Stack order decides which series stay readable

What it is

Task-driven stack ordering treats layer sequence as an encoding choice. In a conventional nonnegative stack, the series next to zero is easiest to compare across positions, outside layers are easier to locate and label directly, and middle layers rely mainly on thickness. Reordering changes neither component values nor the total, but it changes baselines, adjacency, visual competition, and traceability. Order should therefore serve the primary question while remaining stable enough to preserve identity, rather than following source order, randomness, or a new rank at every time point.

Why it happens

Each boundary accumulates the series ordered before it, so reordering changes which fluctuations propagate through later boundaries. The placement of volatile or large layers alters internal wiggle, thickness contrast, and prominence near the outer silhouette. It also determines which colours or patterns touch, affecting boundary separation. For streamgraphs, the baseline algorithm and layer order jointly determine geometry. Byron and Wattenberg's inside-out ordering addressed burst-and-decay time series and a particular balance of legibility and aesthetics; it is not a universally optimal order for area charts.

Studying it

With data held constant, studies can compare task-priority order, fixed domain order, size order, onset order, inside-out algorithms, and interactive reordering. Tasks include locating a target, comparing thickness, finding peaks, maintaining identity, and reading totals. Manipulations should include layer count, volatility, labels, colour or pattern, baseline algorithm, device size, and uncertainty. Beyond accuracy and time, measure target loss, legend lookups, identity confusion after reordering, and confidence. Validate algorithms on the intended distribution; aesthetic preference, rapid detection, and analytical correctness are different outcomes.

Where it stops holding

Semantic order can outweigh perceptual optimization for hierarchies, process stages, or categories with a conventional direction, while consistency across charts and time aids identity. Diverging stacks require a positive–negative split before ordering each side. Missing series should not cause other layers to jump positions from one period to the next. Dynamic value sorting can place a leader at an edge but breaks continuous tracking; it fits explicit rank-change tasks only when transitions and labels preserve identity. When uncertainty dominates small differences between layers, show intervals or data quality rather than implying that a fine ordering is meaningful.

Applying it

• Define the primary task and choose one stable order for the view: place a priority component next to zero, preserve meaningful domain sequences, and evaluate specialized streamgraph orders for dense exploratory flows.
• Do not silently reorder on filtering, missingness, or refresh. If reordering is necessary, retain stable colour or pattern, direct labels, transitions, and a reversible sorting control.
• Check adjacent layers under colour-vision variation, grayscale, and small screens. Expose series order, values, totals, and the ordering rationale to screen readers.
• Compare at least two plausible orders with representative tasks, tracking target loss and wrong answers. Select the order that meets task tolerance while preserving identity, not one universal algorithm.

Related

• Same group: U2.04.1 Only the bottom series of a stacked chart owns the common baseline · U2.04.2 Upper series read only by band thickness, at markedly lower precision · U2.04.3 Stacking suits totals, not per-series trends · U2.04.4 Percent stacking discards the totals
• Nearby: U1.11.1 The same entity keeps the same colour and position across a chart series · U4.01.3 Redundant shape or pattern coding alongside colour
• Search terms: stack ordering · inside-out ordering · streamgraph · layer identity