How to choose the right chart for your data

The chart type is not decoration — it is the argument. Pick the wrong one and even great data falls flat. Here's the framework I use.

Most bad charts aren't bad because of colour or fonts. They're bad because the chart type doesn't match the question the data answers. Start with the question, not the chart.

1. Name the question first

Before touching a charting tool, finish this sentence: "This chart shows that ___." Your answer decides the chart family:

  • Ranking over time — who's rising and falling? → bump chart or bar chart race.
  • Trend & divergence — how have many series pulled apart? → line chart, or a divergence fan with the spread shaded.
  • Flow between things — where does volume go? → Sankey for origin→destination, chord for two-way relationships.
  • Composition of one entity — what's it made of? → a radial HUD, not a pie chart.
  • Geographic concentration — where is it happening? → a hexbin map.

2. Respect the number of things

A chart that works for 5 categories breaks at 50. With many entities, don't try to show them all equally — highlight the two or three that carry the story and let the rest recede to a muted background. The reader's eye needs somewhere to land.

3. Make the encoding obvious

If a viewer has to hunt to understand why some lines are coloured and others grey, you've failed. State the rule in a caption ("coloured = the three fastest risers + the biggest faller"). Use full labels, never cryptic codes. Add a legend.

4. One number, computed — never typed

Every headline figure — a total, a percentage, a "leader" — should be calculated from the underlying data, not hand-written. Typed numbers drift out of sync and quietly become wrong. A wrong number is a failed chart, however beautiful.

5. Avoid the usual traps

  • Pie charts for anything beyond 2–3 slices — people can't compare angles.
  • Dual axes that imply a correlation you haven't proven.
  • Rainbow palettes that encode nothing. Colour should mean something.
  • All-caps titles and decorative 3D — they read as generic, not premium.

The shortcut

You can skip the guesswork entirely: drop your CSV into the Nomogram Lab agent and it proposes three chart types matched to your data — previewed with your real numbers — then builds the one you pick to a production standard. See the work for what "production standard" looks like, or read about the services.

Have a dataset in mind?

Get three chart proposals on your data, free.