Guidelines
Suggest edit

Do not tune single-hue sequential palettes to chase retrieval performance

For retrieve tasks, avoid switching among single-hue sequential color scales on quantitative color encodings to prevent over-interpreting negligible performance differences and address unnecessary palette tuning for viewers comparing values against a legend.

  • purpose:refine
  • basis:empirical
  • task:retrieve
  • data:quantitative
  • quality:fidelity
  • lever:encoding
  • reading-mode:lookup
  • polish:palette

advice

Single-hue palette tuning

Do not swap one single-hue sequential ramp for another to chase retrieval speed or accuracy. For example, changing between greys, blues, greens, and oranges did not produce a significant performance difference in the tested quantitative retrieval task.

reason

Why this tuning does not pay off

Within the tested single-hue family, the palette hue changed the look of the scale more than it changed the retrieval result. Readers performed similarly across the single-hue options, so hue swapping inside that family was not an effective speed-or-accuracy intervention.

Mechanism: When the only edit is changing the hue of a single-hue sequential ramp, the retrieval task still relies on the same basic ordered color progression, so performance stays within the same range.

Evidence: The review records the single-hue family as a weak ordering for accuracy and a tie for time, with no significant pairs; the original study likewise found no significant differences among greys, blues, greens, and oranges for either error or response time in the retrieval task (Zeng & Battle, 2023; Liu & Heer, 2018).

context

Use when this is the decision

  • User Goal: Improve retrieval performance on an existing single-hue quantitative color scale.
  • Task: Compare relative closeness of encoded values using a legend.
  • Data: Scalar quantitative values encoded with a single-hue sequential palette.
  • Chart Setting: The planned revision is only a swap among single-hue ramps.
  • Audience: Readers decoding magnitude from color.
  • Success Criterion: A measurable gain in speed or accuracy.

exceptions

When not to use this rule

Break it when: You are deciding between a single-hue palette and a different palette family. Why: This rule only covers within-family swaps among the tested single-hue sequential ramps.

costs

What this costs

Sacrifice: You give up palette iteration as a primary performance fix within the single-hue family. Risk: Keeping the same palette family can leave a real retrieval problem unresolved. Mitigation: If performance is still weak, compare the current single-hue ramp against a different palette family instead of another single-hue hue.

mistakes

Common failure around this change

Mistake: Recoloring a single-hue ramp from one hue to another after readers miss value comparisons. Why it fails: The tested single-hue ramps were statistically indistinguishable on the retrieval task.

check

How to test the tuning decision

Failure Sign: The chart still produces retrieval errors after a hue-only swap inside the single-hue family. Quick Check: A/B test the current single-hue ramp against another single-hue ramp on the same chart and retrieval question. Stronger Test: Record time and error for both versions; if the difference stays negligible, stop tuning within the family.

fix

What to change

  • Stop rotating among single-hue hues as the main retrieval-performance intervention.
  • Keep one single-hue ramp and compare it against a different palette family if retrieval remains weak.
  • Evaluate any further palette change with the same legend-based retrieval question.

References

Liu, Y., & Heer, J. (2018). Somewhere Over the Rainbow: An Empirical Assessment of Quantitative Colormaps. Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems, 1–12. https://doi.org/10.1145/3173574.3174172
Zeng, Z., & Battle, L. (2023). A Review and Collation of Graphical Perception Knowledge for Visualization Recommendation. Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems, 1–16. https://doi.org/10.1145/3544548.3581349