Nothing happened to Swedish teenagers in 2021

The figures behind this page come from the European Union Labour Force Survey. In 2021 that survey was reorganised: new regulation, new definitions of who counts as living with their parents, new sampling in several member states. The numbers before and after are both correct. They are just not measurements of the same thing.

Eurostat does not hide this. It attaches a flag — the letter b, for break — to the affected observations. In this dataset 32 of the 33 countries carry that flag for 2021. It is not an edge case affecting a stray country. It is almost the whole table.

The trouble is what a flag does and does not do. It marks the data. It does not stop anyone drawing a line straight through it, and a line through a break looks exactly like a line through a trend.

Here is a sentence you could write from this dataset without inventing anything. Between 2005 and 2025, the gap between the earliest and the latest country in Europe grew from 8.1 years to 10.2 years. Young Europeans, you might conclude, are drifting apart.

Two things are wrong with it. The first is the break. The second is quieter and worse: 2005 covers 24 countries and 2025 covers 31. The countries that joined the table in between are not a random sample — they are mostly the ones at the far ends of the range. Some of that "growing gap" is just more countries being measured.

Measure inside a window where the definition held still, and the direction reverses. Between 2022 and 2025, across the 31 countries that report in both years, the spread narrowed from 12.1 years to 10.2. Same dataset. Opposite conclusion.

The obvious fix — drop the break year, join the two ends — does not work either, because the break year is not a midpoint between the old definition and the new one. It is a transition, with countries switching over at different moments and in different ways.

Portugal is the clearest case. It reads 29.9 in 2020, jumps to 33.3 in 2021, and settles at 30.1 the year after. By 2025 it is back to 28.8. Whatever 33.3 measures, it is not the age at which half of young Portuguese people had left home that year.

The same thing shows up in the spread across all countries. In 2020 the distance between the lowest and the highest country was 15.8 years. In 2021 it was 11.8. Europe did not converge by 4 years in one year; the bottom of the range simply moved, from 17.5 to 21.5, when the definition did.

6 of the 30 countries move by a year or more. For a figure that shifts by a tenth or two in a normal year, that is enormous. And the moves do not point the same way: 4 rise and 2 fall, which is what you would expect from a definitional change and not from anything that happened in Europe.

Every page here that shows change over time asks one function for the answer, and that function checks for a break inside the window first. If it finds one, it returns nothing and the section does not render. There is no option to override it and no footnote variant, because a footnote is what failed.

That is why leaving home carries a trend measured from 2022 and not from 2000, even though the data goes back that far. And it is why some pages carry no trend at all.

Where else this bites

The 2021 reorganisation was not specific to one indicator, and other sources have breaks of their own. Across the datasets behind this site:

• Leaving home — comparable since 2022, with a break in 2021
• Household size — the break falls in 2025, the latest year itself, so there is no window to measure over at all
• Working hours — comparable since 2025, with a break in 2024
• Healthy years — the break falls in 2024, the latest year itself, so there is no window to measure over at all
• First child — comparable since 2024, with a break in 2023

Working life is the exception: no break is flagged, so the whole run is comparable. It is the only page here that can honestly show two decades on one line.

So none of the data before 2021 is usable?

It is usable, just not in the same sentence as the data after. The older run is a perfectly good series on its own terms: you can compare countries within 2015, or follow one country from 2000 to 2020. What you cannot do is subtract a number from one side of the break from a number on the other.

Why does Eurostat not just correct the old figures?

Reconstructing twenty years of survey responses under a definition that did not exist when they were collected would mean estimating, and Eurostat publishes what it measured. Flagging the break is the honest option available to them. The gap is not in their statistics; it is in what readers do with them.

How do I check whether another dataset has a break?

In Eurostat's own browser, flagged observations carry a small b. Through the API the flag arrives in the status block alongside the values. Our pipeline reads that block rather than the published table, which is the only way to be sure the flag has not been dropped somewhere in between — the full account is on the method page.

Does this affect the comparison I get when I enter my own age?

No. That comparison uses one year — the most recent one each country reports — so it never crosses a break. Breaks only matter when you look at change over time, which is exactly where this site is most careful and most other pages are not.