Where the axis starts decides what you see.

When you draw a bar chart, you are making a silent agreement with the reader: the length of each bar is proportional to the quantity it represents. Twice as long means twice as much. That is the entire reason bar charts work — the eye compares lengths effortlessly and trusts them. Truncating the axis breaks the promise without announcing it. The bars still look like honest lengths, but they now measure distance from an arbitrary floor rather than from zero, so their proportions no longer match the numbers.

The result is one of the most common and effective ways to mislead with a completely accurate dataset. No figure is wrong. The axis is simply labelled — truthfully — to begin somewhere other than zero, and most readers never look. A two-percent difference becomes a towering one; a flat trend becomes a dramatic ascent; a rival's identical performance becomes a visible deficit. The chart tells the truth and lies at the same time.

Because a bar encodes its value as length, the only honest baseline for a bar chart is zero. Anywhere else, the ratio between two bars’ heights stops matching the ratio between their values, and the visual comparison the chart exists to make becomes false. This isn’t an aesthetic preference; it is a property of the encoding.

The tell is always in the same place, and it takes one second to check: look at where the vertical axis begins. If it starts at a number other than zero on a bar chart — or worse, isn’t labelled at all — the differences you’re being shown are exaggerated by some factor the chart is hoping you won’t compute.

Here the rule earns its nuance, because “always start at zero” is itself an over-simplification that the well-meaning repeat too confidently. It is right for bars and wrong for many line charts. A line doesn’t encode value as length; it encodes change as position and slope. Forcing a global temperature series, a stock price, or a blood-pressure reading to include zero can flatten a genuinely important movement into a meaningless straight line — hiding real signal rather than revealing it.

So the sophisticated rule is not “never truncate.” It is: match the baseline to the encoding, and label it plainly. Bars need zero. Lines may start elsewhere when there’s a real reason — but the moment a truncated line is drawn to make a 0.3-degree wiggle look like a catastrophe, or a steady trend look flat, the same deception is back, just wearing a different chart type.

The truncated baseline is the most common axis trick, but it keeps relatives. Each takes a legitimate charting choice and points it at the reader.

What unites the whole family is that none of them touches the data. The numbers stay pristine; the frame around them does all the lying. That is exactly why these tricks survive fact-checking — there is no false number to catch — and why the only defence is to read the scaffolding of a chart as carefully as its contents.

The one-second habit. Before you absorb the shape of any chart, glance at the vertical axis and find its lowest number. On a bar chart, if that number isn’t zero, mentally deflate the drama you just felt — the differences are smaller than they look, often by a lot. If the axis isn’t labelled at all, trust the picture even less; an unlabelled axis is a chart asking you not to check.

Ask what the marks encode. Length-based pictures — bars, columns, area, bubbles — demand a zero baseline, because their whole message is “how big.” Position-based pictures — lines, dots, slopes — are about “how it changed,” and a labelled non-zero start can be the honest choice. The distortion isn’t the truncation itself; it’s a truncation that fights the encoding.

So the pocket question is almost embarrassingly simple, and it is the one that disarms the whole genre: where does this axis start? Ask it of the next chart in your feed, your slide deck, your morning paper. Then, if you want to feel exactly how much an axis can be made to lie, go and bend one yourself in Lie With This Chart. The rest of the compendium is about numbers that mislead; this is the one where the numbers are innocent and the picture is the culprit.