Showing Uncertainty in Charts: Error Bars, Bands, Fan Charts
A chart that shows only a single line or bar implies a certainty the underlying data usually doesn't have. Error bars, confidence bands, and fan charts each communicate a different kind of uncertainty — measurement error, estimate confidence, and forecast range — and picking the wrong one confuses more than it clarifies.
Quick answer: Use error bars for a single value's measurement uncertainty (typically ±1 or ±2 standard errors) on bar or point charts. Use a shaded confidence band around a line for a statistical estimate's range over a continuous axis, like a regression line's 95% confidence interval. Use a fan chart — bands that widen with distance — for forecasts, since uncertainty about the future compounds the further out you project.
What's the difference between error bars, confidence bands, and fan charts?
Error bars attach a small range to individual discrete points (one bar, one dot); confidence bands are a continuous shaded region around a line showing the plausible range of a statistical estimate; fan charts are confidence bands that widen over time, used specifically for forecasts where uncertainty grows the further ahead you predict.
| Technique | Best for | Typical range shown |
|---|---|---|
| Error bars | Discrete points/bars (survey results, A/B test lift) | ±1 SE, ±1.96 SE (95%), or min/max |
| Confidence band | Regression lines, trend estimates | 95% confidence interval |
| Fan chart | Forecasts, projections | Widening bands (e.g. 50%/80%/95% intervals) |
When should I use error bars instead of a confidence band?
Use error bars when you have a finite set of discrete measurements to compare — bar charts, dot plots, grouped comparisons — where each bar or point needs its own visible range. Use a continuous band when the underlying x-axis is continuous (time, a numeric variable) and a line, not discrete bars, is the right chart type. See scatter plot best practices for adding confidence bands to a regression fit.
Why do forecast charts need a widening fan instead of a constant band?
Forecast uncertainty compounds with distance: a model's confidence about next week is much higher than its confidence about next year, because more unknown factors can intervene the further out the prediction goes. A constant-width band around a forecast implies false precision for the distant periods and false imprecision for the near ones; a widening fan chart represents both correctly.
Near-term forecast (1 month out): narrow band, high confidence
Long-term forecast (12 months out): wide band, low confidence
→ Bands should visibly widen left to right across the forecast horizonWhat if my audience doesn't understand confidence intervals?
Label the band explicitly in plain language ("range of likely outcomes") rather than assuming viewers know what "95% CI" means, and consider a one-line annotation stating the practical takeaway ("Actual sales are unlikely to fall below X"). See confidence intervals explained and overlapping confidence intervals for how to talk about the underlying statistics without misleading a non-technical viewer.
Common mistakes
- Showing error bars without stating what they represent — standard error, standard deviation, and a 95% CI all look identical but mean very different things.
- Interpreting overlapping error bars as automatic proof of "no significant difference," which isn't a valid statistical shortcut.
- Using a constant-width band on a long-range forecast, implying false precision far into the future.
- Cluttering a chart with error bars on every single point when a summary band would communicate the same uncertainty more clearly.
Pro Tip: When in doubt about which interval to show, default to a 95% confidence interval and say so explicitly in the legend or subtitle. Consistency across your organization's charts (always 95%, always labeled) matters more than which specific interval you pick, because it lets viewers build a stable mental model of what the shading means.
← Back to Visualization Tips