Choosing a chart
Match the visual to the question rather than to the data.
Pick the chart that answers the question you are asking. The data usually fits several of them; the question rarely does.
| The question | The visual |
|---|---|
| How much, right now? | Card, KPI |
| How do categories compare? | Bar, Column |
| How has this changed over time? | Line, Area |
| What is this made of? | Pie, Donut, Treemap, Stacked bar/column |
| Where do people drop out of a process? | Funnel |
| Are these two numbers related? | Scatter |
| How close are we to a target? | Gauge, Linear gauge, Bullet |
| What are the actual records? | Table, Matrix |
| Let readers narrow the view | Slicer |
A few rules that hold up
- Bar over pie for comparison. People compare lengths accurately and angles poorly. Reach for a pie when the point is "these are parts of one whole", not "which is biggest".
- Line for time, bar for categories. A line implies continuity between points. Between "France" and "Germany" there is none.
- Start bar and column axes at zero. A truncated axis exaggerates differences, and a reader has no way to tell it happened.
- 100% stacked answers a different question. It shows composition and hides magnitude — two bars can look identical while one represents ten times the volume.
Fewer, larger visuals
A page of six well-chosen charts communicates more than twenty small ones. If a visual is not answering a question somebody actually asks, it is costing attention rather than earning it.