For fifteen years the country argued about why British workers had stopped getting better at their jobs.
Books were written. Select committees sat. Chancellors stood up with graphs. Outrage was had by all!
Even, heaven forbid, someone, somewhere, made a podcast. The “productivity puzzle” got its own name, which is what happens to a problem when nobody can solve it and everyone has to keep talking about it. Then, on 17 September, the Office for National Statistics published a correction to how it counts hours worked, and about half of the puzzle went away.
They didn’t find any mythical missing output, in fact, the hours were wrong. When the survey couldn’t get hold of somebody, the old method wrote down that they’d worked a normal week. Someone on a beach, or halfway up a ladder cleaning the gutters was counted as working hard for the UK economy.
How you lose fifteen years
The Labour Force Survey is a rolling panel. Every quarter or so it goes back to the same households and asks, among other things, how many hours you worked last week. People couldn’t remember, so they gave their contracted hours. If someone else in the house answered on their behalf, that got worse. And when a person couldn’t be reached at all, the method carried forward the last thing they’d said, which for most people most of the time was a normal week.
The ONS’s own example is a person who goes on holiday and therefore doesn’t answer the phone. The record should show fewer hours. It shows the usual ones instead.
The survey comes back round every three months, so a person missed in August is carrying roughly May’s answer through the summer. Response rates fell through the 2010s, and so the “just put what they did in May” problem got worse.
The numbers that moved
Hours worked per week, roughly 30.75 down to 28.75, once leave is counted properly. Average weekly hours over 2007 to 2024 fell by 5.2% on the new method, where the old one had them rising by 1.3%. Productivity growth for 2009 to 2019 revised from 0.7% a year to 1.3%. The UK goes from second-worst in the G7 to second-best, behind the United States.
That’s the Resolution Foundation’s arithmetic. The ONS’s own line is more careful: the puzzle remains, but hours explain about half of it.
Silence is not a zero
The ONS didn’t have a data problem so much as a listening problem. Dave, the guy on holiday. was giving the right answer by not picking up the phone, but the method had no way to hear it, so it filled the silence with the last thing it heard.
Every survey you’ve ever run does the same thing.
Consultation that hardly anyone answered becoming “residents told us”. The staff survey where the people who’d already decided to leave didn’t fill it in, so satisfaction went up. The dashboard field that keeps last month’s value when nobody updates it, so every stale record looks current.
The people who don’t answer are not a random sample of the people who do. They’re missing for a reason, and the reason is often the exact thing you were trying to measure.
No fix, just choosing better
The new method takes employer-reported hours and subtracts leave. It assumes everyone takes all the holiday they’re entitled to.
That’s not a measurement either. It’s a guess. A better guess than “nobody ever goes on holiday”, and one the ONS has made with its eyes open. You can’t count what people did in the week you didn’t ask them. You can only choose what you assume, and know that you chose it.
Most organisations never get that far. The default fills the gap, the number goes on the slide, and the slide goes to the board.
And let’s be honest, the cynical reading is available. The ONS has had a rough few years on this survey, and here’s a revision from the same body that makes the national picture look better. The TaxPayers’ Alliance has already said it won’t help anyone. I don’t buy it.
Who didn’t answer?
Britain is still well behind France and Germany on output per hour, and the revision suggests the gap has older roots than 2008. Half a puzzle is still a puzzle.
But the half that went away went away because somebody finally asked what the silence meant.
So, one question, and take it to your next survey or your next dashboard.
Who didn’t answer, and what would it mean if they were all missing for the same reason?
