Paging candidates costs more than retyping
Aliases: candidate paging · cheaper to retype · candidate page
What it is
If the target is not on the first screen of the candidate window, the user can page to hunt for it, or change the encoding, add letters, or recut the segmentation so it fits on the first screen. Paging is search along a list of unknown length; retyping is shrinking the lattice with a more distinctive code. For most words that are not extremely rare, time spent past the second screen usually exceeds the time to add a few letters and get a first screen again.
Why it happens
A candidate window holds few items per screen (fewer still on touch). A page turn is one visual search plus a tap; the prior on the next screen is only “the target is somewhere later,” with no positional memory, so three turns are three unanchored searches. Changing the code adds a constraint on the lattice: one more final, a tone, a longer word as the composition. Rivals drop exponentially, and the chance the target returns to screen one rises. Fluent users default to “not visible, add code”; novices walk the pages. Paging also hides a cost: each screen can be mis-tapped, and a mis-tap commits a wrong character that must be deleted before composition reopens, which dwarfs one extra letter of code.
Studying it
Place the target on screen 1, 2, or 4, allow or forbid recoding, and log time to the target and path (page / add code / abandon). Independent variables include items per screen, target screen index, and whether users were taught to add code; dependent measures include time, mis-taps, and the fraction who page. Scoring only “did the word eventually appear” writes high-cost paths down as success.
Where it stops holding
For extremely rare characters, or when the user does not know a longer code, paging or handwriting may be the only path. A desktop candidate window can show many items at once, and paging cheapens. Picking the 8th candidate with a number key is not paging; it is same-screen choice. If the list is indexed by radical or stroke, search has an anchor and the unanchored cost of paging does not hold.
Applying it
- Reserve the first screen for a few words likely in the current context, rather than using paging to digest a long flat list.
- Hint in the candidate window that “keep typing to narrow,” so novices do not default to paging.
- Measure take rate on screen two and beyond; if it is high, first-screen constraint is weak—add encoding distinctiveness before enlarging the page-turn control.