Small Multiples
"At the heart of quantitative reasoning is a single question: Compared to • what? Small multiple designs, multivariate and data bountiful, answer • directly by visually enforcing comparisons of changes." — Edward Tufte
What I see and like
A wall of nine identical little maps, one per year, the same region shaded darker as something spreads. Because the frame never changes, the change is the only thing that moves. The eye does the statistics automatically.
Same frame, shifting data
A small multiple repeats one design and varies one variable. Identical scales, identical axes, identical everything — except the data.
2019 2020 2021 2022 +----+ +----+ +----+ +----+ | . | | .. | |... | |....| |. | |.. | |.. | |... | +----+ +----+ +----+ +----+ same axes, same size, same scale -> only the data differs
Why it is the best design
Once the reader learns to read one panel, they can read all of them for free. Comparison becomes effortless and trustworthy.
| Property | Effect on the reader |
|---|---|
| Constant frame | Removes the work of re-orienting |
| Shared scale | Makes magnitudes directly comparable |
| Dense layout | Many comparisons in one glance |
| Repetition | Pattern and exception both pop out |
Index, don't reset
The cardinal sin is letting each panel pick its own scale. The moment scales differ, the comparison is a lie.
WRONG (auto-scaled) RIGHT (shared scale)
panel A: 0..10 panel A: 0..100
panel B: 0..1000 panel B: 0..100
"looks the same" but isn't truly comparable
Key takeaways
- Small multiples answer "compared to what?" by repeating one frame.
- Hold scale, axes, and size constant; vary only the data.
- Learn one panel, read them all.
Checklist
- [ ] Every panel shares one identical scale and axis
- [ ] Panels are small, dense, and ordered meaningfully
- [ ] The only thing that changes across panels is the data