The Case for the Treasury

A student’s readiness for work is first tested in a job interview, years after school. Their readiness for society shows up decades later, in how long they live, or whether they end up in the justice system. By the time either number arrives, it is far too late to do anything about it.

Here is what late costs. In May 2026 Alan Milburn’s review for the UK government put the cost of the country’s almost one million young people who are not in education, work or training at £125 billion, which it called “more than we spend on education each year”. Europe counts the same way. The EU’s own research agency put the cost of its young people outside work and study at €153 billion for 2011.

That is the bill for getting it wrong. To decide whether they are getting it right, governments use a sum. The OECD adds up the extra tax a person will pay over a working life for each level of education they complete, and subtracts what that education cost the state. On average across OECD countries, a man who finishes upper secondary school returns about $46,800 to the state, and a graduate about $127,000. For women the figures are far lower, which says less about their education than about what the sum is made of: tax on earnings, and nothing else.

The economists who build these figures know how narrow that is. The World Bank’s review of the research says an ideal calculation would count benefits that never turn into money, and that it leaves them out because there is too little evidence on them. Governments measure financial return because it is the one thing somebody collected.

There are good reasons for that. Tax records cost nothing extra to collect, they are hard to dispute, and they let a pound spent on schools be compared with a pound spent on hospitals.

That would matter less if the figures stayed in the OECD’s tables. In July 2025 the United States wrote a test of what graduates earn into law. Under the One Big Beautiful Bill Act, every degree course in the country will be judged by what its graduates earn four years after finishing, against working adults in the same state who finished at high school, and a course that fails in two years out of three loses access to federal student loans. The first courses can be cut off from 2028. Most will clear the bar. Among those projected to fail are two-year degrees that train early childhood teachers and master’s degrees in mental and social health services. Nobody in Washington decided that teaching small children or supporting people in a mental health crisis is worthless. They decided to count earnings, and earnings gave that answer.

So the state has a single measure of whether education worked, it has begun to act on it, and it arrives late. A graduate’s tax record says nothing useful until they have been working for several years, and a school leaver who never finds work appears only in the £125 billion. The Milburn review saw where the gap opens. “A school is recognised, regulated and rewarded for examination results,” it says, “not on where the student ends up.” The law asks schools to develop the whole person and they check exam results, because exam results are what they have.

Earlier signs exist, and schools already hold them. In Chicago, a ninth-grader’s attendance and course grades predicted whether they would finish school better than the test scores they arrived with. In North Carolina, a teacher’s effect on attendance, behaviour and grades predicted their students’ later completion better than the teacher’s effect on test scores. Almost none of this is reported upward, because nobody upward has asked for it.

Acting on those signs is harder than spotting them. In the one large American trial, schools using an early warning system cut chronic absence and course failure a little after a year, and a Norwegian trial found nothing.

Nobody can promise, then, that better data would pay for itself, and it should not have to. A fairer system that never showed up as a saving on a Treasury spreadsheet would still be worth having. But a Treasury that spends £122 billion a year on education, and learns whether it worked from tax returns a decade later, is in no position to say what pays for itself.

Better data in schools would tell the state whether any of this is working while the student is still in the room, which is the one time anyone can still do something about it, for the taxpayer and the student both.