I was in a meeting once where someone presented a performance chart showing a line climbing steeply from left to right across 18 months. It looked like dramatic growth — the kind of upward slope that gets applause in boardrooms. When I asked what the Y-axis started at, I noticed the baseline was 94%. The chart represented a change from 94.2% to 96.8% — genuine improvement, but modest. Visually, it looked like something multiplying. The data was accurate. The chart was designed to produce an impression the data didn't support. That moment changed how I read every graph that followed. A chart is a set of design decisions, and those decisions shape perception at least as much as the underlying numbers do.
The Truncated Y-Axis
The most common misleading graph technique is starting the y-axis at a value other than zero. A chart showing company revenue "growing" from £98m to £103m looks dramatic if the y-axis runs from £90m to £110m — a 5% increase fills most of the chart height. The same data on a y-axis starting at zero would show an almost-flat line. Both are technically accurate. One is designed to exaggerate.
The rule of thumb: bar charts should almost always start at zero. Line charts are more contextual — when tracking a variable that varies over a narrow range around a large baseline (stock price, body temperature), a truncated axis is sometimes appropriate and clearly labelled. The question to ask is whether the visual impression matches the actual magnitude of the change.
Area and Bubble Charts That Mislead
When size is encoded as area (circles, squares, or 3D shapes), the visual impression is often wrong. Our brains tend to judge the diameter of a circle, not its area. A circle with twice the diameter has four times the area (πr²), but visually looks roughly twice as large. Charts that encode data in the area of circles but present them as if the visual impression were linear systematically misrepresent ratios.
The same problem applies to 3D bar charts: a bar that's twice as tall in a 3D perspective looks less than twice as tall due to depth distortion. 3D charts almost never improve the accuracy of data communication and frequently distort it.
Cherry-Picked Time Ranges
Selecting a time range that starts or ends at a convenient point is a classic way to show a trend that might not hold over a longer period. A chart of stock performance from a low point to a high point shows impressive gains; the same chart extended by a few months in either direction might show a different story. Temperature charts in climate debates are particularly subject to this — starting at an unusually hot year creates the impression of cooling; starting at an unusually cool year creates the impression of dramatic warming.
Misleading Correlations in Scatterplots
Scatterplots that show correlations are subject to the same correlation-vs-causation issues that affect any statistical claim. A regression line through a scatterplot looks like evidence of a relationship, but the strength of that evidence depends on how much scatter exists around the line, how large the sample is, and whether the relationship has a plausible causal mechanism. A tight cluster of points with a clear trend is more informative than a wide scatter with a line drawn through it.
Our probability calculator and the broader toolkit of probability and statistics help contextualise what visual correlations actually imply in terms of likelihood — separating pattern from noise.
Inconsistent Scales
Comparing two datasets on the same chart using different scales — or using dual y-axes without making this clear — allows two lines to be made to appear synchronised when they may not be. A chart showing sales growth (left y-axis: £0 to £500k) alongside customer satisfaction (right y-axis: 0 to 100) can be positioned so the lines track each other closely regardless of whether they're actually correlated, simply by adjusting the scale of one axis. Two lines that cross dramatically can be made to not cross at all by rescaling.
Reading Charts Critically
For any chart, check: where does the y-axis start? Are there multiple y-axes? What time range is shown, and why that range? Are areas or volumes encoding data that should be encoded as lengths? Is the correlation shown likely to be causal, or is a third variable plausible? What's the sample size or the error bar? None of these questions takes more than a few seconds to ask, and they consistently reveal whether a chart is communicating data honestly or engineering a particular impression.
Why Pie Charts Are Worse at Showing Differences Than They Look
Pie charts are everywhere, but they are genuinely one of the least effective ways to communicate quantitative differences. The reason is physiological: the human visual system is much better at judging length than angle. Bar charts encode values as lengths, which the eye compares quickly and accurately. Pie charts encode values as angles and areas, which the eye compares poorly — particularly when the slices are similar in size. Studies of data visualisation consistently show that people make more errors and take longer to answer questions about the same data when it's shown as a pie chart rather than a bar chart.
The problem worsens with more than four or five categories, when any slice smaller than about 5% becomes visually indistinguishable from adjacent slices without a label. And yet pie charts remain ubiquitous in business presentations and journalism, partly because they look more interesting than bar charts and partly because convention has made them feel authoritative. The practical alternative for most pie chart use cases is a horizontal bar chart sorted by value — it communicates the same information faster, more accurately, and scales to any number of categories without becoming illegible.