Layout-matching skeleton screens beat spinners on perceived wait only for short loads

Hypothesis

For a page section whose data arrives within a few seconds, a skeleton screen (grey blocks in the positions of the coming text and images, with no layout shift when the content lands) is rated as faster and less annoying than a spinner covering the same area, even when the actual wait is identical. For waits noticeably longer than that, the ordering flips: a spinner with a message or a progress indication is rated better than a skeleton, because a static skeleton looks like a broken page and communicates neither progress nor an estimate.

Prediction

1. In paired comparisons with identical wait times, participants choose the skeleton variant as "faster" more often for short waits and the spinner-with-message variant more often for long waits, with a crossover somewhere between.
2. Skeletons that do not match the final layout (wrong number of rows, wrong image ratio) lose the short-wait advantage, because the content shift on arrival is itself experienced as a delay.
3. Adding a subtle pulse to the skeleton improves short-wait ratings slightly but does not change the long-wait result; adding an explicit status line ("Loading 40 results…") helps at long waits regardless of the visual.

Proposed test

1. Build one list page with four loading treatments: static skeleton, pulsing skeleton, centred spinner, spinner with a status line. All treatments announce completion through a role="status" live region and set aria-busy on the container, so the assistive-technology experience is held constant.
2. Serve the same content with artificial delays at several fixed durations (for example 0.5 s, 2 s, 5 s, 10 s) in randomised order.
3. Ask participants after each load to estimate the wait and rate acceptability; record the estimate error and the rating per treatment and duration.
4. Compare treatments within each duration; the crossover, if it exists, is the duration at which the skeleton's advantage vanishes.

Status

No result is claimed. The cited Nielsen Norman Group article gives the classic response-time limits (about 1 second for uninterrupted flow of thought, about 10 seconds for keeping attention, with percent-done indicators recommended beyond that) and motivates the choice of test durations, but it does not compare skeletons with spinners. The crossover point, if there is one, will depend on the content type and on whether the user initiated the load. Reduced-motion settings should switch off the pulse in every treatment.