Nine ways a chart lies

Almost nobody sets out to mislead with a chart. They pick a default, the default distorts, and the distortion is invisible to the person who made it because they already know the answer. Here are the nine I see most.

1. The truncated bar axis#

Bars encode value as length. Start the axis at 80 instead of 0 and a 3% difference becomes a doubling.

Fix: bars always start at zero. If the differences are genuinely small and matter, use a dot plot or a line chart, where position rather than length carries the value and a truncated axis is legitimate.

2. Dual axes#

Two series, two y-scales, one chart. The reader infers a relationship from where the lines cross — and where they cross is entirely determined by the two ranges you happened to pick. Change one and the "relationship" changes.

Fix: two stacked charts sharing an x-axis, small multiples, or index both series to 100 at a common start date so they share one honest scale. There is no situation that requires a dual axis.

3. The rainbow heatmap#

A rainbow scale has no perceptual ordering. Readers cannot tell you whether green is more than orange without consulting the legend, and the hue changes fastest at arbitrary points, manufacturing boundaries the data does not have.

Fix: sequential data gets one hue, light to dark. Diverging data gets two hues with a neutral grey midpoint.

4. Alphabetical sorting#

A bar chart sorted by category name makes the reader do the ranking. It is the single most common wasted opportunity in business reporting.

Fix: sort by value, descending, unless the categories have a natural order — days of the week, size bands, funnel stages.

5. Bar charts of averages#

A bar showing a group mean hides the spread, the skew, the bimodality, and the sample size. Two groups with identical means and wildly different distributions look identical.

Fix: box plots for a handful of groups, strip or beeswarm plots when you have fewer than a few hundred points per group. If you must show means, show the interval too — and never as a "dynamite plot" with a single error bar on a bar, which is the worst of both worlds.

6. Cumulative charts#

Cumulative totals only ever go up. They always look like growth, including when the underlying rate is collapsing.

Fix: show the rate. If cumulative is genuinely the point — progress toward a target — put the rate beside it.

7. Missing denominators#

A map coloured by "number of customers" is a population map. A chart of raw counts by region is a region-size chart.

Fix: per-capita, per-user, per-opportunity. Whatever the denominator is, it belongs in the chart.

8. Cherry-picked windows#

Start the x-axis one month later and a decline becomes a rise. Usually accidental — the window is "whatever data we have loaded" — and the effect is the same as if it were not.

Fix: show the longest window you have, and annotate the period you want to discuss. If you must truncate, say why on the chart.

9. Aggregation that reverses the finding#

Simpson's paradox, in the wild. A relationship that holds within every segment can reverse when you pool them.

Fix: whenever you compute an aggregate relationship, compute it within your most important grouping too. If they disagree, that disagreement is the finding, and you were about to publish its opposite.

Two more that are not lies but are close#

No title, or a title that names the axes. "Revenue by month" describes what the reader can already see. State the finding: "Revenue grew 22%, almost entirely from marketplace." That title survives being screenshotted into a thread with no context, which is what will happen to it.

Precision theatre. "$4,183,927.44" implies a precision you do not have if the number restates for a week. Round to what could change the decision.

The check#

Before you publish, ask two questions.

What would this chart look like if the finding were the opposite? If the answer is "about the same", the chart is not carrying the argument and something else is.

Which decision does this change? If you cannot answer, you have made a decoration. That is occasionally fine, and it should be a decision rather than an accident.

The tools on this site are built to make most of these hard to do by accident: bars start at zero, dual axes are not available, the heatmap is diverging through grey, and every chart ships with a table view underneath. That is not because designers cannot be trusted — it is because defaults are what most charts actually are.