Upper series read only by band thickness, at markedly lower precision

What it is

In a stacked area chart, an upper component is encoded by band thickness: the vertical difference between its lower and upper boundaries at a given horizontal position. Neither the two-dimensional filled area, the upper boundary's height above zero, nor the boundary's slope is the component value. Judging this nonaligned difference is generally harder and less precise than locating endpoints against a common baseline. “Markedly lower” is an empirical tendency under particular tasks and stimuli, not a universal numerical discount.

Why it happens

Thickness inherits localization error from both boundaries. When the edges are nearly parallel and change slowly, vision can form a useful local-width estimate. Diverging slopes, thin bands, dense layers, or weak adjacent contrast make readers establish the vertical measurement direction while suppressing the edges' trajectories. Filled area mainly groups a layer across time; it does not turn the value at one instant into an area judgment. Connecting boundaries likewise express continuity but can tempt readers to follow a cumulative edge instead of taking the difference that encodes the component.

Studying it

Controlled studies can manipulate true thickness, upper and lower slopes, target position, neighbour contrast, sampling density, display size, and uncertainty rendering. Separate tasks should test exact lookup, pairwise comparison, trend direction, and anomaly detection. Aligned length, nonaligned position, direct labels, and interactive probes provide useful controls. In addition to accuracy and time, collect omissions, confidence, and speed–accuracy trade-offs. Large effects and highlighted targets may overstate readability, while stripping away labels and interactions used in the product may understate it.

Where it stops holding

Thickness may be sufficient for detecting presence, a peak period, or a large change. Direct labels and on-demand values can support exact lookup, but they change the reading task rather than proving that an unassisted band is precise. Discrete stacked-bar segments have straight, separated edges and may be easier than continuous curved area bands; the two are not interchangeable stimuli. Both edges of a streamgraph move with its baseline, favouring overall flow over exact values. If uncertainty is drawn as another adjacent band, it can be mistaken for data; it needs a distinct, explained encoding.

Applying it

• Provide small multiples, a standalone line, or a selectable focus view when upper layers require exact values or subtle trends. Treat labels and probes as supplements.
• Preserve boundary and luminance contrast, and add direct labels, patterns, or programmatic names for colour-vision differences and screen-reader access.
• Break or mark missing observations. Do not let interpolation masquerade as observation, or use indistinguishable area fills for both values and uncertainty.
• Test at production size and density, comparing band-derived answers with the underlying data. Use task tolerance rather than a universal error threshold.

Related

• Same group: U2.04.1 Only the bottom series of a stacked chart owns the common baseline · U2.04.3 Stacking suits totals, not per-series trends · U2.04.4 Percent stacking discards the totals · U2.04.5 Stack order decides which series stay readable
• Nearby: U1.02.2 Losing the shared baseline sharply degrades position judgment · U1.04.1 Perceived area systematically underestimates actual area
• Search terms: band thickness · nonaligned position · stacked area · graphical perception