Source
Adapted from my own handwritten and hand-typed notes from Year 1 to Year 3 of my EEE degree
Tufte: there are two goals when presenting data, convey your story and establish credibility. Above all else, show the data.
Effectiveness (Convey the Story)
Mackinlay: a visualisation is more effective than another if its information is more readily perceived. In practice:
- use encodings people decode better, meaning more accurately and faster
- length is easier to compare than area, so bars beat circles or bubbles
- pie charts with many similar slices are hard to read, a bar chart is clearer
Expressiveness (Establish Credibility)
Mackinlay: a set of facts is expressible in a visual if it shows all the facts and only the facts in the data. In practice:
- all the facts: a bar chart axis must start at 0, starting at 50% exaggerates the differences
- only the facts: do not use bars for values that are not quantities from a common baseline, e.g. shop locations by region are better as a dot plot
- tell the truth, and do not lie by omission (Heer)
Data Ink Ratio
Data ink is the ink that shows actual data. Non data ink is gridlines, borders, heavy backgrounds and decoration.
The ratio should be as high as possible: remove clutter so the data stands out.
Choosing a Chart
Start from the data type (numerical, categorical, mixed, map, network, time series) and from the relation to show. The FT visual vocabulary groups charts by purpose:
| Purpose | Typical charts |
|---|---|
| Deviation | diverging bar |
| Correlation | scatter plot |
| Ranking | ordered bar |
| Distribution | histogram, box plot, violin |
| Change over time | line, column |
| Magnitude | column, bar |
| Part to whole | stacked bar (pie only with few parts) |
| Spatial | choropleth map |
| Flow | Sankey diagram |
Good references: FT Visual Vocabulary and data-to-viz.com. The distribution plots used in this course are in Univariate Visualisation.