Dr. Cath Sleeman and Jyldyz Djumalieva

Our recent manifesto on the future of work called on the Government to adopt a live skills taxonomy which would identify the skills needed for different jobs. This latest research demonstrates the potential of such a taxonomy by showing employers and policy makers how workers’ skills vary across the UK. The analysis uses data from 53 million job adverts and our own static skills taxonomy, developed within ESCoE. We also discuss how job adverts might help to inform regional skill mismatches.

This project was supported by the Productivity Insights Network (PIN) which is funded by the Economic and Social Research Council (ESRC). Read the full discussion paper.

Why measure skill mismatches?

Skill mismatches arise when the skills that employers need don’t match the skills of workers and job seekers. Mismatches are costly to firms, workers and governments. Firms may have to retrain under-skilled workers, and struggling to find suitable workers may raise their recruitment costs. For job seekers, a lack of suitable skills may result in spells of unemployment. Workers who are over-skilled may experience reduced job satisfaction and suffer wage penalties. At a national level, skill mismatches can reduce productivity and lower economic growth.

Existing methods for measuring skill mismatches

While skill mismatches can be measured in several different ways, no single measure is perfect, and there is room for innovation. The ideal measure would produce timely estimates that are objective, detailed and accurate. In the UK, the leading measure of skill mismatches comes from the Employer Skills Survey (ESS). Its strength lies in using a large and representative sample of UK businesses. However, the survey is a major undertaking and it is only run once every two years, with a year’s delay between the survey being in the field and the results being released. Moreover, the survey can only tell us about skill mismatches in very broad areas, such as ‘advanced or specialist IT skills’ which may make it difficult for course providers to change their offerings, and for learners to adapt their choices, in a way that would counter skill mismatches.

Alternative approaches to measuring skill mismatches include using macroeconomic indicators, such as wage growth and employment rates (where sharp wage growth might indicate a skills shortage). However, these metrics can be influenced by other factors, such as union membership, and so they are noisy indicators of skill mismatches.

Another approach is to measure the under and over-qualification of employees. While qualifications may be an objective indicator of skills, an employee’s past qualifications are not necessarily an accurate indicator of their current skills. The success of this approach also relies on how we determine the ‘right’ qualification level for each role.

Can big data help?

This study used a dataset of 53 million online job adverts, collected between January 2012 and December 2018, and provided by Burning Glass Technologies. For each advert we have information on the job title, salary, location, education and experience requirements mentioned within the advert. We also have a set of approximately 10,500 keywords that were extracted from the adverts. We refer to these as 'skills' and they include the likes of ‘communication’, ‘management’ and ‘listening’.

Measuring skill supply

The dataset of job adverts was used to identify the skills required in different occupations, which enabled us to estimate the stock of skills potentially held by workers.

To map the skills required in different occupations, each job advert was first sorted into an occupation group (from the ONS’s classification of occupations which contains approximately 360 occupations). Assignment was based primarily on the job title in the advert, and we only used those adverts that could be confidently assigned to an occupation. The next step was to identify the mix of skills required within the assigned adverts. We grouped these skills using our taxonomy, in which the 10,500 keywords have been clustered into 135 skill groups.

After mapping skills to occupations, we then used official statistics on ‘regional employment by occupation’ to estimate the stock of skills demanded by employers in each region. This is a measure of ‘potential’ rather than ‘actual’ skill supply as it assumes that every worker is perfectly skilled for their current position (and that the skills required for an occupation don’t vary). This assumption is not overly limiting as the ESS suggests that only 4.4% of the workforce lack full proficiency at their job (p14). The geographical units that we use are Travel to Work Areas (TTWAs) whose boundaries are drawn to capture local labour markets, where the bulk of their resident population work within the same area.

Limitations of using job adverts to measure skill supply

Job adverts are not without limitation. First, the skills listed in an advert may not accurately reflect the skills required for the job. Second, not all vacancies are advertised online. And without knowing the ‘true’ number of vacancies for each occupation (official vacancy statistics aren’t broken down by occupation) and the ‘real’ skills required, it is difficult to know how severely these limitations affect our estimates. Other methods for measuring skills suffer similar drawbacks, in so far that employers may misjudge the skills that they need, and employees may misreport their own skills.

Another data limitation relates to measuring regional employment. The 2011 Census is the latest available source for providing estimates of employment across TTWAs for the detailed set of occupations that we use. We take these employment estimates and adjust them to ensure that the distribution of employment across broader occupation groups in each region matches more recent data. And finally, owing to the small populations in some TTWAs, we focus our analysis (below) on just those TTWAs that have more than 250,000 economically active residents.

How does the mix of workers’ skills vary across the UK?

The data visualisation below shows the distribution of potential skills held by workers who live in the UK’s most-populated Travel to Work Areas (TTWAs).

In order to compare the skill mixes, we first calculated the distribution of potential skills (held by workers) in each region, and then normalised these results by the same distribution for the whole of the UK. This produced a set of location quotients for each skill and in every region. For a given skill, the coloured bar shows the interquartile range of the location quotients for that skill in the 32 TTWAs, which each have an economically-active population of over 250,000. If the bar sits to the right of the ‘UK wide’ line, it indicates that these larger TTWAs have a greater prevalence of that skill than the rest of the UK. If the bar sits to the left of the line, it indicates that the larger areas don’t tend to have a large concentration of that skill. The place names in each row (and corresponding dots) indicate the areas (and their location quotients) with the greatest concentration of that skill among the largest TTWAs.

As discussed, there is substantial uncertainty around these estimates, and they should be viewed as indicative only. They do however provide a sense of the detailed regional estimates of skill supply that might be possible with further refinement of the method.

Notes: A log scale is used to show the location quotients along the x-axis. The darkest portion of each bar indicates the median location quotient amongst the largest 32 Travel to Work Areas.

Measuring skill diversity

To summarise the mix of skills in each region, we created a ‘skill diversity’ metric. This novel indicator was inspired by measures of species diversity from the field of ecology. A higher score for skill diversity indicates that the region either has a potentially wider range of skills (or a greater abundance of several rare skills) compared to other areas. Greater economic diversity may help to mitigate the effect of economic shocks.

Not forgetting the various limitations of the approach (described above), five locations with high skill diversity (and an economically active population over 250,000) are Guildford and Aldershot, Reading, Cambridge, Slough and Heathrow, and London. Several of these areas have a relatively greater supply of IT skills than other large areas. Reading also has a concentration in data engineering skills. Brighton and London have a predominance of skills in design, PR and marketing, while Cambridge has a clear specialty in science.

Can we accurately measure skill mismatches using job adverts?

Having mapped the potential supply of skills from current workers, we can in theory compare this to the flow of skills demanded in job adverts. We trial this approach in the research paper, contrasting the potential supply of a particular skill (as a proportion of all skills supplied) with the demand for that skill from job adverts (as a proportion of all skills demanded).

These metrics cannot be used to directly estimate skill mismatches because we have assumed that workers are ‘perfectly skilled’ for their jobs. Instead, we are effectively comparing employers’ stock of skill demands with their flow of skill demands. If the flow of demands looks radically different from the stock, then it suggests that skill-needs are changing and so this measure may still help to signal a forthcoming skill shortage.

Unfortunately, it may be a noisy signal as there are a number of extraneous factors that can affect the flow of skill demands from adverts. First, differences in turnover rates between occupations may distort the apparent demand for skills. Specifically, a skill required in a high-turnover occupation will appear to be in greater demand than a skill required in an occupation with a lower turnover rate. The solution is to estimate the stock of job adverts by obtaining information on the number of days that each advert was live. This should capture not just the variation in churn, but also variation due to location and the time of year.

Second, differences in the likelihood of an occupation being advertised online may also skew the results. For example, many new teaching roles are not advertised and are instead filled directly through teacher training programmes. If we use job adverts to measure the demand for teaching skills, we may be left with the erroneous impression that there is an oversupply of these skills. This limitation could be tackled by gathering information to estimate the proportion of an occupation’s vacancies that are typically advertised online, and using these to adjust our estimates.

A third, and more fundamental challenge, is that a large number of unfilled vacancies does not necessarily indicate a skills shortage. These unfilled vacancies may instead reflect poor working conditions or low wages. Similarly, a lack of demand for a particular skill (within job adverts) does not always mean that the skill is in ample supply. High recruitment costs may be forcing employers to retrain existing staff rather than hiring new recruits.

And finally, comparing the aggregate demand and supply of a skill will not tell us about the quality of individual matches. We cannot determine whether individual workers have the appropriate skills needed for their positions. Assessing that would require access to the personal data of employees.

In summary, comparing the potential skills supplied by workers with the skills being demanded by employers (from adverts) may give some insight on forthcoming shifts in skill mismatches that are more granular and timely than current metrics. On the flip side, we acknowledge that the measure is noisy, further refinement is required, and any findings should be corroborated with other metrics.

The next step: creating a national skills taxonomy

A key ingredient in this research has been our skills taxonomy. The taxonomy provided a standard approach for grouping skills and enabled us to link skills to occupations. Identifying these links would allow students, workers and job seekers to better understand both the skills needed for different jobs, and the ‘extra skills’ required to transition between occupations.

If a taxonomy was created, the most important aspect would be keeping it up to date. The Government would need to secure ongoing access to an open dataset of job adverts, which would enable them to track how skills are changing. Keeping the taxonomy ‘live’ would help to ensure that it is trusted and widely adopted by other users of labour market information.

Our recent manifesto contains more recommendations for supporting workers. Improving the breadth and depth of labour market intelligence is just one way to help workers navigate their way to better jobs.