MDE Calculator (Minimum Detectable Effect)

This calculator answers the opposite question to sample size. Instead of starting from the effect you want to detect, it starts from the traffic you already have and returns the smallest lift you can prove. It is the reality check before you switch a test on: if your MDE is +12% and the change you made usually returns +3%, the test will end inconclusive for lack of sensitivity, and it is better to know now. When the effect is the goal and traffic is the answer, use the sample size calculator.

How to use it

1. Enter your current conversion rate (the control's, in %).
2. Put in the visitors per week (the test total) and the duration in weeks you plan to run.
3. Pick the number of variations: more variations means less sample per arm and a higher MDE.
4. Leave confidence 95% and power 80% (the defaults) or adjust if you know what you are doing.
5. Read the result: relative MDE, absolute MDE (in points) and the target rate the variation must reach.

How it works: the formula

The calculator inverts the sample size equation. Starting from the available N per variation, it isolates the smallest detectable effect via the two-proportion normal approximation, using the pooled variance (p2 close to p1):

Where δ is the minimum effect in points, p is the baseline rate, n is the sample per variation, z_α is the confidence critical value (1.96 for 95% two-sided) and z_β is the power one (0.84 for 80%). The relative MDE is δ divided by p.

Worked example (reproduces the default result)

With the values already filled in: baseline rate 5%, 10,000 visitors per week, 4 weeks, 2 variations, confidence 95% and power 80%, two-sided. Total traffic is 10,000 × 4 = 40,000, divided by 2 variations gives 20,000 visitors per variation. Plugging into the formula:

δ = (1.96 + 0.84) · √(2 · 0.05 · 0.95 / 20,000) = 2.80 · √(0.00000475) = 2.80 · 0.0021794 = 0.006106.

That is 0.61 percentage points (the absolute MDE). In relative terms, 0.006106 ÷ 0.05 = +12.2%. The target rate to beat is 5% + 0.61 pp = 5.61%. This is exactly what the tool shows above when you open the page.

How to read it and where it misleads

The MDE is the smallest lift worth chasing with your traffic. Smaller effects exist, but the test lacks the power to separate them from noise within the deadline, so you finish with no verdict. That makes the MDE a business decision, not just a statistical one: if your MDE is +12% and most of your optimizations return less, you either gather more traffic or test bolder changes.

Limits to keep in mind: the math assumes stable traffic and a binary metric (converted or not). It does not cover revenue per visitor (higher variance, higher MDE), strong seasonality, or novelty effects. The pooled approximation is the market convention for MDE and slightly overestimates on very large effects, but across the useful decision range it agrees with the sample size calculator. After running, check the result with the significance calculator and read the guide on A/B testing statistical significance.

Good practice when sizing by MDE

The MDE is the sanity check you run before investing weeks of traffic. Treat it as the first step of planning.

• Compare the MDE with the lift your change usually returns. If the MDE is larger, the test is born to end in a tie.
• Fewer variations concentrate the sample and lower the MDE. Only open an A/B/n test if the traffic can fund every arm.
• Run full weekly cycles so you do not mix the MDE with day-of-week noise.
• If the MDE only drops to something acceptable over many months, raise the target effect or gather more traffic before starting.

Frequently asked questions

What is minimum detectable effect (MDE)?

It is the smallest conversion lift your test can actually prove with the traffic you have. If the MDE is +12% relative, a smaller improvement will likely go undetected: the test lacks the sensitivity to tell it apart from noise. The MDE is the floor of your ruler, and it drops as you accumulate more visitors.

How is this different from a sample size calculator?

They are the two ends of the same equation. The sample size calculator starts from the effect you want to detect and returns how many visitors you need. This one starts from the traffic you already have (visitors per week and duration) and returns the smallest effect it can detect. Use this when traffic is fixed and you want to know if a test is even worth it; use the other when the effect is the goal.

Relative or absolute MDE: which one do I read?

Both say the same thing in different units. The relative one (+12.2%) is the percentage change over your current rate, easier to compare across pages. The absolute one (0.61 points) is the difference in percentage points, useful when the baseline rate is very low. The target rate to beat ties them together: it is the conversion your variation must reach for the test to call a win.

My MDE is too high. What do I do?

A high MDE means your traffic can only prove large wins. Three ways out: accumulate more traffic (extend the duration or gather more visitors), reduce the number of variations to concentrate the sample, or accept that only big-swing changes are worth testing. Chasing a +2% lift with traffic that only detects +12% burns weeks to conclude nothing.

Why confidence 95% and power 80%?

They are the market conventions. Confidence 95% accepts a 5% chance of a false positive; power 80% gives an 80% chance of detecting the effect if it exists. Raising either one makes the test stricter and, on the same traffic, raises the MDE (you can now only detect larger effects).

Keep going

Know the smallest effect you can detect? If it fits your goal, size the test with the sample size calculator and plan the calendar with the A/B test duration calculator. When it ends, read the verdict with the significance calculator. The full context is in the guide what is A/B testing.