Flattening moves choice cost from depth to breadth

What it is

Pressing a tree flat does not cancel choice; it changes how the bill is paid. Depth is serial: one judgment, one wait, one commitment per level. Breadth is parallel: scan the same screen, discriminate pairs, exclude. Flattening moves choice cost from depth to breadth means the total is still due, under a different heading. Fewer clicks and fewer loads are bought with a longer row of candidates and one heavier discrimination.

Reading “flat” as “less work” drops the transferred line. Whether the transfer is worth it depends on how expensive the waits used to be, and whether the unfolded row can still be scanned.

Why it happens

Depth cost stacks along time. Each judgment depends on the previous submission; loads slice decision-making so the next level cannot be previewed during the wait. Breadth cost stacks along space. Options share a screen; Hick–Hyman roughly describes reaction time rising with the log of set size, but in navigation the larger head is visual search and label discrimination: near-synonyms can bend the log toward a nearly linear read-every-item.

The error forms differ too. A depth error needs backtracking and is revealed only after the next load; a breadth error is a miss or a mix-up, committed on the same screen. Flattening kills the delayed “I learn I was wrong after the page arrives” error and creates the immediate “I scanned past the right item” error. Scent comparison shifts from “does this branch smell like the goal?” to “who in this row smells most like it?”—the comparison set grows, so relative differences must be stronger to remain usable.

Studying it

Split total time into between-level waiting and within-level search. After flattening, the first should fall and the second should rise; otherwise only the skin changed.

• Paradigms: deep tree versus flat list on the same tasks; eye tracking or sequential highlighting to count within-level inspections; throttle network delay to see whether the flat advantage grows as waiting gets dearer.
• Independent variables: number of levels, on-screen option count, between-level delay, near-synonymy of labels.
• Dependent variables: summed wait time, within-level search time, backtracks versus same-screen re-selections.
• Methodological note: lab menus often switch instantly and will understate waiting cost in real depth, hence understate flattening’s value on slow networks. Conversely, highly distinctive numbers or icons as labels will understate discrimination cost on vague copy. Set both delay and labels to the product’s real ones.

Where it stops holding

Native desktop apps with almost no between-level wait already have cheap depth; moving the cost to breadth may not pay. On a very narrow viewport, “flat” becomes a long vertical list and breadth cost gains a scroll term, so the transfer is not clean. Experts on familiar paths short-circuit both costs with spatial memory; the transfer is visible on novices. Voice navigation has no same-screen scan, so the cost structure does not cut on this axis.

Applying it

• Before flattening, measure the current depth bill: mean levels, load per level, backtrack per level. Then estimate option count and confusable label pairs on the target screen.
• If the depth bill is mostly waiting and backtracking, flattening is worth doing. If depth is already shallow and the pain is indistinct items, unfolding only lays the indistinction on screen one.
• After flattening, use separators or group headings to split peers into scannable clusters and shrink the discrimination set, without smuggling a fake hierarchy back in.
• Verify on the same tasks by logging “clicks + loads” separately from “dwell and re-selection on the nav.” The first falling and the second rising is the transfer happening; both rising means the flattening was done wrong.

Related

• Within the group: G2.02.1 A flat structure cuts depth and grows the sibling set · G2.02.2 Flat navigation fits limited, parallel content
• Adjacent: G2.04 Depth versus breadth tradeoff · E5.15 Number of navigation items · G1.04 Labeling systems
• Search terms: choice cost · Hick-Hyman · visual search in menus