Diverging palettes suit data with a meaningful midpoint

What it is

A diverging palette is two sequential ramps joined at a centre: one colour at the middle (usually neutral grey or white), each side deepening toward its own hue. The question it serves is "how far from the benchmark, and in which direction" — the data must carry a semantically meaningful midpoint: zero (profit-loss, year-over-year change), the mean (deviation from average), a target (completion versus 100%), a base period (index versus baseline). The two sides compare symmetrically (how much above versus below), and diverging encoding hands "direction" to opposing hues and "magnitude" to lightness distance — exactly this data's shape.

Why it happens

The diverging structure's advantage is rendering the benchmark visually explicit: the centre colour paints the benchmark onto the ramp, and reading becomes two steps — first which side of centre (direction; hue judgement is fast and categorical), then how far from centre (magnitude; lightness distance). A sequential palette doing the same job forces readers to remember "which value is the benchmark" and compute deviation in the head — implicit, internal. Diverging also naturally spotlights extremes (both ends deepest and most salient), fitting "find anomalous deviations"; the neutral mid-band lets the mass of "no significant deviation" data recede quietly, concentrating attention at the two ends.

Where it stops holding

The precondition is a midpoint that genuinely means something: total-ordered data without a natural benchmark (temperature, income, rank) under a diverging palette erects the median into a false "dividing line," and readers hallucinate "two classes, high and low" — a carved-up continuum; such data belongs to sequential palettes. The midpoint semantics must be statable ("relative to the national average"); an unstatable midpoint equals none. Data with two zero points (independent zero directions, like two-party vote shares) take stacked/flow solutions; one-way data (positive deviations only) reverts to sequential.

Applying it

• Answer two questions before diverging: does the data have an agreed benchmark? does the question care "which way and how far"? Two yeses enter diverging.
• Declare the midpoint with the chart ("midpoint = industry mean, 47"), with symmetric value extents on both arms (third leaf).
• Verification: ask readers "what does the centre line stand for?"; a blank answer means the diverging failed — read as a sequential ramp, direction information lost.

Related

• Same group: U4.03.2 Midpoint placement changes the conclusion · U4.03.3 Diverging ends must be symmetric in luminance
• Nearby: U4.02.1 Sequential palettes must increase monotonically in perception · U2.10.3 Colour scale choice directly shapes the conclusion
• Search terms: diverging palette · midpoint semantics · deviation map · reference value