Breadth beats depth when categories are distinctive

What it is

If the classes on a level can be told apart before anyone clicks, people can finish excluding on that screen and skip the decisions and waits below. Breadth beats depth when categories are distinctive is a claim about navigation performance: the same set of endpoints, laid out as a wide menu of glanceable labels, is usually faster and needs less backtracking than several murky descents. It is not about how a taxonomy should be cut in cataloguing. It is about already-chosen names, used as a menu: go wide, or keep narrowing level by level.

When distinctiveness is low the result flips. Spreading indistinguishable items across a screen makes people treat the screen as several unknown-depth paths and open them one by one; breadth becomes many depths in parallel.

Why it happens

Navigational choice is exclusion. When two labels contrast in preview (“Personal / Business,” “Flights / Hotels”), exclusion ends at the current level, the levels below need not exist, and a whole slice of unit cost—decision plus wait—is deleted. The wide-menu advantage in Larson and Czerwinski, Snowberry and others, rested on separable materials: digits, clear topic words, not marketing near-synonyms.

When contrast is weak, every item on the level keeps a residual scent of “maybe this,” and the rational strategy is to sample by opening. Each opening pays the unit cost of depth; more items, more samples; the wide menu’s bill exceeds a narrow deep one. Distinctiveness is therefore the switch for a breadth advantage, not decoration: it decides whether “fewer clicks” can be cashed. Mutual exclusivity in the catalogue does not automatically make on-screen words separable. The navigation tradeoff is judged on the words users see, not on definitions in the classification table.

Studying it

Treat “how many on a level” and “can these be told apart in preview” as two factors. Any replication of a breadth advantage must report the second.

• Paradigms: pairwise judgments or a confusion matrix for label separability first; then wide-shallow versus narrow-deep menus or tree tests stratified by high/low distinctiveness.
• Independent variables: depth, width per level, distinctiveness measured in advance.
• Dependent variables: time to complete, backtracking, first click, dwell between confusable pairs.
• Methodological note: “wide beats deep” on highly separable lab materials (digits, country names) does not transfer to “Solutions / Services / Solutions we offer.” Confusion structure differs across languages; measure distinctiveness in the target language and with target users. Cataloguers’ “these two classes have different definitions” is not a substitute for users’ preview judgments.

Where it stops holding

Experts have finer discrimination; classes that are inseparable for novices may be separable for them, so a breadth advantage appears earlier. Conversely, “must go deep” on a novice sample should not be pasted onto expert tools. Icons or example images can raise preview separability and change the tradeoff without recutting the taxonomy. When the screen cannot scan that wide, distinctiveness cannot cash a breadth advantage; the scannable-count cap binds first. Known-item lookup by a proper name does not travel this path at all.

Applying it

• Before trading width against depth, test the labels: show a level’s candidates pairwise and ask which one, unopened, is more like target X. Pairs that cannot be told apart should not share a level.
• On a separable level, prefer going wider and dropping a level, deleting that level’s decision and wait wholesale.
• On an inseparable level, do not go wider to “save a click.” Rename or recut first, then talk about depth.
• Verify with one distinctive wide menu and one near-synonym deep menu on the same endpoints. High first-click and low backtracking on the wide version means the breadth advantage landed; serial probing on the wide version means fix distinctiveness before adding more width.

Related

• Within the group: G2.04.1 Each added level costs one more decision and one more wait · G2.04.2 Breadth is capped by how many items can be scanned
• Adjacent: G1.03 Taxonomy and hierarchy · G1.04 Labeling systems · Q2.12 Tree testing
• Search terms: category distinctiveness · breadth versus depth · information scent