Rule of thumb · MathematicsNº 105 / 122

“No significant difference” usually means “too small to tell”

An underpowered test cannot distinguish a real effect from none, so a null result from one carries almost no information.

Why it works

Work out the smallest effect your actual sample could have detected. If that number is a 25% lift, then a null result has ruled out nothing anybody hoped for — almost no real change moves a conversion rate that far. The test did not show the change does not work; it showed the test could not see whether it did, which is a different sentence with a different conclusion.

When it fails

A null result from a well-powered test is genuinely informative, and treating every null as an artefact is the opposite error. If the test could reliably have caught a 3% lift and did not, that is real evidence the effect is small — which is often exactly what you needed to know.

Do it exactly

Estimate with the rule, then check it against the calculator that models it properly.

Open A/B Test & Sample Size →

More in Mathematics

← PreviousHalving the effect quadruples the sample Next →A loop that only overshoots on big steps is winding up, not mistuned

Why did my A/B test find nothing?

An underpowered test cannot distinguish a real effect from none, so a null result from one carries almost no information. Work out the smallest effect your actual sample could have detected.

Browse all 122 rules of thumb →