Math · A/B testing
Minimum detectable effect calculator.
Already know how much traffic you have? Find the smallest lift your test can reliably detect. Enter the baseline rate and visitors per variant. Everything runs in your browser.
With a 10% baseline and 5,000 visitors per variant, a two-sided test at 5% significance and 80% power can reliably detect a rate of 11.74% or higher: a minimum detectable effect of 1.74 percentage points, or a 17.4% relative lift. Smaller true lifts will often be missed.
Your test
Your result
Fill the form and press Calculate.
What it computes
The minimum detectable effect (MDE) is the smallest true improvement that the test will flag as significant at least power of the time (80% means 8 times in 10). It comes from the same relationship as the sample size calculator, solved the other way round: with baseline p1, rate to detect p2 = p1 + δ and pooled p̄ = (p1 + p2) / 2, find the δ for which
n = (zα √(2p̄(1−p̄)) + zβ √(p1(1−p1) + p2(1−p2)))2 / δ2
equals your visitors per variant. The equation has no closed-form solution for δ because p̄ depends on it, so the calculator finds it by bisection. The z values are the standard normal values for the significance level (1.960 at 5% two-sided) and the power (0.842 at 80%).
Worked example
Baseline 10%, 5,000 visitors per variant, 5% two-sided, 80% power. Trying δ = 0.0174 gives p2 = 11.74%, p̄ = 10.87%, and n = (1.960 × 0.4402 + 0.842 × 0.4400)2 / 0.01742 ≈ 5,022, a little more than 5,000, so the exact solution is slightly larger: δ = 1.744 points (11.74% versus 10%), which is a 17.4% relative lift. With 20,000 visitors per variant the MDE falls to about 0.86 points: quadrupling traffic halves the MDE because it scales with 1/√n.
How to read it
If your real lift is smaller than the MDE, the test is underpowered for it: the chance of a significant result falls below your chosen power. A non-significant result on an underpowered test is weak evidence of no effect. The MDE assumes a fixed sample, independent visitors and a normal approximation, and only describes the planned comparison.
Sources, opened 8 October 2026: NIST/SEMATECH e-Handbook of Statistical Methods, 7.3.3 How can we determine whether two processes produce the same proportion of defectives? (the pooled two-proportion z-test the power calculation is built on); NIST/SEMATECH e-Handbook, 1.3.6.7.1 Cumulative Distribution Function of the Standard Normal Distribution (1.645, 1.960); NIST/SEMATECH e-Handbook, 7.1.3.1 Critical values and p values (significance level and rejecting the null hypothesis). The power equation and its numerical inversion are standard normal-approximation algebra and are not quoted from these pages.
Frequently asked questions
What is a minimum detectable effect?
It is the smallest true change in the conversion rate that your test would detect (reach significance) at least as often as your chosen power. With 80% power it is the lift you would catch 8 times out of 10.
What is the minimum detectable effect for 5,000 visitors per variant?
On a 10% baseline at 80% power and a 5% two-sided significance level, 1.74 percentage points (10% to 11.74%), a relative lift of 17.4%.
How do I make the MDE smaller?
Collect more visitors. The MDE falls with the square root of the sample size, so four times the traffic halves it. Lower power, a one-sided test or a looser significance level also shrink it but increase the risk of wrong calls.
Is MDE absolute or relative?
Both are shown. Absolute is in percentage points (10% to 11.74% is 1.74 points). Relative divides that by the baseline (1.74 / 10 = 17.4%).
What if the true lift is smaller than the MDE?
The test can still show a significant result by chance or luck, but less often than your chosen power. Treat a non-significant result as inconclusive rather than proof of no effect.
Planning and learning aid
Planning estimate using a normal approximation for one comparison with a fixed sample. It does not adjust for multiple variants, multiple metrics or repeated looks at the data.