Small multiples replace overlay with repeated like-for-like panels

What it is

Stack many series into one plot, or give each series its own like-for-like panel and tile them into an array — small multiples take the second road. The wager: repetition with identical structure is itself readability. Each panel carries only one or two things (one series, one group, one period); between panels everything but the splitting dimension is identical (axes, scales, colours, marks); once the eye learns the first panel, reading all the rest is nearly free. The crossing, occlusion, and legend bloat that plague overlay charts do not exist structurally in small multiples.

Why it happens

Small multiples shift the burden of comparison from perception to layout: overlay demands that readers separate multiple visual groups within one plot (suffering crossings and occlusion), while small multiples give each group its own panel and turn comparison into explicit panel-to-panel contrast — precisely the pattern matching the visual system excels at ("which panel differs" is nearly preattentive). Structural identity also removes relearning: axes and legends are read once, and attention goes entirely to data differences. The price is per-panel resolution: each panel holds a fraction of the page, and detail and read-off capacity shrink accordingly (fifth leaf).

Where it stops holding

The precondition is a splitting dimension that is categorical and moderate in count (a few dozen panels at most — fourth leaf); when the splitting dimension is continuous (time), animation or sliding facets approximate the idea. A small amount of within-panel overlay (three lines per panel, say) is permissible, with the ceiling returning to line-tracking constraints. One array serves one comparison task; multiple independent questions go back to "one question, one chart" (the task family's first group) rather than one faceted array answering everything.

Applying it

• Default to small multiples beyond three or four series, groups, or periods; keep overlay only within two or three.
• Enforce structural identity in the facet template: shared axis ranges and ticks, same palette and mapping, same chart type — only panel titles differ.
• Verification: ask readers to spot "the panel with the most anomalous shape" and "the key difference between two panels"; instant success means like-for-like comparison works.

Related

• Same group: U2.15.2 All facets must share the same scale to be comparable · U2.15.3 Facet order should carry information, not alphabetical habit · U2.15.4 Too many facets shrink panels below readability · U2.15.5 Small multiples compare shapes, not precise values
• Nearby: U2.03.3 Beyond a few lines, tracking collapses · U2.01.3 The same data supports several tasks; draw a chart for each
• Search terms: small multiples · faceting · trellis · panel chart