Radar-chart axis order changes the polygon's shape and area

What it is

The radar chart draws dimensions as axes radiating from a centre, one record's values joined into a polygon. And it carries a fact nobody usually questions: axis order is a free parameter — five dimensions admit 5! circular arrangements, each drawing a differently shaped, differently sized polygon while the data stands untouched. Placing "looks strong" dimensions adjacently yields a plump convex polygon; interleaving yields a sawtooth — the impression dances with the arrangement. In scoring scenes (competency models, product ratings), radar charts get silently repriced by this arbitrariness: two products each choosing flattering orders turn the polygon comparison into retouched portraits side by side.

Why it happens

A polygon's visual quantities (shape class, area, concavity) are decided by adjacency: the line between adjacent axes is the only visible "relation between dimensions," yet dimensions have no order (they are parallel categories) — the sequence is imposed. Readers take bulges as "coherent strengths," dents as "weaknesses," and form an overall area impression ("bigger polygon, better overall") — but area varies with axis order (the same values convex-packed versus sawtoothed) and with per-axis scales (stretching one axis lengthens its neighbouring edges). One set of scores can therefore render as "comprehensive dominance" or "each with weaknesses," the choice resting entirely with the plotter.

Where it stops holding

Even the radar chart's few legitimate uses must declare this leaf: self-comparison (one object, two time points, polygons overlaid) is sturdier than mutual comparison, since axis order acts equally on both polygons; few dimensions (4–6) with business-meaningful order (process stages, body-check regions) anchors the arbitrariness semantically. When comparing objects, at least fix one shared order and publish it; better alternatives are parallel coordinates (dimensions on parallel linear axes — order still shapes adjacency but is interactively rearrangeable) or faceted bars (one group of bars per dimension, dissolving the shape illusion).

Applying it

• Order radar axes by business stages or a fixed published template; forbid "whatever looks good" or conclusion-driven reordering.
• For multi-object comparison, overlay on one chart (shared order) and append a parallel-coordinates or faceted-bar version for verification.
• Verification: render one record under three axis orders and ask readers to rate "overall performance"; ratings shifting more than half a grade with order means arrangement is pricing the impression.

Related

• Same group: U3.08.1 Polar coordinates map one variable to angle, the other to radius · U3.08.2 Equal radius increments give unequal areas; outer rings are systematically enlarged · U3.08.3 Polar coordinates suit inherently periodic variables · U3.08.5 Magnitude comparison under polar coordinates is harder than in Cartesian
• Nearby: U2.14.3 Axis order decides which patterns can be found · U2.14.5 Each axis needs independent normalization
• Search terms: radar chart · spider chart · axis ordering · parallel coordinates