← Back to editorials
✍️

When zero is the lie

When zero is the lie

Last week I went through a small gallery of charts that cut off the bottom of the axis to turn a bump into a cliff. I also confessed what triggered it: the PISA article in De Standaard of 8 September, with its "dramatic" headline and its line charts running from 450 to 550. I was, frankly, ready to write a piece about how our newspapers exaggerate to sell clicks.

So I did what I always tell people to do. I checked.

What did they actually do?

The article has five charts. Three of them show percentages - the share of top and low performers in reading, maths and science, and the share of students who don't reach the basic reading level, split by home language. All three start at zero. No tricks.

The two charts I had noticed are the ones showing the average PISA score for Flanders versus Europe, from 2003 to 2025. Those run from 450 to 550. And here I had to stop and think, because the question is not "does the axis start at zero". The question is "does zero mean anything here".

The scale nobody has ever scored zero on

A PISA score is not a count of anything. It is a scale that the OECD calibrated so that the average across OECD countries at the first measurement was 500, with a standard deviation of 100. Nobody scores zero. Nobody scores 1000. In practice the whole thing lives between roughly 350 and 600. The OECD's own rule of thumb, which De Standaard quotes, is that about 20 points is one year of schooling.

Now look at the numbers, which I took from the Vlaams Indicatorenboek and the UGent PISA 2025 report. Reading in Flanders went from 530 in 2003 to 467 in 2025. Maths went from 553 to 489. That is 63 and 64 points. By the OECD's rule of thumb, roughly three school years (some Flemish academics use a more careful conversion and arrive at "at least a year and a half", which is still a lot). Two thirds of a standard deviation. Our 15-year-olds went from well above the original 500 benchmark to well below it, and only look "above average" today because most other countries slid too.

That is what the picture at the top of this article shows. On the left, the data as De Standaard drew it. On the right, the same data on the "honest" axis from zero - the one I was going to demand. And on that honest axis, the drop is a barely visible tilt. Three lost years of schooling, flattened into a line that says "nothing to see here". That chart would have been the lie.

So what is the actual rule?

This turned out to be a proper rabbit hole, and the people who think about charts for a living are remarkably consistent about it.

Bars must start at zero. Always. A bar encodes a value in its length, so a bar that starts at 28 is simply drawing the wrong length. Datawrapper, the tool half of the world's newsrooms use, refuses to let you do it. The worst offenders from last week - NOS, Fox, VRT, the N-VA chart - were all bar charts, or had no axis at all.

Lines and dots don't have to. A line encodes change in its slope, not in its distance from the floor. Statistician Andrew Gelman puts it nicely: if zero is in the neighbourhood, invite it in - and if it isn't, don't drag it in just to look virtuous. Kaiser Fung of Junk Charts says the same, and his example is SAT scores, which is about as close to PISA as you can get: nobody scores zero, so zero tells you nothing.

And say what you did. A line chart with a cut axis is fine when the axis is visible and labelled, and the reader can see the reference point. De Standaard drew a dashed line at 500 and labelled it as the OECD average at the first measurement. That is exactly what you are supposed to do.

Sanne Willems wrote a lovely Dutch explainer for the Dutch statistical society that arrives at the same place: moet je altijd bij het begin beginnen? No. You have to begin at a logical point, and you have to be honest about which one.

What I got wrong, and what I still think

I got the reflex wrong. I saw a cut axis, matched it to last week's gallery of horrors, and nearly wrote a piece accusing a newspaper of manipulation for doing its job properly. That's a useful lesson in itself. Scepticism about charts is a good habit, but scepticism is not the same as knowing the rule. The rule is about bars versus lines, and about whether zero means something - not about zero as a magic number.

What I still think is that last week's offenders deserve every bit of the criticism. The NOS care-cost bars and the Fox News enrolment bars were counts, drawn as bars, with the bottom sawn off. There is no defence for that. The De Morgen family chart was a line, so by the letter of the rule it gets more slack - but a headline about families "fleeing" the city, over a line whose axis was picked to make a few hundred families look like an exodus, is the same sin in a different chart type. The rule is a tool. The test is still whether the picture matches the numbers.

And there is a third group that worries me more than either of those: readers who, having learned that "the axis should start at zero", now dismiss any chart that doesn't. That is how you get a public that shrugs at three lost years of schooling because the line "didn't look that steep". Positron is not about distrusting the news. It is about reading it well - and that cuts both ways. Sometimes the honest chart is the alarming one.

For what it's worth: the drop is real, it's big, and the fact that the chart showing it was drawn correctly is the least of our problems.

Hope this was a useful reflection - as always, would love to hear what you think.

Cheers

Rik

Share