Midpoint placement changes the conclusion

What it is

A diverging palette carries one parameter with no default answer: where the midpoint sits. Mean, median, zero, target, last year, industry benchmark — every candidate is legitimate, and each draws a completely different map: move the midpoint down and half the sheet "goes red"; move it up and the screen "floods green." The data is unchanged, but the conclusions about where is abnormal, what counts as on-target flip with the midpoint. This is not a mathematical defect of the palette — it is the semantic core of diverging encoding: it visualizes a choice of reference, and choosing the reference is choosing the frame the conclusion lives in. That choice must be explicit and auditable, not buried in a colour function's default arguments.

Why it happens

The midpoint is the origin of the diverging map's coordinate system: every cell is split into "above benchmark" versus "below benchmark," and the opposing hues render that split as the most salient visual category on the page (a red-green opposition out-competes any gradient). So one move of the midpoint translates the category boundary wholesale — every region it sweeps flips from one class to the other, and the map's whole "layout" is redrawn. The midpoint also controls how contrast is allocated: when the value range is skewed relative to the midpoint (most values above the mean, say), one arm saturates quickly while the other sits idle, and the deepest colours land on only a handful of cells — whether "anomaly" develops at all is decided by the midpoint's position relative to the range.

Where it stops holding

Mechanical sensitivity and selection discipline around the midpoint are the design-side concern; deliberately moving the midpoint to manufacture alarm or optimism — the intent and its consequences — belongs to the ethics domain. A legitimate midpoint follows the data's semantics: deltas take zero; completion rates take 100% or the target; deviations take the chosen reference (mean / base period / control value), stated numerically in the chart title. Beware skewed distributions when the mean is the midpoint — extreme values drag it, and a median or trimmed mean holds steadier. A sensitivity check should be routine: nudge the midpoint one notch up and down and see whether the conclusion flips; if it flips, the conclusion must be stated as an interval, not a point.

Applying it

• Every diverging deliverable carries three items: the midpoint value, the midpoint's rationale (why this one), and the value extents at both arms; missing any one fails review.
• Sensitivity self-check: re-render with the midpoint shifted ±10%; if the "anomalous region" changes area or position dramatically, switch to interval conclusions and disclose the sensitivity.
• Verification: give readers the same data in two versions with different midpoints and ask each for their "main finding"; a demo where the two findings contradict each other is the training case for exactly this failure.

Related

• Same group: U4.03.1 Diverging palettes suit data with a meaningful midpoint · U4.03.3 Diverging ends must be symmetric in luminance
• Nearby: U2.10.3 Colour scale choice directly shapes the conclusion · U3.04.1 Dual-axis relative position is a human choice
• Search terms: midpoint choice · diverging scale · sensitivity analysis · reference framing