interacting with computers specifically, an additional assumption is imposed, stating that this task becomes important for more jobs, based on an adjustment in the importance threshold for this task.15

Modelling the income effect

The income effect occurs as a result of increased demand for goods and services and offsets the displacement effect on employment. Where that demand falls, i.e. how people spend their extra income – is independent of where technology is generating productivity gains. To estimate the income effect, we took GVA growth forecasts for Australia from the Oxford Economics Global Industry Model and derived the implications for individual occupations using the industry-occupation employment matrices.

In the real world, the displacement and income effect occur in parallel. However, technically we estimated the displacement effect of our technology scenario first. We take the post-displacement employment for occupation o in industry ( ), then apply the GVA forecast for industry ( )16 and sum across all industries to estimate the net change in employment level for occupation o ( ).

The aggregate income effect on FTE employment was calculated as the sum of all occupational impacts and, for modelling purposes, was constrained in FTE employment terms to equal the aggregate displacement effect, as expressed in (6). In fact, the income effect exceeded the displacement effect, but the residual is presumed to be driven by other, non-technology related productivity gains that are correlated to digitalisation.

Skills Matching Model

The Skills Matching Model was designed to match vacant positions in the labor market with viable candidates, iteratively filling the vacancies until no surplus workers remain and equilibrium is restored. The matching algorithm considers both the viability of a move from one occupation to another in terms of job compatibility, as well as the wage dynamics that would play out in practice, to incentivize the movement of labor in the evolving economy.

Changes in the demand and supply of labor brought about by the displacement and income effects will affect wages in each occupation. For occupations that experience high growth in labor demand, wages will be expected to rise and the probability of workers in those occupations wanting to move will fall. For occupations that experience a large reduction in labor demand, the opposite will happen: workers in these occupations will be willing to accept lower wages and they will be more likely to move jobs.

¹⁵ Importance threshold is lower to 25 (out of 100)
¹⁶% growth in GVA over the next 10 years (2018-2028)

These dynamics are embedded in the historical job moves data to an extent but given the uniqueness of the labor market “shock” we are modelling (in which we process ten years’ worth of technological change concurrently), the wage dynamics surpass those experienced in the past. For that reason, we incorporate a Wage Adjustment Factor( WAF ) into the model equation. The WAF for a sending occupation (i), in the context of a move to a receiving occupation (i), reflects the difference in wage pressures between the two (%^Emp, -%^Emp,), relative to the wage pressure being felt in other, competing, sending occupations.17 When the difference in wage pressure between a receiving and sending occupation is greater than the average wage pressure differential related to that receiving occupation, the WAF will be greater than one, and the likelihood of that move taking place will therefore be greater than the historical data suggests. When the relative wage pressure is smaller than average for a move to the receiving occupation, the WAF will be less than one and the likelihood of the move taking place will therefore be lower than historical data suggests.

The model solves iteratively by filling one vacancy18 at a time and recalculating the relative wage-adjustment factors after each iteration. Each iteration starts by taking the occupation with the greatest upward wage pressure (i.e.the greatest proportion of vacancies), calculating the relative wage-adjustment factors for all potential sending occupations and therefore deriving the probability that the vacancy is filled by those occupations. Once that vacancy is filled, the wage pressure experienced by that occupation will fall. Therefore, the wage pressure is recalculated, and the next iteration begins. The model will step through each iteration until all vacancies have been filled and the labor market has returned to equilibrium.19 As wage pressures dissipate, the WAF tends to one, meaning the probability matrix tends back to historical probabilities

¹⁷ Relative wage-adjustment factors are constrained so the probabilities sum to 100%
¹⁸ Due to computational constraints 10% of the initial required vacancies for a given occupation are filled in each iteration rather than one single vacancy.
¹⁹ In order to prevent workers moving back and forth between occupations, a rule was set up that prevents a vacancy being filled by an occupation with a similar wage pressure.