How much does the automatability of your job affect your employment prospects? Roughly half of US jobs might be a risk from automation. However, a worker’s employment opportunities do not depend solely on the number of available jobs in their current occupation, but also on the availability of jobs they can apply for in other occupations – their ‘occupational mobility’. Our recent study suggests that transition possibilities are as important to consider as automation risk.
For example, according to previous estimates, statistical technicians are likely to be replaced by automation. Yet, if statistical technicians were displaced, their existing skills would most likely allow them to transition into other mathematics-related jobs, which have lower automation risk (see figure 1).
In contrast, although childcare workers may not be directly displaced by automation, this does not mean their employment opportunities are unaffected. As automation displaces people in other occupations, many of these workers could apply for childcare jobs and may consequently increase competition for, and reduce the job opportunities of existing childcare workers.
In other words, even though the direct risk of automation may be larger for statistical technicians, once we account for possible occupational transitions and labour demand reallocation, childcare workers could face a greater risk of unemployment.
Estimates of automatability in the occupational mobility network. (A) A histogram of the probability of computerization for occupations. (B) The occupational mobility network, where nodes represent occupations and links represent possible worker transitions between occupations. Red nodes have high automatability, and blue nodes have low automatability. The size of the nodes indicates the logarithm of the number of employees in each occupation.
To understand workers’ job opportunities at an occupation level, one can use networks to identify and visualise occupations that share skills and work activities. In turn, these characteristics allow us to understand how a worker can transition from one occupation to another and help design career pathways. We built on these studies and designed a network model of the labour market to better understand the impact of automation on employment.
So what do occupation-level employment prospects look like once we take both automation risks and occupational mobility into account? On average, we can expect workers in occupations with decreasing labour demand to face higher unemployment spells. However, due to differences in occupational mobility, workers in occupations with very similar automation risk can end up facing very different unemployment rates and spells. For example, dispatchers and pharmacy aides face roughly the same automation risk, but while pharmacy aide workers are likely to decrease their unemployment spells, our modelling suggests dispatchers will most likely increase theirs (see figure 2).
Impact of automation on unemployment and long-term unemployment at the occupation level. (a) The percentage change in the unemployment rate versus the automation level for each occupation, and (b) the same thing for the long-term unemployment rate.
In addition, since low-wage occupations are at a higher risk of automation than high-wage occupations, there has been concern that automation will increase inequality. Occupational mobility is unlikely to solve this issue, as low-wage occupations at risk of automation tend to be clustered: in other words, transition possibilities for low-wage workers losing their job tend to be in occupations that also have low wages and a high automation risk. These bottlenecks may lead to higher rates of unemployment and, concerningly, higher rates of long-term unemployment.
Of course, these results depend strongly on estimates of the occupation-level risk of automation, which are highly uncertain. But they do suggest that targeted retraining policies might be necessary. Our research found there is scope for retraining policies to leverage the occupational mobility network structure to reduce displaced workers' unemployment spells. Support and skills development programmes should be directed towards workers who are more likely to face longer periods of unemployment, considering available options of occupational mobility, rather than simply to workers who face a direct threat of job loss. Our model can help policymakers identify who these workers are.
The Covid-19 pandemic has caused a global recession with unequal impact across workers of different occupations. To avoid a Jobless Recovery, it is therefore even more important that worker support be targeted effectively.