Percent stacking discards the totals

What it is

A 100% stacked chart divides each component by the total at its category or time point, making every stack reach 100%. It preserves component shares and their sum-to-one constraint while removing absolute magnitude: groups with different totals but identical composition produce the same stack. The form is therefore suited to “how does composition differ?” but cannot alone answer “how large is it?” or “how did the amount change?” This loss is a consequence of normalization, not automatically a data-processing error; it matters when the decision needs the denominator.

Why it happens

Dividing each position by its own total changes the unit from an absolute measure to a proportion and flattens the silhouette. A layer can thicken because its numerator grew, other components shrank, or both occurred. Share change cannot identify absolute change. Area and colour help maintain component identity, but only layers next to 0% and 100% have a shared edge; middle layers still require thickness judgments. When sample count, exposure, or eligible population differs across periods, equal-width 100% stacks can also make unequal evidence volumes look alike.

Studying it

Studies can pair datasets with equal shares but unequal totals, or rising shares and falling counts, and compare percent stacks, absolute stacks, a total line plus composition, and denominator-labelled versions. Separate tasks should test shares, absolute values, causal attribution of change, and decisions. Outcomes include accuracy, time, confidence, and whether readers seek denominator information. Sample size and decision consequence belong in the design. A study asking only proportion questions cannot show that readers will avoid absolute inferences, while an always-visible denominator may not represent an unlabelled product chart.

Where it stops holding

Percent stacking is compact when absolute totals are outside the task, or when equal denominators are already known. A zero total makes proportions undefined. With negative values, “percentage of total” may exceed intuitive bounds or become meaningless through cancellation; direct 100% stacking is usually inappropriate. Missing components change the visible denominator, so the display must state whether shares use a complete population, known components only, or renormalized available records. Ratio uncertainty depends on numerator and shared denominator and should not be represented by simply stacking independent component intervals to 100%.

Applying it

• Decide whether absolute scale affects the decision. If it does, add a total line, denominator labels, an absolute view, or a unit toggle instead of asking the percent stack to serve both questions.
• Label the unit as percent and define the denominator, time window, missing-data treatment, and weighting in titles, axes, or accessible descriptions.
• Show “proportion undefined” for zero totals. Use diverging absolute values or separate net, positive, and negative totals when signed data make composition incoherent.
• Expose totals and component counts to keyboard, screen-reader, and download paths. Test with a case where share rises while count falls to detect numerator–denominator confusion.

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 · U2.04.5 Stack order decides which series stay readable
• Nearby: U1.08.5 The same field can take different types under different analytical tasks · U2.17.1 A mismatched chart forces readers to compute the answer mentally
• Search terms: 100% stacked chart · normalization · denominator effect · composition versus magnitude