We hold fifteen years of European employment figures, broken down by occupation group. Looking for something to write about, we did the obvious thing: compare 2011 with 2025 and sort by the biggest change.
Top of the list of declines: Other clerical support workers, down 40%. Two and a half million people in 2011, one and a half million now. A ready-made headline about the office job vanishing from Europe.
It is not true. And the figures are correct.
The shape, not the endpoints
Here is the same group, year by year, in thousands of people:
| year | employed (thousands) | change |
|---|---|---|
| 2011 | 2,615 | — |
| 2012 | 2,433 | −7% |
| 2013 | 1,964 | −19% |
| 2014 | 1,612 | −18% |
| 2015–2024 | ≈1,600 | flat |
| 2025 | 1,574 | −1% |
The entire fall happens in three years and then stops dead for a decade. Jobs do not behave like that. Automation is slow, retirement is slow, industries decline over decades. A category that loses 40% in three years and then holds perfectly still for eleven has not lost anyone: it has been redrawn.
The proof is in the neighbours
If people are being moved between categories rather than out of work, the groups next door should move the opposite way at the same moment. They do.
| group | year | change |
|---|---|---|
| Numerical and material recording clerks | 2012 | +24% |
| General and keyboard clerks | 2012 | −15% |
| General and keyboard clerks | 2020 | +18% |
| Numerical and material recording clerks | 2020 | −13% |
The same two clerical groups swap roughly a million people between them in 2012, and swap them back in 2020. Nobody changed job. A boundary moved, and the people standing near it were counted on the other side.
The same years show other jumps of the same kind — one group of agricultural workers rises 56% in a single year, which is not something agriculture does. Breaks cluster in particular years across unrelated categories, which is the signature of a survey being revised rather than of an economy changing.
What a real trend looks like
The contrast is the useful part. Over the same fifteen years, in the same dataset:
| group | 2011 | 2025 | change |
|---|---|---|---|
| ICT professionals | 2,377 | 5,423 | +128% |
| Business and administration professionals | 4,913 | 10,104 | +106% |
| Science and engineering professionals | 5,252 | 7,905 | +51% |
| Skilled agricultural workers | 6,964 | 4,523 | −35% |
Europe more than doubled its ICT professionals and lost two and a half million farmers. Both of those are real, and you can tell because the year-by-year series walk there: 2,377, 2,429, 2,511, 2,595, 2,716, and on up, one step at a time, with no year doing the work of a decade. The decline in agriculture is just as steady in the other direction, without a single reversal.
That is what a trend looks like from the inside. A boundary change looks like a cliff followed by a plateau.
Why the wrong version is the usual one
Comparing a start year with an end year is how almost everyone reads these figures, including us until an hour before writing this. It is fast, it produces a single number, and that number is arithmetically correct.
It is also blind to the one thing that matters: whether the change happened. Two points cannot tell a slope from a step. You need the line.
So the check is cheap and worth making every time. Before believing any figure of the form X% over N years, ask for the years in between. If the change arrived all at once and then stopped, you are not looking at the labour market. You are looking at the people who measure it, changing their minds about where to put someone.
What we did with it
We did not publish the clerical headline. We did keep the two real trends, and they now sit on the occupation pages they belong to, labelled with the group they describe rather than the single job the page is about — which is a different way of being wrong, and one we have written about separately.