The UK needs a Skills Map
www.nesta.org.uk/blog/the-uk-needs-a-skills-map/
www.nesta.org.uk/blog/the-uk-needs-a-skills-map/
An automated Skills Map would benefit both workers and policymakers
This blog describes a prototype Skills Map, which is available here
With tools like Google maps and Citymapper, getting from point A to point B has never been easier. The same cannot be said for navigating the labour market. Put simply, there is no map.
What would a Skills Map look like? It would show students the skills (streets) they need to reach a particular occupation (town). It would show existing workers the new paths (skills) they need to transition between jobs. And policy makers would use the map to gain a bird’s eye view – enabling them to spot jobs and skills that are emerging, as well as identify those skills that are becoming redundant. Most importantly, the map would be largely automated, providing up to date information and requiring little manual input.
The need for a Skills Map is growing. Recent research forecasts that around 35 per cent of current jobs in the UK are at high risk of computerisation over the next 20 years. To stay ahead of automation, workers will need to become lifelong learners. But to do so, they will need to easily identify relevant new skills. And policy makers will need to design policies that encourage workers to acquire these skills.
The best insights on skills that are currently available come from the Employer Skills Survey (ESS). The survey asks over 91,000 employers about a range of issues relating to skills and training. The survey has two strengths. The first is its emphasis on obtaining a representative sample of establishments. The survey employs a careful system of quotas and weights that allows survey results to be compared over time and across the UK labour market. The second strength is the survey asks about skills shortages, as opposed to just skills demanded, and the former is arguably more important for policymakers.
While the ESS is a useful resource for policymakers, individuals (particularly students) are more likely to turn to career websites to learn about skills and occupations. There is a plethora of such sites – one example being the National Careers Service. These offer job profiles, which tend to list a small set of broad skills that are needed for that job, as well as providing information on topics such as salaries and hours. These job profiles typically require manual updating.
Insights from online job adverts could usefully supplement existing skills information. Job adverts describe the skills that employers are currently searching for in their next employee. While the skills mentioned in any one advert may not be representative, millions of these adverts can paint a detailed picture of the skills required for a specific occupation. In the same way that satellite images provide a ‘first look’ at unmapped areas, online job adverts could act as a base layer to form the Skills Map.
There are several potential benefits of using online job adverts.
Granularity. The skills mentioned in job adverts are more granular than those used in surveys or found on career websites. Workers and policy makers need detailed skills information in order to act. For example, if a worker can discover that they’ll need a specific programming language in their desired job (such as Python or R) they can then find courses that teach that language. If instead they can only find that their job requires ‘programming skills’ then it is more difficult for the worker to take action.
Timeliness and automation. While the prototype Skills Map is based on a static dataset, the map could in principle be automated. When an advert is placed on a job search website the skills contained in the advert could be automatically extracted and fed into the Skills Map. This would ensure that the information in the map is both up to date and independent - coming directly from employers.
In contrast, most career guidance sites are updated irregularly and require an individual’s judgement on the skills and occupations that should be featured. Similarly, the Employer Skills Survey is currently only conducted every two years. This means that new skills may only be recognised with a delay.
This prototype Skills Map aims to show the potential strengths and limitations of such a tool. It was constructed using processed data provided by Burning Glass who collect online job advertisements. The dataset contains 37 million adverts for UK positions placed online between 2012-2016 (inclusive). There is an average of five skills or software programs in each advert and over 11,000 unique skills. The skills are broad and include types of knowledge, abilities and work activities.
The user begins by entering their job title. The Skills Map then shows the skills that are most frequently requested for their occupation group, and compares these to the skills required for all occupations. For each skill, an average salary range is reported, based on the salaries offered in adverts containing that skill. The map also shows the skills that have experienced the fastest and slowest growth in mentions between 2012-14 and 2014-16, as well as the software programmes that are most commonly requested. The map concludes by providing occupations that require similar skills to the user’s own occupation.
The key function of the Skills Map is to provide insights that are tailored towards individual occupations. However, the map also tells us about skill demands at the national level.
Across all occupations, many of the most frequently mentioned terms in job adverts relate to inter-personal skills, such as customer service and team-work. A number of other top skills are core competencies such as writing, mathematics, and problem solving. Others relate to resource management, and they include organisational skills and planning. Finally, some terms could be described as core work activities, such as sales, management, teaching and research.
A rise in the number of times that a skill is mentioned may be driven by several different factors. A wider range of occupations may now require this skill. Or the occupations that use the skill may have experienced above-average growth. Finally, the skill may simply be a new term that has become a more popular way of describing an existing skill.
A number of the skills that have grown fastest fall into two broad groups. The first group of skills relate to caring for others, and they include patient care, care planning, mental health, working with patients who have dementia or learning disabilities. The second group relate to the opportunities and threats that come from living in a more digitally connected world. Skills in this group include big data, digital marketing, social media, information security and firewalls.
On the flip side are those skills that have experienced the largest falls in mentions between 2012-14 and 2014-16. One skill in this group that stands out is ‘basic internet skills’ - whose appearance may seem counterintuitive in an ever more digitised world. It appears because employers are now more likely to take this skill for granted and so no longer need to mention it in job adverts.
The other skills are rather mixed, but upon reflection a number of broader skill groups emerge. One group of skills in decline appear to relate to consumer based finance, and include the likes of financial planning, insurance underwriting, mortgage sales, mortgage advice and financial advising. Another group of skills relate to working with industrial machines, such as mechanical design, machining, computer numerical control and mechanical engineering. A third group of skills are connected to logistics - warehouse management, store management and logistics analysis. A final group relate to using the telephone (perhaps reflecting its decline) and include telemarketing and cold calling.
While job adverts could create a valuable Skills Map, there are a number of hurdles still to overcome in ensuring that the data from job adverts is as accurate and useful as possible.
Not all jobs are advertised online. This makes it impossible to cover all occupations. For the prototype Skills Map a substantial number of occupation groups were excluded because they contained too few job adverts.
Job adverts are not exhaustive. The space in a job advert is limited and this means that employers are unlikely to specify every skill required from an employee. Instead, the skills mentioned are likely to be those that the employer suspect applicants may lack. As a consequence, analysis of job adverts may underplay the importance of basic skills (e.g. mathematics).
Collecting text from websites can be a messy business. It may be difficult to distinguish where one advert ends and another starts, causing the adverts to become merged. It may also be difficult to distinguish skills mentioned in the advert from other words mentioned on the webpage. Essentially, the inherent messiness of a web page and variation across job sites, means that the wrong skills may become associated with an occupation.
In the prototype Skills Map a number of skills were removed from the dataset because they were often incorrectly associated with jobs. However, it is not possible, nor desirable, to check every advert for incorrect skills. A light-touch approach was taken, and it is inevitable that incorrect skills will remain. One method to prevent this problem is to look at the context surrounding a skill-term when it is collected.
The same job advert may appear on multiple sites. If these jobs are double-counted then it would exaggerate the importance of skills in jobs that are advertised more widely. While a lot of effort goes into deduplication, this process is necessarily based on rules. For example, if the same job advert appears on a different website within a given number of weeks, then the second advert is judged to be a duplicate and will be deleted from the dataset. However, the length of time that a job advert remains online will vary by website and on other factors such as economic conditions.
The same word can have several different meanings. One example is the term ‘Chef’ - it may relate to cooking but it may also refer to a programming language called Chef. Acronyms are particularly susceptible to this problem. For example ‘GPS’ is read as both ‘Global Positioning System’ and as General Practitioners (GPs). Most of these issues can be overcome in the raw data, for example, by distinguishing between upper and lower case and by looking at other skills mentioned in the advert.
Given the limitations identified above, there are several ways in which researchers could improve the accuracy and usefulness of job advert data for a UK Skills Map.
Improving occupation classification. Placing job adverts into occupation groups is not simple. At present, the title of the job advert is used to choose the most appropriate occupation group. However, words can have multiple meanings and if enough jobs are consistently misclassified this can give a misleading picture of the skills required for an occupation. For example, an advert for ‘NHS Manager – Salary Band 6’ may be placed into the ‘Musicians’ occupation group due to the word ‘band’ which is also associated with music. If this misclassification happens enough times then a Skills Map may suggest that musicians require medical knowledge.
One way to improve the classification would be to use information inside the advert to place the jobs into occupation groups. Even then however, it is difficult to judge the success of an alternative approach. Even when the classification of job adverts is performed manually people often fail to reach consensus on the ‘right’ occupation group (Schierholz, Gensicke and Tschersich, 2016).
Developing a skills taxonomy. Many of the skills extracted from job adverts are not independent of each other - some are broad skills (such as ‘maintenance’) that encapsulate more specific skills (such as ‘lawn mowing’). This means that the fastest or slowest growing skills can be an odd mix of very specific and very broad skills. Ideally, one would want to group specific skills under broader skills. This is called a hierarchical skills taxonomy. Examples of existing skill taxonomies include ONET and ESCO. The taxonomy would allow analysts to determine whether an increase in mentions of a specific term reflects greater demand for its broader skill group, or simply reflects a change in the preferred term used to describe the skill.
The challenges mentioned above are both being tackled as part of Nesta’s work as a partner in ESCoE, which is the ONS’s new Economic Statistics Centre of Excellence. The Centre’s broader aim is to provide the ONS with research that addresses the challenges of measuring the modern economy.
Creating a weighting system for online adverts. The Employer Skills Survey (ESS) uses quotas and weights to ensure that their findings are representative of UK establishments (that have at least two people on the payroll). There is no weighting system yet for online job adverts. This means that industries that are more likely to advertise online (such as IT), and jobs that have higher turnover, can overly influence metrics such as the ‘most in-demand’ skills. A weighting system would account for variations in the tendency to advertise online and so would ensure that insights were representative. The weights would also allow analysts to construct alternative measures of labour market statistics such as the vacancy rate.
Adding data. Online job adverts show only the skills demanded by employers. Policy makers are arguably more interested in skill shortages. One approach to estimating shortages would be to collect information on the length of time that a job advert remained online. An alternative approach would be to collect data on the skills supplied by job seekers, from online CVs, and compare these to the skills demanded in job adverts. Ideally, the Skills Map would also direct workers to resources where they could find out how to acquire the skills that they have just identified. This information could include links to online courses and approved training providers.
To conclude, online job adverts have the potential to form the foundation layer of an automated Skills Map that could benefit both workers and policy makers. However, there are still a number of hurdles to overcome to ensure that the data is as accurate and useful as possible. It is likely that any map will always require some manual input.
