Small multiples

Small multiples are one chart design drawn many times, once per category, and laid out in a grid. Edward Tufte gave them the name and the argument: once the reader has decoded the first panel, every other panel is free, because they all share a design, a scale, and an axis treatment. The alternative is putting all those categories into a single chart as separate series, which works for two or three and then collapses. Six lines in one frame cross each other, fight for the same colours, and need a key the eye has to keep visiting. Six panels of one line each need no key at all, because the panel title does the work a legend was doing.

The rule that makes the layout work is a shared scale. When every panel is drawn against the same domain, position means the same thing everywhere, and comparison becomes a matter of looking across the grid rather than reading numbers. Per-panel scales quietly destroy exactly that. Each panel fills its own box, every trend looks equally dramatic, and a category with a tenth of the volume of its neighbour renders at the same height. The chart still looks like small multiples and no longer works like them, which is the most expensive kind of failure, because nothing about it looks broken.

The layout it is most often confused with is the dashboard grid, and the two are opposites dressed alike. A dashboard grid holds different charts answering different questions, sized by importance, and a reader moves between panels to change subject. Small multiples hold one chart repeated so that moving between panels changes only the category, which is what turns comparison into a positional judgement. If your panels need individual titles explaining what each one measures, you have a dashboard. If the only thing that changes from panel to panel is which slice of one measure is drawn, you have small multiples.

Two practical decisions carry most of the quality. The first is panel order: alphabetical is almost never right, and sorting by the value the reader came for (the latest figure, the total, the rate of change) makes the grid itself say something. The second is how much apparatus each panel keeps. Axis labels repeated twelve times are noise, so the usual answer is to label the outer edges of the grid and let the inner panels carry only their title, the same economy that lets a sparkline drop its axis entirely. Keep the panels genuinely small. A grid of six charts each big enough to study on its own is just six charts.