Young people and work explorer

The model follows one year-group of 16-year-olds, three months at a time, to age 41. It tracks shares of the group, not individuals: how many are in education, in work, or out of work (looking, caring or ill), how long they have been out, whether they get a health benefit, and what time out has cost them in pay.

It starts from today's official figures: about 13 in 100 16 to 24-year-olds out of work and learning (981,000 people), who they are, how many get health benefits, and how long they stay on them. Some of the flows behind those figures are not published, so the model solves for them until it matches.

Each policy is set so that, for the group it was tested on, it gives the effect that test found. Each carries a grade: A is official data or a randomised trial, B a robust survey or a study comparing like with like, C modelling or a secondary source, D an assumption. Click a grade to see the source. Where the evidence is thin, the assumption is a control under "Assumptions the evidence does not settle".

The range comes from running each choice three times: the weakest reading of the evidence (with the most displacement), the central one, and the strongest. It shows how far readings of the evidence differ. It is not a statistical interval.

What I could not make fit

Two official sources disagree. Government analysis says young people out of work and learning for under a year usually leave within the year (about 65 in 100), but those out for longer rarely do (about 25 in 100). Applied to today's numbers, that means about a third of a year-group is out for a year or more before 25. A study of pupils who were in Year 11 around 2012 found about 12 in 100. They cover different generations and the rate has risen, but I cannot show that explains all of the gap.

I kept today's numbers because they are the latest and the ones the policies act on. So the page shows how much more likely one group is to be out of work and learning than another, never a group's own rate. Young people on free school meals for three years or more are about three times as likely as others, in the study and in the model. The model's rate for that group is too high to show.

This also makes the number of people on a programme, and so its cost, too high. The gain is too high by about the same share, so they shrink together and the sign of the result holds. The job guarantee's eligibility rule (set below) changes them a great deal.

What it leaves out

It does not know whether someone has ever worked, so it does not reproduce the official figure that six in ten have never worked. It counts full-time education only, and not its cost. After 24, education empties a little faster than the census shows (3 to 4 in 100 still studying at 30), a small gap. It has no regions or separate nations, no effect of more people in work on wages or jobs, no housing costs beyond an average, and nothing on AI and entry-level jobs. It stops following people at 41, so the country's figures leave out older people's benefits and tax. Nothing is shown after 41, because nobody has measured whether time out of work stops costing pay. Whether being out of work causes ill health, or the other way round, is unresolved, so that link is left out. The 2026 benefit changes are too new to have results. A move into education shows no education cost.

Early years help, such as Sure Start, is not a lever. The best evidence, from the Institute for Fiscal Studies, is on childhood outcomes and exam results at 16. Nothing yet measures its effect on being out of work and learning, and any effect would take about 16 years to arrive. The IFS estimates the government eventually recoups about 90% of the cost, mostly as tax on higher pay, which this page does not follow past 41.

Money is in 2026/27 prices, with 2026/27 income tax and National Insurance rules for England. Year-group sizes follow the ONS 2024-based projection to 2060.

Each grade note links to its source.

Numbers in this debate are often a different measure from the one quoted. These are the ones I checked against the Milburn review's own annex and other published data. The page uses the underlying figures.