Group or aggregate when categories multiply

What it is

When category count passes the discriminability ceiling (previous leaf), the right move is not more colours but fewer categories. Two roads: aggregate — roll up the business hierarchy (dozens of cities into a few regions, hundreds of SKUs into product lines), returning colour count inside the ceiling with comparisons at the region level; group — keep the categories but subcontract the second level (major category gets a colour, members distinguished by position or facets), or the focus mode of "key categories each get a colour, the rest collapse into 'Other'". The choice follows where the question lives: at region level, aggregation is the destination; the question must reach detail, and grouping plus interactive drill-down is the way out.

Why it happens

Both moves exploit hierarchy as natural information compression: business data almost always carries a category tree (city → province → region, SKU → line → division), whose upper node counts naturally fit palette capacity. The focus mode (top-N plus Other) reallocates attention: the categories readers actually care about concentrate in the head (Pareto distribution), the tail's independent identity is rarely needed by any question, and a uniform grey "Other" preserves the head's contrast while honestly declaring the tail's existence and volume. Colour shifts from "an identity per category" to "an identity per question-relevant category," aligning encoding resources with attention.

Where it stops holding

Aggregation has a semantic price: roll-ups mask within-group variance and outliers (one city's collapse absorbed into the regional average), so analytical questions need drill-down as a safety net; static reports declare the aggregation level in the caption. The "Other" bucket has discipline: no more than about 30% of the total (when "Other" exceeds every named category, the wrong level was chosen); bucket colour is neutral grey, never a hue that could read as a real category. Do not aggregate preemptively when the count fits the ceiling — discriminable means give every category its colour; aggregation carries costs of its own.

Applying it

• Over the ceiling, trigger the decision tree: question level permits roll-up → aggregate and declare the level; detail is required → group/facet plus drill-down; head-heavy concentration → top-N plus grey "Other."
• "Other" bucket rules: volume ≤ 30%, always neutral grey, composition listed in the caption (or drill-down provided).
• Verification: after aggregating, ask readers the original question; answering without asking "what's in the bucket?" passes the level check — frequent probing means the granularity crushed something the question needed.

Related

• Same group: U4.01.1 The number of distinguishable categories has a ceiling · U4.01.3 Redundant shape or pattern encoding alongside colour
• Nearby: U2.12.3 With huge magnitude gaps, small flows vanish — long-tail aggregation helps · U6.05 Drill-down and roll-up
• Search terms: category aggregation · top-N others · colour grouping · hierarchical colour