A donut chart is a pie chart with its center removed. The primary encoding is arc length — the angle each segment subtends — not area. Human perception ranks arc length as more accurate than area for ordered comparison, making the donut a more honest composition chart than a filled pie. The hollow center is not decorative: it creates a stable focal point for contextual annotation, displaying a portfolio total at rest and a hovered segment's value on interaction — dual functionality that would require a separate callout in any other chart type.
Donut charts are suited to part-to-whole stories: when categories are mutually exclusive, exhaustive, and sum to a meaningful whole. The recommended maximum is five segments; beyond that, small arcs become difficult to compare and a bar chart or treemap is more readable.
This chart encodes $4.80B in humanitarian AI investment across five application domains — Healthcare & Medicine (34%), Climate & Environment (22%), Education & Access (18%), Economic Development (15%), and Crisis Response (11%). The five categories are mutually exclusive and exhaustive, satisfying the precondition for a part-to-whole chart.
Segments are ordered clockwise from 12 o'clock by descending value. This is deliberate: starting the largest segment at 12 o'clock anchors the viewer's reference point and makes the rank order immediately legible. The walnut-to-mist colour ramp is drawn from the hai palette, with shape and label providing redundant encoding so no information is carried by colour alone. Hover any segment to surface its value in the center ring.
A Donut Chart is a Pie Chart with its center removed. The primary encoding is arc length — the angle each segment subtends — not area. This is a meaningful distinction: human perception ranks arc length as more accurate than area for quantitative comparison. The hollow center is not decorative; it creates a stable focal point for contextual annotation. Here it displays the portfolio total by default, and the hovered segment's value on interaction. That dual function would require a separate callout annotation in any other chart type.
The data tells a composition story: five mutually exclusive, exhaustive categories that together constitute total humanitarian AI investment. The categories sum to a meaningful whole — the $4.80B figure — which is the precondition for any part-to-whole chart. With five segments (the maximum recommended before an alternative becomes more readable), the donut renders without clutter. A horizontal bar chart would show values more precisely but would not communicate "these are parts of one thing" as immediately or as honestly.
A bar chart encodes values as position on a common scale — the most accurate perceptual channel. For comparison tasks it would be strictly superior. But the message here is composition, not comparison. A bar chart implies "rank these" rather than "these are pieces of something." Viewers looking at bars do not naturally compute the total or perceive the parts as a unified whole. The donut's circular closure is not decorative — it is the visual argument that the dataset is exhaustive.
FT Visual Vocabulary: Part-to-whole. Abela quadrant: Composition — showing how individual parts make up the whole at a point in time. Tufte: the donut's removed center reduces non-data ink versus a filled pie while opening space for genuine data display.
The one decision worth knowing: segment order
starts at 12 o'clock and runs clockwise by descending value.
This is not default D3 behaviour — pie.sort(null)
preserves data order, and the data is pre-sorted. Starting the
largest segment at 12 o'clock anchors the viewer's reference
point and makes the rank ordering immediately legible.