All you can plot
Which chart should I use?
Start from the question your data answers, not from the chart you like. Here is how to match the two — and the mistakes worth avoiding.
Start with the job
Every chart encodes numbers as something you can see: position, length, angle, area or colour. People judge these with very different accuracy. In Cleveland and McGill’s ranking, position along a common scale is read most accurately, then position on unaligned scales, then length and angle, then area, with colour shading last (Cleveland & McGill 1984). So a good chart puts the comparison that matters into the most accurate encoding.
That is why the first step is to name the job: are you comparing amounts, showing change over time, showing parts of a whole, relating two measures, or showing how values are distributed?
| Your question | Reach for | Why |
|---|---|---|
| How do these amounts compare? | Column chart | Lengths on a common baseline are the easiest comparison. |
| How do many categories rank? | Bar chart | Horizontal bars leave room for long labels. |
| How has this changed over time? | Line chart | Slope shows the rate of change. |
| What is this whole made of? | Doughnut chart | For a few parts with one clear message. |
| How does the mix differ between groups? | 100% stacked chart | Shares line up against common edges. |
| Do these two measures move together? | Scatter plot | Each row becomes a point; patterns appear. |
| How are these values spread? | Histogram | Bins reveal the distribution’s shape. |
| How do several groups’ spreads compare? | Box plot | Medians and quartiles side by side. |
| What got us from this number to that one? | Waterfall chart | Floating bars bridge two totals. |
| Where do people drop out? | Funnel chart | Each stage is a subset of the last. |
| Which few causes matter most? | Pareto chart | Sorted bars plus a cumulative line. |
Change over time
Time almost always runs left to right. For many points, use a line; for a few periods, columns work and make each value easy to read. Stacked areas show a total and its parts over time, but only the bottom layer can be read precisely.
Never put two measures with different units on two y axes. The alignment of the two scales is arbitrary, so any apparent crossing or correlation is an artefact of where the axes were set. Use two charts, or convert both to an index.
Parts of a whole
Pies and doughnuts work for two to six parts and a simple message. With more parts, or parts of similar size, sorted bars are clearer. To compare the make-up of several groups, use a 100% stacked chart rather than a row of pies.
Check that the parts really add up to a whole. Survey questions that allow several answers, or categories that overlap, don’t.
Colour with a purpose
Give each series one colour and keep it, chart after chart. Don’t colour bars by value when the categories have no order — length already shows the value. And don’t use more colours than readers can tell apart: around eight at most, fewer for scatter plots. Chart Buffet’s palettes marked “colour-blind safe” are tested so that neighbouring colours stay distinguishable for the most common forms of colour blindness.
Label directly where you can. A legend makes readers look back and forth; a label at the end of a line or on a bar is read at once.
Keep it honest
Bars and areas must start at zero, because their length or area is the value. Lines need not, since their slope is the message, but say so if you cut the axis.
Remove what doesn’t carry information — heavy gridlines, borders, 3D effects, shadows. Tufte called the ratio of data to everything else the “data-ink ratio”, and argued for raising it (Tufte 1983).
References
- Cleveland, W. S. & McGill, R. (1984). Graphical perception: theory, experimentation, and application to the development of graphical methods. Journal of the American Statistical Association 79(387), 531–554.
- Bertin, J. (1967). Sémiologie graphique. Paris: Gauthier-Villars; La Haye: Mouton.
- Tufte, E. R. (1983). The Visual Display of Quantitative Information. Cheshire, CT: Graphics Press.
- Munzner, T. (2014). Visualization Analysis and Design. Boca Raton, FL: CRC Press.
- Friendly, M. & Denis, D. J. (2001). Milestones in the history of thematic cartography, statistical graphics, and data visualization. Web document.