What we did
By 2030 the job market will look dramatically different. Previous Nesta research has predicted that about 10 per cent of workers are in occupations that are likely to grow as a share of the workforce and 20 per cent will shrink. As for the remaining jobs, their outlook is more uncertain. Although unsettling, this disruption needn’t be disastrous for the workforce. There is an opportunity for employees in uncertain or shrinking occupations to improve their prospects by investing in the right skills.
Policymakers consider digital skills to be a top priority for investment. They are seen as offering people greater employability and job resilience. But are all digital skills created equal?
Our analysis, described in detail below, shows that not all digital skills will be equally important in the future. In fact, occupations which we are more certain will have poor prospects, are more likely to require a digital skill than the occupations that are most likely to grow by 2030. This is because the relationship between the digital intensity of an occupation and its potential for growth is not straightforward: there are occupations that are currently not digitally intensive, but are expected to grow in the next 10-15 years, as varied as teachers and chefs. The type of digital skills needed in a job also makes a difference: the digital skills most likely to be needed in growing occupations are ones that are used in non-routine tasks, problem-solving and the creation of digital outputs.
This is exploratory analysis that takes a novel approach. At Nesta we will continue to study the demand for skills and the future of work, which is in itself a shifting landscape.
Job adverts can help us to measure the demand for digital skills
To find out which digital skills will be most needed in the future, we investigated employer demand for digital skills using online job adverts. Online job postings are a great source of near real-time information on the labour market. In addition to vast volumes (our dataset contains 41 million UK adverts), job adverts also offer more granular data than skill surveys as they allow employers to describe their skill needs more precisely.
The data that we used for the analysis was collected between 2012 and 2017 by Burning Glass Technologies, a labour market analytics software company. For each job posting, Burning Glass identified and extracted skills that had been mentioned in the advert. There are 11,425 unique skills mentioned across all job adverts. In this context, ‘skills’ is a broad term that refers to all employer requirements including knowledge and competences.
We use machine learning methods to identify digital skills
The definition of digital skills can vary depending on the objective, audience and context and this makes it challenging to determine which skills are digital and which are not. For this analysis, we start with a list of skills that are actually software programmes, such as Maya or PeopleSoft. Next, we add skills which tend to be mentioned alongside software skills in job adverts. Examples of these skills are 3D animation, Data mining and Agile development. To find these skills we use a machine learning technique called word embeddings*. We train a word embeddings model on all job adverts and generate vector representations of skills, which we can then use to measure skill similarity. Two skills will have a high similarity if they tend to co-occur in the same job advert. For example, Frontend technology is often mentioned in the same adverts as the software framework Angularjs, so we add the former to the digital skills list. In total, we identify 1,358 digital skills: 756 because they are types of software and 602 additional skills via word embeddings.
Given the large number of digital skills, we group them into clusters using a skills taxonomy, which we have developed as part of Nesta’s contribution to a research centre called ESCoE (Economic Statistics Centre of Excellence).2 In some instances, the name of a skill cluster might refer to a broader range of skills than just digital. For example Taleo or Applicant tracking systems are digital skills used in HR Management which is a cluster that also includes non-digital skills. In these instances, we only focus on the digital skills within the cluster. See Table 1 for examples of the most popular digital skills in each skill cluster.
We focus on occupations most likely and least likely to grow
In the Future of Skills report, Bakhshi et al., estimate the probability that individual UK occupations will grow or shrink, as a share of the workforce, over the next two decades. They arrive at these estimates using a combination of expert judgement and machine learning. The authors find that around 30 per cent of workers are in occupations either highly likely to grow as a share of the workforce (probability of growth >= 0.70) or highly likely to shrink (probability of growth <= 0.30). The prospects for the remaining 70 per cent of jobs are much more uncertain. Our analysis focuses just on the first two groups in order to better contrast digital skill needs across occupations.
In order to explore how demand for different digital skills varies between the two groups of occupations we fit a logistic regression model and analyse which skill clusters are most closely associated with each group.
*Jurafsky, D. and Martin, J.H. (2017) ‘Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition.’ London: Pearson.