Methodology
The survey begins with four colour-vision screening plates and one unrecorded practice question. It then presents 29 measured figure questions covering the design comparisons shown below.
Factorial cluster — Channel Family Fit × Channel Accuracy
Crosses Channel Family Fit (are the six arms kept as separate categories, or forced into a part-to-whole stack?) with Channel Accuracy (is the rate read off position -- the most accurate channel -- or area, one of the least?). Caution when interpreting the interaction: the two “fit = baseline” cells below use different violations of expressiveness -- a false part-to-whole stack (top-left) vs. a false trend implied by a connected line (top-right) -- to keep each cell's own accuracy channel consistent. A significant interaction term may reflect that these are different kinds of mistakes, not a clean fit×accuracy effect.
| Channel Accuracy: baseline | Channel Accuracy: treatment | |
|---|---|---|
| Channel Family Fit: baseline |
Stacked bar: forced part-to-whole framing (fit violated). Segment length is still a fairly good channel, but stacking undermines the comparison. |
Connected line across unordered arms: implies a false trend between categories (fit violated), even though position is the most accurate channel (accuracy correct). |
| Channel Family Fit: treatment |
Bubble chart: categories kept separate (fit correct), but circle area is a low-accuracy channel (accuracy violated). |
Sorted, separated bars: categories kept separate and position/length used to encode the rate (both correct). |