Stacked and Composed Charts

Stacked charts answer how a total divides into contributions. They work best when the total and a small number of stable components both matter. Interior layers do not share a baseline, so their individual values are harder to compare than the first layer or the total.

Use a normalized stack when proportion matters more than magnitude. Use a mosaic when both column width and internal height carry part-to-whole meaning. Use a streamgraph only when changing shape is the primary story and exact values remain available elsewhere.

Choose the composition#

Single-value bar and area channels stack implicitly. Use layout: stack() when the order or offset must be explicit; supply interval endpoints when the application has already computed them.

Preserve totals with a stacked area#

A stacked area combines a shared ordered x domain with one length per series. The top boundary carries the total; the thickness of each layer carries its contribution.

areaY(rows, {
  x: 'date',
  y: 'value',
  color: 'series',
  layout: stack({ order: ['Core', 'Services'] }),
})

Keep series order stable across updates. Reordering layers can make unchanged values appear to move substantially and breaks the reader's spatial memory. When one series needs precise comparison, place it on the shared baseline or give it a separate aligned view.

The original value remains available to tooltips and selection while the mark derives its stack endpoints.

Center an ordered response scale#

Survey counts are nonnegative, but an ordered response scale often reads as diverging. Group the observations into counts, then anchor the stack on the neutral category:

const counts = groupBy(responses, {
  by: { question: 'question', response: 'response' },
  outputs: { count: { reduce: 'count' } },
})

barX(counts, {
  x: 'count',
  y: 'question',
  z: 'response',
  color: 'response',
  layout: stack({
    order: ['Strongly Disagree', 'Disagree', 'Neutral', 'Agree'],
    anchor: { series: 'Neutral', fraction: 0.5 },
  }),
})

The grouping transform owns counts and lineage. The stack owns ordered endpoints and the per-question zero translation. Response order, neutral choice, and anchor fraction remain explicit chart meaning.

Compare proportional mix#

A normalized stack gives every x position the same total height. It answers which series gained or lost share, but deliberately removes the original total magnitude.

Format the quantitative guide as a percentage and state the denominator. Keep raw totals available in a tooltip, table, or companion view when the reader may otherwise mistake stable share for stable volume.

Use layout: stack({ offset: 'normalize' }). Normalization is resolved independently at each x position.

Emphasize changing shape#

A streamgraph offsets and orders layers to reduce visible oscillation around a central baseline. It is effective for the broad shape of many positive series, but the displaced baseline makes precise values and totals difficult to read.

The source can stay in tidy form; stacking belongs to the area definition:

areaY(industries, {
  x: 'date',
  y: 'unemployed',
  z: 'industry',
  color: 'industry',
  layout: stack({ offset: 'wiggle', order: 'inside-out' }),
})

Keep the offset and order stable across revisions, preserve series colors, and provide exact values through Tooltips and Focus.

Use an ordinary stacked area when totals or baselines are part of the question.

Show shares as fixed units#

A waffle chart trades precise length comparison for countable, equal units. Use it when one tile has a clear meaning, such as one percentage point.

waffleY(alphabet, {
  y: 'frequency',
  color: 'letter',
  unit: 0.01,
  round: true,
  gap: 2,
  radius: 2,
})

waffleY expands source values internally and preserves each source row for tooltips and selection. unit defines one complete cell; round: true rounds cumulative boundaries so category rounding does not change the overall total. Leave columns unset for responsive square-cell packing, or set it when the grid dimensions are part of the encoding.

Encode two part-to-whole dimensions#

A Marimekko chart uses column width for one categorical total and vertical composition for a second. Each cell is an explicit rectangle with both horizontal and vertical interval endpoints.

The two dimensions have independent denominators: response totals determine each question's column width, while response-category shares determine height within that question. Use groupBy to state the count or sum, then mosaicY to allocate both normalized interval dimensions. Ordinary rect and text marks render the result. Small cells may need a tooltip or adjacent table rather than unreadable direct text.

The transform contract is defined in Data Transforms, and rectangle endpoint semantics are defined in Bar and Rect Marks.

Production checks#

• State whether the chart preserves totals or normalizes every group.
• Keep series order, category order, colors, and keys stable.
• Include zero for ordinary positive stacks and an explicit zero rule for diverging stacks.
• Preserve raw values alongside derived endpoints and proportions.
• Avoid too many layers; group minor categories only when the aggregation is defensible and disclosed.
• Use a legend or direct labels that remain meaningful in light and dark themes. See Legends and Color.
• Verify keyboard focus and exact-value access with Accessibility.

Marks and Layering explains how the interval areas, rules, labels, and highlights compose into one chart.