How it works
The MDE is the sample size formula run backwards. For your baseline p₁ and visitors per group n, the calculator searches for the target rate p₂ at which the required sample exactly equals n:
It uses bisection to 200 iterations, which is far more precise than you'll ever need. The answer round-trips exactly: plug the MDE into the sample size calculator and you get your n back.
Example
A store gets 14,000 visitors a week to its product pages and converts 3%. The team can give a test three weeks.
- Visitors per group: 14,000 × 3 / 2 = 21,000
- Smallest detectable lift: about +16.1% (3.00% to roughly 3.48%)
With one week instead of three the MDE climbs to about +28.7%. That number tells you which ideas are worth testing. A new trust badge isn't going to lift conversion by 16%. A new offer structure or a redesigned product page might.
Using MDE to choose what to test
Write down your test ideas and, for each, a realistic guess at the effect. Anything with a realistic effect well below your MDE is a poor test candidate. Either bundle several small changes into one bolder variant, ship it without testing if the risk is low, or move it to a page with more traffic.
This is the single most useful check in a testing program, and it takes ten seconds.
Common questions
What is minimum detectable effect?
The smallest true difference between variants that your test will flag as significant with the power you chose, usually 80%. Effects smaller than the MDE can still win, but your test will miss most of them.
Why is the detectable drop different from the detectable lift?
Because the variance of a conversion rate depends on the rate itself. Near small baselines a drop and a rise of the same relative size aren't equally easy to detect. At 3% with 20,000 per group, you can detect a lift of about 16.6% or a drop of about 15.3%.
What MDE is realistic for e-commerce tests?
Many well-run tests on mature stores produce lifts in the low single digits, if anything. Bold changes to pricing, offers or page structure can move more. If your MDE is 25% or higher, assume most tests will come back inconclusive unless you test big ideas.
Should I lower my standards to get a smaller MDE?
Moving from 95% to 90% confidence shrinks the MDE a little, at the cost of more false positives. It's a legitimate choice for low-risk changes. Write it down before the test starts rather than deciding after you see the data.