Broad-and-shallow versus narrow-and-deep depends on category distinctiveness

What it is

A tree is shallow when one layer holds many categories that can be told apart at a glance. It is deep when a layer holds a few indistinguishable buckets that then split below. Broad-and-shallow versus narrow-and-deep depends on category distinctiveness: breadth only buys “one fewer click” when names can be distinguished without opening them. If the difference is visible only after opening, breadth merely spreads confusion across the first screen. Distinctiveness is contrast among labels, not an aesthetic of how many categories there are.

This is about how the classification tree is cut, not about how many controls a navigation bar can hold. Fitting on the chrome is not the same as being recognizable.

Why it happens

Every descent is a decision that also waits. If categories are already separable in preview, elimination finishes inside one layer, total decisions drop, and a shallow tree pays off. If categories nest in each other or are near-synonyms (“Solutions” and “Services,” “News” and “Updates”), more options in a layer still fail to eliminate, and people open each to look—breadth becomes parallel depth. Breadth/depth studies such as Larson and Czerwinski found wide trees performing well under distinguishable names. Flattening indistinguishable labels does not reproduce that result.

Distinctiveness comes from contrast, not from precision. Two precise words that point at the same thing (“employee” and “staff”) are not distinct; two coarse but opposed words (“personal” and “business”) are. Get contrast first; only then ask how wide the layer can run.

Studying it

Treat “how many in a layer” and “can these be told apart” as separate factors. Do not vary depth alone.

• Paradigms: tree tests comparing wide-shallow with narrow-deep, but only after pairwise judgments or name-the-card tasks have measured distinctiveness, then stratify high versus low.
• Independent variables: depth, width per layer, distinctiveness of names (quantifiable in advance with a confusion matrix).
• Dependent variables: direct success, backtracks, dwell between similar categories, and whether errors concentrate on confusable pairs.
• Methodological note: many “breadth wins” results used distinctive numbers or clear topic words. Applying “wider is better” to vague marketing categories on a corporate site drops the premise. Confusion structures differ in Chinese and English; measure distinctiveness in the target language.

Where it stops holding

Expert users who remember paths offset extra deep-tree decisions with spatial memory, so narrow-and-deep is not always slower. On very narrow screens or in spoken menus that can read only a few options, physical width is capped even when categories are distinct, so the tree must deepen and rely on stronger preview. Distinctiveness also shifts with learning: after a few uses, a once-confusable pair may be remembered by location, and a first-glance distinctiveness test will underestimate veterans’ tolerance for depth.

Applying it

• Run a names-only distinction test on each layer: can someone uninvolved state the difference between any two names. Rename or merge those that fail before deciding to widen.
• Only distinctive categories may share a layer. Six clear ones beat twelve that rewrite each other.
• Every layer of a deep tree must still cut on a distinctive opposition. Do not copy an indistinguishable bucket downward.
• Verify: if tree-test errors cluster on one pair, that is a distinctiveness problem; widening or deepening will not help. Make that pair eliminable in opposite directions, then compare broad-shallow with narrow-deep again.

Related

• Within the group: G1.03.1 A hierarchy is predictable only when categories are exclusive and exhaustive · G1.03.2 Unclassifiable content exposes the wrong dimension
• Adjacent: G2.04 Depth versus breadth tradeoff · G1.04 Labeling systems · G2.01 Hierarchical navigation
• Search terms: breadth versus depth · category distinctiveness · taxonomy width