Colour as a signal

Claim: colour links what belongs together and is never the only cue; data colour maps are perceptually ordered.

What it asks

  • Give one colour to one kind of thing, and keep it across the deck.
  • Never let colour be the only cue. The slide stays readable in greyscale (W3C 2023c; Rector 2019; Microsoft 2026d).
  • Use few colours: at most 4 colours per chart, one per category, in a fixed order (ScotGov 2024, convention).
  • For ordered data, choose a map by lightness, such as viridis, not a rainbow-like map (Crameri 2020; Matplotlib 2026).
  • Keep photographs in their real colours (Bramao 2011).
  • Show position and length before colour: draw relations.

Why

Ozcelik 2009 (n = 52, abstract) found that colour-coding text and picture parts raised retention and transfer. Eye tracking showed that readers located the corresponding parts faster.

Jet-like maps can distort over 7% of the displayed data variation. An estimated 8% of men and 0.5% of women have a colour-vision deficiency (Crameri 2020, full text). Viridis gave consistently low error in 56 subjects (Liu 2018, full text).

Limits

  • A hue background that signalled lesson sections helped a structure test, d = 0.41, and not a fact test, d = -0.17 (Clariana 2009). A background colour as retrieval cue did not help recall (Meusel 2024).
  • A diverging blue-orange scheme erred when a comparison crossed its centre (Liu 2018). Rainbow maps can help some inference tasks, so the ban is a default, not a law (Reda 2021, secondary).
  • Pleasant colours help a little in lessons. Retention d+ = 0.387 across 33 samples (Brom 2018), but only 5 of the 33 changed colour alone, and no moderator held after correction.
  • Colour harmony is preference only: no study of harmony and learning was found (Schloss 2011). Hue meaning, such as red as alarm, is untested for a live audience.
  • Colour aids object recognition at d = 0.28 overall, and d = 0.18, marginal (p = .06), for objects that are not colour-diagnostic (Bramao 2011).
  • A claim that colour decides up to 90% of quick purchase judgments cites a study without a working link (Ethos3 2026, 2.5 of 20, example only).