Annotating Outliers in Charts Without Distorting the Scale

A single outlier can force a chart's scale to stretch so far that every other data point flattens into an unreadable line. The fix is almost never to hide the outlier — it's to call it out with a text annotation and, if it truly breaks the scale, use a broken axis or a separate inset rather than silently rescaling everything around it.

Should I remove an outlier from a chart or keep it and annotate it?

Keep it, unless it's a known data error (a sensor glitch, a duplicate record) rather than a real event — removing a genuine outlier without disclosure hides real information and can mislead viewers about the actual range of outcomes. If you do exclude a data error, say so explicitly in a footnote rather than silently dropping the point.

How do I annotate an outlier without cluttering the chart?

Use a short callout label pointing directly at the point, stating what happened in a few words, not a paragraph — "Server outage, 2hr" tells the viewer everything they need without a legend lookup. Reserve full explanations for a footnote or the surrounding text, not the chart itself.

Good annotation: "Mar 14: Payment outage (2hr)" — short, on the point
Bad annotation:  A dense paragraph in a text box covering a third of the chart

What if one outlier is compressing the rest of the data unreadably?

Use a clearly marked broken axis (a visible jagged gap in the scale) or add a second, zoomed-in panel showing the non-outlier data at readable resolution, and reference the full-scale chart alongside it. Never quietly clip the axis to hide the outlier — a broken axis must be visually obvious, or it becomes the same misleading truncation covered in axis truncation ethics.

Situation	Recommended approach
Real event, moderate scale impact	Keep it, add a text callout
Real event, extreme scale impact	Broken axis (clearly marked) or a second zoomed panel
Confirmed data error	Exclude, with a visible footnote explaining why
Unconfirmed anomaly	Keep it, flag it, investigate before deciding

How do I tell if a point is really an outlier and not just natural variation?

Use a statistical rule rather than eyeballing it — a common threshold is more than 1.5x the interquartile range beyond the box in a box plot, or more than 2-3 standard deviations from the mean for roughly normal data. See outlier detection methods and z-scores explained for the underlying math before deciding a point deserves a special callout.

Common mistakes

• Silently deleting outliers from the underlying data before charting, with no disclosure.
• Annotating every unusual-looking point, which trains viewers to ignore the annotations entirely.
• Using a broken axis without a visible break marker, which reads as ordinary truncation.
• Explaining an outlier with a guess ("probably a data error") instead of confirming the cause before publishing.