This dossier used to open with a promise: that the OECD's 2025 skills report, read the right way, would tell a candidate which skills to learn before the market moved. It then carried more than twenty figures in support, credited to the OECD, the World Economic Forum, LinkedIn, McKinsey and others. When we opened the documents those figures were credited to, most of them were not there, and the report at the centre of the argument turned out to be about something else.
The OECD Skills Outlook 2025, published on 9 December 2025 under the title Building the Skills of the 21st Century for All, is not a forecast of skills demand. It is a measurement of who gets to build skills, and of what those skills are worth once built. Its first finding is that socio-economic background is the strongest and most pervasive driver of skills disparities. That is a harder finding than the one this dossier used to print, and a more useful one, because it tells a candidate where the gap opens and which part of it can still be closed.
This is the rewritten dossier. It rests only on what eight dated documents print, and it reaches one conclusion the evidence supports: the skills gap opens before the first job, it is carried into the labour market through education, adult learning and measured skills, and those three channels are the only part of the gap a candidate can act on.
What the report actually measures
The Skills Outlook draws on the 2023 Survey of Adult Skills, which the OECD published a year earlier, on 10 December 2024, as Do Adults Have the Skills They Need to Thrive in a Changing World? That survey tested adults in 31 countries and economies in literacy, numeracy and adaptive problem solving. It is a test, not a questionnaire about confidence: adults sit tasks and are scored on a common scale.
The survey's headline results are plain. On average across participating OECD countries, nearly one fifth of adults are low performers, scoring at or below Level 1 in all three domains. France is one of eleven countries that consistently perform below the OECD average in all skills domains, with an average of 255 points in literacy and 248 points in numeracy, against OECD averages of 260 and 263.
The finding the Outlook builds on is the one about origin. Having at least one tertiary-educated parent is associated with an advantage of 50 points in literacy, 49 points in numeracy and 42 points in adaptive problem solving, compared with adults with no parent who attained upper secondary education. Since the previous assessment, the literacy gap between adults with low and highly educated parents has widened in half of the countries.
The gap measured before any job
50 points
literacy advantage of adults with at least one tertiary-educated parent over adults whose parents did not attain upper secondary education
Measured on a test, not a self-report, across the 31 countries and economies of the 2023 Survey of Adult Skills. The same comparison gives 49 points in numeracy and 42 points in adaptive problem solving.
Source 1 McKinsey & Company
Where the old reading went wrong
The earlier version of this dossier treated the Outlook as a demand forecast and borrowed the vocabulary of an earlier edition, whose subject was the green and digital transition. It credited the report with a count of jobs to be transformed by 2030 and with a share of adults holding basic digital skills. Neither figure is printed in the 2025 report. It described the report as covering the OECD's member economies, when the evidence underneath it is a survey of 31 countries and economies. We have withdrawn those figures, each with a dated note saying what the document actually prints.
The same happened across the source list. Figures credited to the OECD Employment Outlook 2024 on training budgets, staff turnover and productivity do not appear in that report; the only 2.5% figure we found in it concerns carbon tax revenues as a share of GDP. A curriculum figure credited to Education at a Glance 2024 does not appear there either. Figures credited to McKinsey, the International Labour Organization, Indeed and a LinkedIn skills report had no citation row that carried them, or a row that opened a collection page instead of a document. The cybersecurity gap was printed as a round number smaller than the one ISC2 published. All of them are listed on the withdrawals page, and the dossier's panel says so quietly at the foot of the page.
What remains is narrower, and it is enough to reason with.
How the gap travels into pay
The Outlook's second finding is about what happens to the gap once people start working. On average, adults from tertiary-educated families earn 11% more per hour than peers with similar socio-demographic characteristics whose parents are not tertiary educated. The difference falls to less than 1% once the analysis also accounts for educational attainment, participation in non-formal learning and measured skills. In the report's own summary, over three-quarters of socio-economic wage disparities can be explained by differences in learning trajectories and core skills.
Employment follows the same pattern. Adults whose parents did not attain tertiary education are less likely to be employed, 78% against 86%, but there is almost no difference after controlling for their own education and skills.
The arithmetic matters more than either number on its own. Of an 11 point hourly advantage, more than 10 points disappear once three things are held equal: how far someone went in education, whether they kept learning as adults, and how they score on a skills test. Family background is not paid for directly. It is paid for through those three channels.
Two OECD documents, two earnings questions
A candidate who reads earnings figures from the OECD needs to know that two different questions are being answered, and that the answers are not interchangeable.
Education at a Glance 2024, published on 10 September 2024, answers the descriptive question: how much more do people with a given qualification earn than people with upper secondary education, before anything else is held equal. Across OECD countries, workers with a tertiary education earn on average 56% more. Adults with a short-cycle tertiary degree earn 20% more, those with a bachelor's or equivalent 42% more, and those with a master's, doctoral or equivalent degree 90% more. The advantage grows with age: 25 to 34 year olds with tertiary attainment earn 39% more, and the difference rises to 68% among 45 to 54 year olds.
The Skills Outlook answers the adjusted question: what does one more unit of a skill or a qualification add when other things are held equal. Its answer is smaller, because much of the raw premium belongs to the things that travel with a degree. A one standard deviation increase in numeracy proficiency is associated with about 5% higher hourly earnings. An upper secondary qualification with a vocational orientation adds 3% over otherwise similar adults who did not complete upper secondary education.
Two earnings questions, answered by two OECD documents
| Measure | Value as printed | Publisher |
|---|---|---|
| Tertiary against upper secondary, all workers, raw | 56% more | OECD |
| Short-cycle tertiary, raw | 20% more | OECD |
| Bachelor's or equivalent, raw | 42% more | OECD |
| Master's, doctoral or equivalent, raw | 90% more | OECD |
| Tertiary, aged 25 to 34, raw | 39% more | OECD |
| Tertiary, aged 45 to 54, raw | 68% more | OECD |
| Tertiary-educated family, similar peers, per hour | 11% more | OECD |
| Same, after education, adult learning and skills | less than 1% | OECD |
| One standard deviation more numeracy, per hour | about 5% more | OECD |
Sources 2 Organisation for Economic Co-operation and Development · 4 Organisation for Economic Co-operation and Development
The worked reading is short. The raw tertiary premium of 56% is the price the labour market puts on everything a degree carries with it, including the family that paid for it. The adjusted figures isolate a narrower thing. Between the ages of 25 to 34 and 45 to 54, the raw premium moves from 39% to 68%, an increase of 29 points: a qualification's value compounds over a career rather than being paid once at hiring. The adjusted numeracy figure says what a candidate can still buy after the degree is settled: a measurable skill, paid, in the OECD's adult-skills reading, at about 5% an hour for each standard deviation.
Why adult learning widens the gap instead of closing it
If the gap travels through education, adult learning and skills, adult learning should be the channel that closes it. The Outlook finds the opposite. Participation in non-formal learning is about 43% among adults on average, but 61% among tertiary-educated adults and 19% among those with below upper secondary education. Adults from disadvantaged backgrounds are over-represented in courses on operating machinery, equipment and security protocols, while those from advantaged backgrounds are more likely to take upskilling courses such as project management, foreign languages and numeracy.
That is the mechanism of the whole report in one table. The people who already hold the most skill are the ones who keep building it, and they build the kind that pays. The Employment Outlook 2024 reaches a parallel conclusion from a different direction: in its reading of the net-zero transition, low-skilled workers will require substantially more reskilling than high-skilled workers to move out of emission-intensive occupations.
What the demand documents can add
None of this makes demand irrelevant. It makes it a separate question, answered by different documents.
The World Economic Forum's Future of Jobs Report 2025, published in January 2025, surveys employers. Analytical thinking remains the most sought-after core skill, with seven out of 10 companies considering it essential in 2025, followed by resilience, flexibility and agility, and by leadership and social influence. Employers expect 39% of workers' core skills to change by 2030. This is a statement about what employers say they will pay for, not about what any one employer is hiring for next quarter.
The Outlook's own reading of technology is careful in the same way. It cites research estimating that as many as 80% of the US workforce could have at least one in ten tasks affected by large language models. The research it cites, by Tyna Eloundou, Sam Manning, Pamela Mishkin and Daniel Rock, first published on 17 March 2023, says in its abstract that it does not make predictions about the development or adoption timeline of such models. Exposure is not displacement, and a share of tasks is not a count of jobs.
The two platform measures in the source list read the same way. ISC2's first look at its 2024 workforce study, published on 11 September 2024, puts the global cybersecurity workforce gap at 4.8 million, up 19% on the year, against an active workforce that stalled at 5.5 million. LinkedIn's Global Climate Talent Stocktake of September 2024 finds that 7.7% of job postings on its platform were for a green job or required green skills, up from 7.3% in 2023, and that demand for green talent grew 11.6% from 2023 to 2024 while supply grew 5.6%. Both are real measurements. Both are global stocks with no place and no employer attached.
Where the two lines of evidence meet
Put the demand documents beside the supply documents and one reading survives both. Employers rank analytical thinking first. The OECD measures the capacity to reason with written and numerical information directly, finds it distributed by family background before anyone applies for a job, and finds that it is paid for through education, adult learning and measured skills.
So the most useful thing a skills report can tell a candidate is not which skill is fashionable. It is which part of their position is already fixed and which part is still for sale. Parental education is fixed. The qualification a candidate already holds is mostly fixed. Measured proficiency in numeracy and literacy, and the decision to keep learning the kinds of skills that pay, are not.
What the evidence does not show
The evidence has limits that a careful reader should keep in view.
- It does not rank jobs. None of the eight documents says which occupation a candidate should enter, in which city, in which year.
- It does not prove cause. The adjusted earnings figures hold other measured things equal. They do not show that raising one person's numeracy score will raise that person's pay by the same amount.
- It is an average. The survey covers 31 countries and economies, and the shares range widely: the share of low performers runs from 7% in Japan to 44% in Chile.
- It is dated. The survey was fielded in 2023, and the demand documents describe 2024 and 2025. A reader in a later year should look for the next edition before reusing any figure here.
What this means for a candidate
Read the demand documents to choose what to learn, and read the OECD's measured returns to decide how much it is worth. The World Economic Forum tells you that employers prize analytical thinking. The Skills Outlook tells you that the version of it the labour market can observe, numeracy and literacy on a scored test, is paid for, and that the payment is modest per unit and compounds over a career.
Treat adult learning as the one channel you control, and choose its content deliberately. The report's evidence is that the adults who gain most are the ones who take courses that build transferable skill, such as numeracy, languages and project management, rather than courses on the operation of one machine or one protocol.
Produce evidence of skill, not only evidence of attendance. The report's employers and its wage models both respond to measured proficiency. A candidate who can show scored work in a relevant task carries a signal that a certificate of participation does not.
What this means for an employer
An employer that recruits on qualification alone is paying the raw premium: the price of everything a degree carries, including the family behind it. An employer that assesses the skill directly pays the adjusted premium, and widens its pool to adults whose proficiency is real but whose path was not. The survey's finding that nearly one fifth of adults are low performers is also a finding that four fifths are not, and that many of them hold no degree that says so.
The reading
The 2025 Outlook was never a forecast of which skills would win. It is a measurement of where the gap between people opens and how it is carried into pay. The gap opens before the first job, it is paid for through three channels, and a candidate can act on two of them: what they learn after school and how well they can show it. Everything else in a global skills report is context.