Free tool Minimum detectable effect calculator
When traffic and time are fixed, the real question is what the test can see. Enter your baseline rate and either the visitors per group or your daily traffic and test length: the calculator solves the smallest lift that the test would detect at your significance level and power.
- Inverse of the sample size formula
- MDE for 1–8 weeks
- Runs locally
What can this test detect?
The MDE is solved numerically from the same two-proportion formula the sample size calculator uses, so feeding the result back into that calculator returns your sample.
The control’s current rate. Enter 4 for 4%.
Daily traffic and test length, or the sample each group will get.
Eligible visitors per day before the split. 12k and 1,200 both work.
100 if every eligible visitor is in the test.
Whole weeks are recommended: 14, 21, 28…
The sample each group, control included, will reach.
Control plus variants. 2 for a classic A/B test.
The false-positive rate you accept. 5% is the usual choice.
The chance of detecting an effect of exactly the MDE. 80% is common.
Relative MDE
+9.7%
Smallest lift relative to the baseline.
Absolute MDE (points)
0.39
The same lift in percentage points.
Rate the variant must reach
4.39%
Baseline + absolute MDE.
Visitors per group
42,000
Daily test traffic × days ÷ groups.
MDE by test length
Longer tests see smaller effects, but slowly: four times the sample only halves the MDE.
| Test length | Per group | Relative MDE | Absolute | Variant rate |
|---|---|---|---|---|
| 1 wk | 10,500 | +19.8% | 0.79 pp | 4.79% |
| 2 wk | 21,000 | +13.8% | 0.55 pp | 4.55% |
| 3 wk | 31,500 | +11.2% | 0.45 pp | 4.45% |
| 4 wk | 42,000 | +9.7% | 0.39 pp | 4.39% |
| 6 wk | 63,000 | +7.9% | 0.32 pp | 4.32% |
| 8 wk | 84,000 | +6.8% | 0.27 pp | 4.27% |
MDE by sample size
A quarter, half, double and four times your sample. Your row is highlighted.
Share these inputs: the values go after the # in the link, so they never reach a server.
https://tapcub.com/tools/minimum-detectable-effect-calculator
Version v1.0Updated Update history
What the numbers mean
The MDE is a property of the test design, not of the result. Work it out before launch and compare it with the lift you realistically expect.
Relative MDE
The smallest lift, as a share of the baseline, that the test detects with the chosen power. With 42,000 visitors per group on a 4% baseline, that is +9.7%.
Absolute MDE
The same lift in percentage points: 0.39 points, so the variant needs about 4.39% for the test to pick it up reliably.
The length table
MDE shrinks with the square root of the sample. One week detects +19.8%, four weeks +9.7%, eight weeks +6.8%: each doubling of time buys less.
When the MDE is too big
If the MDE is larger than any lift the change could plausibly cause, the test is likely to end "not significant" even if the change works. Redesign it before you run it.
Power is the probability of detecting an effect exactly the size of the MDE. Bigger real effects are detected more often; smaller ones less often, not never.
Check a test before you commit to it
Use it when the calendar or the traffic is fixed and you need to know whether the test is worth running.
-
1
Enter the baseline
The current conversion rate of the step you will test.
-
2
Describe the traffic
Either daily visitors, the share in the test and the number of days, or the visitors each group will get.
-
3
Keep α and power standard
5% and 80% are the usual defaults. Raising power to 90% raises the MDE for the same sample.
-
4
Compare with your expectation
If the change could plausibly lift conversion by 3% but the MDE is 10%, extend the test or test a bolder change.
Solving the sample size formula for the effect
The sample size formula for a two-sided, two-proportion z-test gives n from p₁ and p₂. There is no closed form for p₂ given n, so the calculator solves it numerically (bisection on the relative lift until the formula returns your sample).
- Visitors per group = ⌊ daily visitors × traffic share × days ÷ groups ⌋
- Only increases are solved. Detecting a drop of the same size needs a slightly different sample, because the variance p(1 − p) changes with the rate
- The significance level is not adjusted for multiple variants; with several groups, consider the 1% level
- Same assumptions as the sample size calculator: fixed sample, read once at the end, independent visitors
Sample formula
n(p₂) = ( z₁₋α/₂ √(2p̄(1 − p̄)) + z₁₋β √(p₁(1 − p₁) + p₂(1 − p₂)) )² ÷ (p₂ − p₁)²
Decreases as p₂ moves away from p₁
Solve for p₂
find p₂ > p₁ such that n(p₂) = n
n = 3,000 × 28 ÷ 2 = 42,000 → p₂ = 4.39%
Report
relative MDE = p₂ ÷ p₁ − 1; absolute MDE = p₂ − p₁
+9.7% relative, 0.39 points absolute
What 1 to 8 weeks can detect
Baseline 4%, 3,000 visitors a day, all of them in a two-group test, α = 5%, 80% power.
| Test length | Visitors per group | Relative MDE | Absolute MDE | Variant rate |
|---|---|---|---|---|
| 1 week | 10,500 | +19.8% | 0.79 pts | 4.79% |
| 2 weeks | 21,000 | +13.8% | 0.55 pts | 4.55% |
| 4 weeks | 42,000 | +9.7% | 0.39 pts | 4.39% |
| 6 weeks | 63,000 | +7.9% | 0.32 pts | 4.32% |
| 8 weeks | 84,000 | +6.8% | 0.27 pts | 4.27% |
Going from 4 to 8 weeks only moves the MDE from 9.7% to 6.8%. Calculated with this tool.
How the MDE gets misread
The MDE is the most misunderstood input in test planning.
Reading a relative MDE as absolute
A 10% MDE on a 4% baseline means 4.0% → 4.4%, not 4% → 14%. Check which one a tool reports before you compare.
Treating the MDE as the expected effect
The MDE is what the test can see, not what the change will do. A test with a 10% MDE can still return a real 3% lift — usually as "not significant".
Calling a null result proof of no effect
A non-significant result from a test whose MDE was 20% says only that the effect is probably below about 20%, not that it is zero.
Counting traffic that never sees the change
Visitors who never reach the tested element dilute the effect. Use the traffic of the page or step where the change lives.
Where the baseline and traffic come from
TapCub does not split traffic or run experiments — your testing tool does that. TapCub supplies the baseline and the daily traffic, and can show conversion per group.
Baseline from funnels
Conversion between the steps you will test, over a fixed time window.
Learn moreDaily traffic per page
Visitors per page and per day, with bots counted separately.
Learn moreBreak down by variant
Send the group as an event or user property, then break funnels and trends down by it.
Learn moreMinimum detectable effect questions
Still have a question?
Chat with the team behind TapCub — we usually reply within a few hours on business days.
What is the minimum detectable effect (MDE)?
The smallest true difference that a test is designed to detect with a given power, usually 80%, at a given significance level. Effects smaller than the MDE can still exist; the test is just likely to miss them.
How do I calculate the MDE?
Take the sample size formula for two proportions and solve it for the variant rate given your sample. There is no closed form, so it is solved numerically. With 42,000 visitors per group and a 4% baseline, the MDE is +9.7% relative, or 0.39 points.
What is a good MDE for an A/B test?
One that is smaller than the lift the change could realistically produce. A redesigned checkout might plausibly move conversion by 10%; a button colour rarely does. If the MDE is far above what you expect, the test is not worth running as designed.
Relative or absolute MDE?
Both describe the same effect. Relative is the lift as a share of the baseline (+10%), absolute is the difference in points (+0.4). The calculator shows both.
How do I lower the MDE?
More visitors per group: run longer, put more traffic into the test, or use fewer groups. Testing a step with a higher baseline rate also helps, because the same relative lift is easier to see on 20% than on 1%.
Why does doubling the test length not halve the MDE?
The MDE shrinks roughly with the square root of the sample. Doubling the sample divides it by about 1.4; you need four times the sample to halve it.
How does power affect the MDE?
Higher power means a larger MDE for the same sample. In the default example, raising power from 80% to 90% moves the MDE from +9.7% to +11.2%.
Update history
New features, fixes and data refreshes for this tool, newest first.
-
· v1.0
LaunchedFirst release.
Missing something or found an error? Tell us — requests that come up often go into the next update. All tool updates
Turn the MDE into a plan
Once the MDE is acceptable, fix the sample and the calendar, and check the split when the test ends.