Higher education routes to higher income , but employment status reshapes the path Figure 51.1 , Higher education routes to higher income , but employment status reshapes the path

Parallel Sets encode the joint distribution of multiple categorical variables simultaneously. Each dimension is represented as a vertical axis; the categories within each dimension appear as proportional segments (blocks) on that axis. Ribbons connect adjacent dimensions: each ribbon represents the population subset that belongs to a specific category in dimension A and a specific category in dimension B. Ribbon width is proportional to that subset's count.

Both charts use proportional ribbon widths to show flow. The critical difference is structural. A Sankey Diagram shows flows between nodes in an arbitrary network , nodes can appear at any position, flows can cross and split in any direction, and the topology is defined by the data's actual directed graph. It is designed for process flows: energy systems, supply chains, user journeys with multiple possible paths. A Parallel Sets chart imposes a strict rectangular axis structure.

The message involves the joint relationship between three categorical variables: Education, Employment Status, and Income. These are not process steps in a causal sequence , they are co-occurring attributes of a population. The question is distributional: "what proportion of the population occupies each combination of these three categories?" A grouped bar chart could show one pairwise relationship at a time but would require six separate charts to cover all dimension pairs. A mosaic plot (Marimekko) could show two dimensions simultaneously.

A Parallel Coordinates Plot , the visually closest alternative , encodes each observation as a polyline crossing multiple continuous axes. It is designed for continuous data: each observation's exact value on each axis is plotted and connected. Applied to categorical data, all observations within a category collapse to the same line, producing overplotted bands that are indistinguishable from one another without transparency hacks. Parallel Sets solves this by encoding aggregated counts as ribbon widths rather than individual observations as lines.

Paste this into Claude Code to generate a working version of this chart, plus its data file. The result will not be a perfect replica , the goal is that the reader can run the prompt, get a chart of this type, and read its source. Generate a complete, self-contained parallel sets in D3 v7. Two files:

The ideas in this chapter didn't appear from nowhere. John B. Sparks published the Histomap of Evolution (1932) and Histomap of World History (1931) , five-foot fold-out charts that ran from antiquity to the present, with each civilization rendered as a horizontal ribbon whose width tracked its relative power. Multiple categorical bands, time on one axis, magnitude in band-width, the whole thing readable as a single image: the parallel-sets form, drawn by hand on cardstock. John B. Sparks, circa 1933.

The ideas in this chapter didn't appear from nowhere. John B. Sparks published the Histomap of Evolution (1932) and Histomap of World History (1931) , five-foot fold-out charts that ran from antiquity to the present, with each civilization rendered as a horizontal ribbon whose width tracked its relative power. Multiple categorical bands, time on one axis, magnitude in band-width, the whole thing readable as a single image: the parallel-sets form, drawn by hand on cardstock. John B. Sparks, circa 1933. AI-generated portrait based on a public domain photograph (Wikimedia Commons).