unspurious.

The presentation illusions · An interactive tool

Lie with this chart.

The data never changes. Only the axis does. Drag the baseline off zero and watch an honest, boring difference turn into a damning one.

Quarterly revenue — up 3.3% on last quarter Two real numbers: £4.80M then £4.96M. Both bars are honestly drawn.
…or drag the baseline handle on the chart.
Real change+3.3%
What the eye sees+3.3%
Exaggeration factor1.0×
An honest chart. The bars are in true proportion — dull, and trustworthy.
QUICK SCENARIOS
The trick. Bar length is the oldest visual encoding there is, and the eye reads it as quantity from zero. Start the axis somewhere else and the picture quietly stops matching the numbers — the bars now encode the distance from an arbitrary line you chose. Nothing is mislabelled; every value is correct. That's exactly why it works.

01 · Why bars are different from lines

A bar promises you it starts at zero

There's a genuine debate hiding here, and it's worth being precise about. A bar chart encodes value as length, and length is only meaningful from a baseline — so a bar chart that doesn't start at zero is making a visual promise it doesn't keep. A line chart encodes value as position, and is read for slope and shape; lines can often, defensibly, start somewhere other than zero, which is why stock charts and temperature records do it without scandal.

So the rule isn't "all axes must start at zero." It's narrower and sharper: length-based encodings must. The tool above is rigged with bars precisely because that's the case where truncation crosses from a judgement call into a distortion.

Ask not "does this axis start at zero?" but "is the eye being asked to compare lengths? If so, where is zero — and why isn't it shown?"

02 · The other dial: dual axes

Two series, two secret rulers

Switch the tool to dual axis and a subtler lever appears. When two series are each given their own invisible vertical scale, the designer chooses where they cross, which one looks higher, and whether they appear to move together. A "striking correlation" or a "dramatic crossover" can be manufactured purely by sliding one hidden ruler against the other — the data underneath never moves.

This is why careful publications are wary of dual-axis charts entirely: the relationship you see is partly the data and partly an editorial decision about two scales, and the reader is given no way to tell which is which.

03 · Field notes

Where you've already seen this

Cable news and the truncated y-axis is the canonical pairing — a famous 2012 chart showed a tax rate rising from 35% to 39.6% with an axis running from 34 to 42, turning a 4.6-point change into what looked like a fivefold leap. The numbers were correct. The baseline did the lying.

Product launches love the "maximum drama" preset: benchmark bars that begin at 90%, so a 2% lead reads as a landslide. Election-night graphics and financial results reach for the same move whenever a modest change needs to look decisive.

The defence is a one-second habit: before reacting to the shape of a bar chart, find where the axis starts. If it isn't zero — or isn't shown at all — mentally redraw it. The honest version is usually waiting underneath, and it's usually far less exciting than the chart wanted you to feel.

Keep going

More from the compendium