# DST - Labour Market Insights (LMI) Discussion Group

[TOC]

## Agenda
- Introductions
- Jude Hillary
- Senior Research Director, NFER
- Split into 2 x breakout rooms
- silent reflections/note taking (5 mins)
- address themes in relation to 4 questions (25 mins)
- Feedback and group discussion

## Introductions
:::info
Introduce yourself, your organisation, your current relevant project/research/work and your email if you wish
:::
:::warning
- Bridget, The Alan Turing Institute working on National Skills/secretariat of the DST/keen to explore LMI as a foundation for role descriptors and gaps in organisational data readiness, contact me at [e-mail]
- Ayesha, The Alan Turing Institute. Currently working on the [Data Science and AI Educators' Programme](https://www.turing.ac.uk/data-science-and-ai-educators-programme), contact me at [e-mail]
- Harsh Shah, East Midlands Chamber. Work as data lead (Knowledge Transfer Partnership Associate) to develop regional business intelligence unit (economy, skills, international trade, sustaonability) to bring more investment in East Midlands region. Contact: [e-mail]
- Katharina, Financial Services Skills Commission, my role is to find out what future skills our industry needs
- Alex, statistician/data scientist at DCMS. Published some research on worker movements into/out of the digital sector. Currently looking at how we can distinguish between digital and non digital jobs.
- Mark Fuller, Stakeholder engagement consultant representng Bright Data. I help run their collaborations with UK universities and other education organisations, as well as partnerships related to the NDS (including membership of the NDS Forum)
- Frank Bowley, Head of the Unit for Future Skills, a new team within the Department for Education looking to provide data and information in the education system on what skills and training is needed. We have lots of job and skills data, need more, and need to think how we can get more insight of the data we have. I'm an economist who is mostly not allowed to. [e-mail]
- Emily Eisenstein, Head of Stakeholder Engagement and Research, Unit for Future Skills., DfE. [e-mail] Working at the moment on developing our research programme for the next year, particularly interested in bringing researc insights to policy audiences both in DfE and across departments. Also working on UFS work on future skills.
:::

---

## Discussion
:::info
Consider how the themes below have impacted the development and delivery of your project/research/work in relation to the following four questions
:::

### Theme 1: Acquiring and maintaining data sets
*Purchasing closed data sets, accessing open data sets, scraping, ensuring information is relevant and up-to-date. . .*

1. In your experience, what has worked well in this area? This can include examples that you have seen done elsewhere.

- BN:
- FB: Most of the data sets we use are government admin
- HS: Transparent conversations with stakeholkders about the potential use of data (all parties involved); Exploring datasets and engaging with businesses about the potential usage;
- FB: Job vacancy data widely used and commercialised
- MF: Bright scraped data from around 5 recruitment sites last year in a pilot for colleagues at DCMS that gave a good snapshot of data skills needs based on job titles and specs. It was a fairly rough and ready excercise but gave a good indication of what coyukld be possible as a repeatable scraping excercise.
- As a plug - they are happy to do similar, free of charge others, so get in touch if it sounds interesting!
- KE: ONS are great with user requested data on occupations, very timely turnaround but the data isn't always super recent
- JH: Most of the data we use currently is government data or available vis AKDS or ONS. We don't often buy data.

2. What barriers have you encountered?

- KE: cost is a barrier - EMSI/Burning glass licences are expensive; Adzuna data used to require data science skills due to lack of user interface wich we don't have
- ON: integrating data from multiple sources and across skills supply and demand is challenging for digital - different definitons are used for sectors/ job categories and data is realsied at different schedules (e.g. yearly/ quarterly/ monthly)
- AB: In some cases costs (Burning Glass), in others the length of time required for contractual negotiations limits the use of the final product
- FB: can be difficult to find data sets
- FB: cost is a problem for government when you need large set for multiple purposes and make our data public
- The main issue is the amount of time it takes to get access to government data, but generally you do get it eventually.
4. What is missing/where are the gaps?

- easily accessible vacancy data which can be explored both by industry and by occupation, ideally matched to an established skills taxonomy (there are too many taxonomies out there and many hidden behind commercial licences)
- HS: web scraping becomes a problem when websites structures are changed yearly along with names of datasets and no API service available from the DfE data data; data consistencies - geography, time period;
- FB: common standards
-
5. What can we learn from?
- [enter response here]
- AB: for government data sets - to be more responsive to external analysis/critique - thinking specifically of the reweighting issues in the labour force survey data that were flagged early on but not actioned for quite a while
- initiative by ONS to provide more local geography datay engaging with data users
-
-
_Discussion notes_
- Government data can be quite narrow but private sector data sets are slightly broader. Private sets are monetised in some way - business models don't allow for larger data sets that are used flexibly.
- Two-way collaboration
- Conflict between publicly accesible data (but not always up-to-date) and private data sets. Initiatives where there is useful collaboration. Atzuna data? More initiatives where more up-to-date data is accessible for free, it will support LMI and analysis.
- JH: uses DfE/ONS data sets. Some higher education data sets (e.g. UCAS) charge for data sets and is expensive. Web scraping: only as useful as the data that you can find.
- Web scraping: ask colleges and schools to put their data in a certain format rather than scraping websites. Links to next theme.

### Theme 2: Taxonomies and descriptors
*Understanding employer needs and recruitment trends, machine-classification of skills and roles, consistency of definitions. . .*

1. In your experience, what has worked well in this area? This can include examples that you have seen done elsewhere.
- [enter response here]
- FB: many analytical projects have used pre-existing systems like O-NET
- KE: it's good to have international consistency in taxonomies like through ESCO or O*Net, useful when we can compare UK data to competitor markets
- FB: Possibly Singapore and Australia are good examples of what can be done to produce skills taxnomies.
-
-
2. What barriers have you encountered?
- [enter response here]
- BN: inconsistency of definitions, lack of definitive classifications and inaccurate job descriptions/outdated role understandings
- EE: different taxonomies being developed, and these don't always interrelate. Taxonomies from other countries don't always fit UK labour market or skills system
- AB: inconsistent/inaccurate application of standard classification frameworks.
- KE: inconsistency of definitions, each commercial provider uses a different skills taxonomy, plus there are ESCO, O*NEt - it's hard to find the right balance of granularity and aggregation
4. What is missing/where are the gaps?
- [enter response here]
- FB: Most taxonomies match jobs to skills. There seems less from qualification/training to skills.
- FE: difference between soft skills and knowledge, and soft skills feel vague.
- KE have I accidentally deleted something? wanted to second point about general population datasets missing information on skills
-
5. What can we learn from?
- EE Other countries eg Australia/ US have invested and developed taxonomies ( skills: occupation: qualifications)structuring data. UFS are currently commissioning reseach to build a UK skills taxonomy.
-
-
-
-
_Discussion notes_
- Map skills but found challenging because of the way information is presented: data is mapped in different ways. Knowledge, skills and behaviours: there is no 'standard' for how this is organised.
- Company's house data: no proper validation checks as employees input their own data - barrier.
- What is the way forward on Skills taxonomies? Where is the one that everyone can work with? Some existing examples e.g ONet.
- DfE, Unit for Skills: research project looking at building a UK-specific taxonomy. First phase - investigate methodology. Second phase - build taxonomy. Will be used as _the skills taxonomy_ across government data and then made publically available. Developed in an iterative process.
- Can we map this back to compare to other markets? Not yet considered but good question and will take back to research team leading on this work. Might be able to hone in on where the significant differences are.
- Some taxonomies developed for other labour markets don't always fit the UK market.
- Lots of work happening in career guidance too, especially around a common language.

- [ONS Adzuna dataset](https://www.ons.gov.uk/economy/economicoutputandproductivity/output/datasets/onlinejobadvertestimates)
- Here's [Cedefop data](https://www.cedefop.europa.eu/en/tools/skills-online-vacancies/skills/occupations) which allows for cross-EU comparison based on jobs ads, based on Esco or O*Net data (this still includes UK data at the moment)

### Theme 3: Synthesising and visualising data
*Identifying audiences and user-specific insights, developing dashboards/interfaces between the data and its intended users. . .*

1. In your experience, what has worked well in this area? This can include examples that you have seen done elsewhere.
- [enter response here]
- BN: interactive visualisations/dynamic sites
- HS: Discussions with different audiences - Careers Enterprise, Businesses, Schools and Colleges etc.; Development of user journeys;
- EE: making data available that others can draw into their own data tools/ visualisations
- Australia JEDI dataset to bring together jobs and skills data together in a common framework
-
2. What barriers have you encountered?
- [enter response here]
- BN: different needs and preferences for different audiences
- FB: It is difficult to join together different data sets (e.g. skills to jobs)
- FB: Where does the data sit? Platforms vs open data
- HS: Experimentation and validating the user interactions
-
4. What is missing/where are the gaps?
- [enter response here]
- BN: identifying and highlighting career pathways for individuals
-
- FB: Data infrastructure/ecosystem to bring job and skills data
-
5. What can we learn from?
- [enter response here]
- BN: Cath Sleeman at NESTA, something very interesting in the pipeline re: storytelling
-
-
-

_Discussion notes_
- HS: User journeys approach (different users in 6 different, specific groups): look at different dashboards depending on their group/data. Track users and improve user journeys - engage with actual users and also by validating/tracking it from the user journeys on the website.
**- DfE: small datasets brought together makes a more powerful data set. How are we bringing all of this data together to link taxonomies, infrastructures? Australia has something called JEDI (jobs and education data infrastructure).**
- Would like to follow up on this at a later date.
- DfE: on strategic level, trying to establish all the needs and cater to these.

### Theme 4: Delivering and measuring impact
*Gathering data on interactions with the information, how intended audiences are engaging, improving accessibility. . .*

1. In your experience, what has worked well in this area? This can include examples that you have seen done elsewhere.
- [enter response here]
-
-
-
-
-
2. What barriers have you encountered?
- [enter response here]
- BN: effectively and ethically collecting data on users - still unsure of best methods to do this?
-
-
-
-
4. What is missing/where are the gaps?
- [enter response here]
-
-
-
-
5. What can we learn from?
- [enter response here]
-
-
-
-
-
_Discussion notes_
- HS: don't want users to misinterpret data, so they see on the dashboard info on the data (when it was updated etc). Once they've seen these, they can go into the dashboard. Helps ensure that nothing is miscommunicated. Backend: when scraping data, they do lots of testing. Communicate back to users too.
- Transparency is key.
---

## Conclusions