A law's ordering of alternatives is more credible than its absolute value

What it is

Interaction laws are often better at predicting which alternative has lower relative burden under matched conditions than at stating how many exact milliseconds a real task will take. This relative prediction can support early ranking and trade-offs; absolute values are affected by device, user, task stage, and unmodelled processes.

Why it happens

When similar alternatives are compared, many fixed costs and system errors are shared and partly cancel in the difference, stabilising directional relations driven by the model variable. Absolute time requires accurately covering search, choice, movement, waiting, strategy, and device delay; any omission shifts total value. Relative order still needs validation in target conditions.

Studying it

Use a model to rank candidates, then test representative tasks to check whether rank agrees with observed time, errors, or key outcomes. Report prediction–observation association, rank agreement, and uncertainty rather than only absolute residuals. When alternatives differ little, use individual variation and practical significance to decide on further testing.

Where it stops holding

If alternatives change task structure, user population, device, or an unmodelled stage, even ranking can fail. Absolute estimates still matter for capacity, service level, and accessibility timing, but should carry ranges and conditions. Ranking neither implies universal preference nor covers safety, understanding, and trust.

Related

• Same group: B1.17.1 Laws calibrated under separate conditions double-count preparation and response when added · B1.17.2 Parallel operations cannot be estimated by adding time · B1.17.3 Adding group-average predictions amplifies rather than cancels error · B1.17.4 A predicted difference smaller than individual variation is not design evidence
• Nearby: B1.01 Fitts's law · B1.04 Hick-Hyman law
• Search terms: relative prediction · design ranking · model validation