Some visual properties are processed before attention arrives. A single red dot among fifty grey ones is found in about the same time whether there are fifty distractors or five hundred, because the search does not scan. Properties with that behaviour are called pre-attentive, and Colin Ware's Information Visualization: Perception for Design is the standard treatment. The working list for charts is position, length, size, luminance intensity, hue, shape, orientation and enclosure.
You can ask two separate questions of that list, and they have different answers. Which attribute does a reader decode most accurately? That decides what carries your quantity. Which attribute survives distance and colour blindness? That decides what carries your emphasis.
Which attributes does the eye decode most accurately?
For accuracy, the reference is Cleveland and McGill's 1984 paper in the Journal of the American Statistical Association, which ranked elementary perceptual tasks by measured error. Position along a common scale was decoded most accurately. Next came position along non-aligned scales, then a single band holding length, direction and angle together, then area, then volume and curvature, then shading and colour saturation. That paper doesn't order length, direction and angle against one another. They share the third rank. What puts length ahead of angle in practice is the pair of experiments behind it, position against length and position against angle, and that's why a sorted bar beats a pie for a comparison message. Carry the whole thing as a ladder. The interesting cases are the ones where you trade a step of accuracy for something else.
The Suryagarh channel-mix data makes the trade concrete. Modern trade holds 22.4% of revenue and general trade 21.5%, a difference of 0.9 points. As angle on a pie, 0.9 points is 3.24 degrees out of 360, which no reader resolves. As length on a 640-pixel bar axis, the same 0.9 points is 25.7 pixels, which every reader resolves instantly. The data didn't change and the difference didn't change. The encoding changed, and with it your reader's ability to see the thing the chart exists to show.
Advertisement
Which attributes survive distance and colour blindness?
For emphasis the ranking changes. The test is what survives degradation. Position and length survive everything: distance, projection, greyscale printing, every form of colour vision deficiency. Luminance intensity survives almost everything. So the accent-versus-neutral formula is built on a lightness gap, and a hue difference only supports it. Size survives distance well and greyscale perfectly. Hue survives distance if the hues are far apart in luminance as well as in hue, and fails badly otherwise. Shape and orientation fail at distance first. At eight metres a triangular marker and a square marker are the same grey smudge.
That gives you a rule you can apply without thinking. Encode the quantity with position or length. Encode the emphasis with luminance, supported by hue. Never encode anything that matters with shape, orientation or area, and never with hue alone.
Why can only one mark stand out?
One caution catches people who have just learned the idea. The effect only works when one thing, and only one, differs. Colour six of the eight bars in six different hues and highlight one in red, and the red bar is no longer pre-attentive. There's no uniform field left for it to pop out of. Emphasis comes from the contrast between one mark and its context. It isn't a property of the mark. That's the mechanism behind the colour formula, and it's why a second accent usually costs more than it buys.