Question

Ask what a nonsignificant study could have missed

An uncertain answer should not be promoted to a confident zero.

When it fits

  • A report says there was no effect because a test did not cross a significance threshold.

When to avoid it

  • Do not replace the analysis with post-hoc storytelling about power; examine the actual estimate, interval and design.

A question to ask

Which effect sizes remain compatible with the interval and assumptions? · Was the design capable of resolving the change that matters? · Was equivalence actually tested with justified bounds, or merely asserted after a nonsignificant result?

Why it matters

Inspect the effect estimate, interval and the study's ability to detect a meaningful difference. Power depends on a specified effect and design. A small or noisy study may leave both useful benefit and important harm unresolved even when its p-value is large.

An example

An imprecise comparison cannot establish that two tools perform equally just because neither is declared a winner.

Check your result

The conclusion distinguishes evidence of little difference from insufficiently precise evidence.

Keep this limit in mind

  • Do not replace the analysis with post-hoc storytelling about power; examine the actual estimate, interval and design.

Evidence and sources

Supports

Statistical power concerns the probability of rejecting the null under a specified alternative and design.

Power is not a single property independent of the effect size, variability, sample size and analysis.

Quantitative Techniques · Power and type II error

All sources (1)