Math calculators

Statistical Significance Calculator

Updated Aug 29, 2026 By Jehan Wadia
Rate Formulas
A/B Test Data
Enter conversions and visitors for Group A and Group B. Conversion rate is calculated automatically.
Group Conversions Visitors Conversion Rate
Group A
Group B
Test Settings
Confidence Level
Test Type
Expected Traffic Split
Group B expected share: 50%
Results
The difference is not statistically significant.

P-Value
Z-Score
Relative Uplift
Observed Difference
Min. Required Difference

Confidence Meter
Achieved confidence: Required threshold: 95%
0%50%100%
Conversion Rate Comparison
Test Statistic vs. Critical Value
Step-by-Step Solution

Introduction

This Statistical Significance Calculator tells you if your A/B test result is real or just luck. Type in how many people saw each version and how many of them converted. The tool does the math for you right away.

It uses a two-proportion z-test, the standard way to compare two conversion rates. You get a clear yes or no answer, plus the p-value, z-score, uplift, and the smallest gap you would need to call a winner. Charts and a step-by-step breakdown show you how the numbers were found. If you want to dig into a single statistic on its own, try our p Value Calculator or Z Score Calculator.

You can pick your confidence level (80%, 90%, 95%, or 99%) and choose a one-tailed or two-tailed test. The calculator also runs a sample ratio mismatch (SRM) check to warn you if traffic did not split the way you planned, which often means a setup or tracking bug.

Use it to test landing pages, emails, ads, buttons, or headlines, and stop guessing which version really works better. Pair it with our Conversion Rate Calculator and CTR Calculator to track the metrics behind each test.

How to use our Statistical Significance Calculator

Enter the conversions and visitors for both test groups, pick your test settings, and the calculator shows your p-value, z-score, uplift, and whether your A/B test result is statistically significant.

Group A Conversions: Type how many people in your control group did what you wanted, like buying or signing up. This number cannot be bigger than your Group A visitors.

Group A Visitors: Type how many people saw your control version. The tool uses this to work out the Group A conversion rate for you.

Group B Conversions: Type how many people in your variation group converted. Again, keep it equal to or lower than the Group B visitors.

Group B Visitors: Type how many people saw your variation. The Group B conversion rate fills in on its own.

Confidence Level: Pick 80%, 90%, 95%, or 99%. This is how sure you want to be before you call a winner. Most A/B tests use 95%. To turn that level into a range around your result, use the Confidence Interval Calculator.

Test Type: Choose Two-Tailed to check for any difference between the groups, or One-Tailed to check only if one group is better. One-tailed tests reach significance faster. Our Critical Value Calculator shows the cutoff z-scores behind each option.

Expected Traffic Split: Enter the percent of traffic you planned to send to Group A. Group B's share fills in on its own, and the tool runs a sample ratio mismatch (SRM) check to spot split errors in your test setup.

Click Calculate to see your results, charts, and the step-by-step math. Use Reset to load the sample numbers again, or Clear Fields to start fresh.

What Is Statistical Significance?

Statistical significance tells you if a result is real or just luck. In an A/B test, you show two versions of a page. Group A is the control. Group B is the variation. One version may get more sign-ups or sales. But small gaps happen by chance all the time. A significance test checks if the gap is big enough to trust.

How the Test Works

This tool uses a two-proportion z-test. It compares the conversion rate of each group. The conversion rate is conversions divided by visitors. The test then looks at how far apart the two rates are, and how much data you have. More visitors means less noise and a clearer answer. The z-test leans on the bell curve, which you can explore in the Normal Distribution Calculator.

Key Terms

  • P-value: The chance of seeing a gap this big if both versions were really the same. A small p-value means strong proof.
  • Z-score: How many standard errors apart the two rates are. Bigger numbers mean a stronger signal. See how the spread is built with our Standard Error Calculator and Standard Deviation Calculator.
  • Confidence level: How sure you want to be. At 95%, you accept a 5% chance of a false alarm.
  • Relative uplift: How much better the winner is compared to the loser, as a percent. The Percent Change Calculator works the same way on any two numbers.
  • Percentage points (pp): The plain gap between two rates. 51% minus 49% is 2 pp. Our Percent Difference Calculator helps keep points and percents straight.

One-Tailed or Two-Tailed

A two-tailed test asks, "Is there any difference?" It can catch a win or a loss. A one-tailed test only asks, "Is this version better?" One-tailed tests hit significance sooner, but they miss drops. Most teams should pick two-tailed.

Sample Ratio Mismatch (SRM)

If you planned a 50/50 split, each group should get close to half the traffic. When the real split is far off, something is broken. It could be a tracking bug, a redirect, or bad bot filtering. The SRM check uses a chi-square test to spot this, the same math you will find in our Chi Square Calculator. If it fails, fix your setup first. The test results cannot be trusted.

Tips for Better A/B Tests

  • Pick your sample size and confidence level before you start. Our Sample Size Calculator and Margin of Error Calculator help you plan ahead.
  • Let the test run for full weeks. Traffic changes by day.
  • Do not stop early just because a result looks good. That causes false wins.
  • "Not significant" does not mean "no difference." It may mean you need more visitors.
  • A tiny win can be significant with huge traffic. Ask if it is worth the work by checking the Effect Size Calculator and running the numbers through a ROI Calculator.

Testing means instead of rates? Use the t Test Calculator for two groups, the ANOVA Calculator for three or more, and the Statistics Calculator for a quick summary of any data set.


Formulas used

Conversion rate for each group
p_A = \frac{c_A}{n_A}, \quad p_B = \frac{c_B}{n_B}
Pooled conversion proportion
\hat{p} = \frac{c_A + c_B}{n_A + n_B}
Standard error of the difference in proportions
SE = \sqrt{\hat{p}\,(1-\hat{p})\left(\frac{1}{n_A}+\frac{1}{n_B}\right)}
Two-proportion z-statistic
z = \frac{p_A - p_B}{SE}
P-value (two-tailed and one-tailed)
p_{\text{two}} = 2\left[1-\Phi(|z|)\right], \quad p_{\text{one}} = 1-\Phi(|z|)
Significance level and decision rule
\alpha = 1 - \frac{C}{100}, \quad \text{significant if } p < \alpha
Minimum difference required for significance
\Delta_{min} = z_{crit} \times SE
Relative uplift of the better group
\text{Uplift} = \frac{p_{\max} - p_{\min}}{p_{\min}} \times 100\%

Frequently asked questions

What p-value counts as significant?

It depends on the confidence level you pick. The rule is that the p-value must be smaller than alpha, where alpha equals 1 minus your confidence level.

  • 95% confidence: p must be under 0.05
  • 90% confidence: p must be under 0.10
  • 99% confidence: p must be under 0.01
  • 80% confidence: p must be under 0.20

The calculator does this check for you and shows a green or orange verdict box.

Why does my p-value show "< 0.0001"?

That means the p-value is smaller than one in ten thousand. The gap between your two groups is very unlikely to be luck. The tool shows it this way to keep the number easy to read. You can still see the exact math in the step-by-step section.

My result is not significant. What should I do?

Keep the test running and collect more visitors. Look at the Minimum Required Difference box. If your observed gap is close to that number, more data may push you over the line. If the gap is tiny, the two versions may just be about the same, and you can move on to a bolder test idea.

How many visitors do I need per group?

There is no single number. It depends on your baseline rate and the size of the lift you want to spot. Small lifts need lots of traffic. As a rough guide, many teams aim for at least 1,000 visitors and 100 conversions per group before they trust a result. Plan this before you start the test.

Why is my uplift large but still not significant?

Because uplift ignores sample size. A jump from 2 sales out of 20 to 4 sales out of 20 is a 100% uplift, but it is built on tiny numbers. The z-test looks at both the gap and the amount of data. With few visitors, even a big-looking uplift can be random noise.

What does the Minimum Required Difference tell me?

It is the smallest gap between the two conversion rates that would count as significant with your current visitor counts. It is shown in percentage points. If your observed difference is bigger, you have a winner. If it is smaller, you do not, at least not yet.

Does it matter which version I put in Group A?

The math works either way. Only the sign of the z-score flips. Still, it is best practice to put your control or original version in Group A and your new version in Group B. That keeps your results easy to read and share.

What does a negative z-score mean?

It means Group B had the higher conversion rate. A positive z-score means Group A was higher. For significance, only the size of the number matters, not the sign. The tool tells you which group won in plain words under the verdict.

Can the two groups have different visitor counts?

Yes. The two-proportion z-test handles uneven group sizes just fine. Just make sure the Expected Traffic Split field matches the split you actually planned. If you planned 70/30, enter 70 so the SRM check does not raise a false alarm.

I got an SRM warning but my split looks close. Why?

With a lot of traffic, even a small gap can fail the check. A 51/49 split on 500 visitors is fine, but the same split on 500,000 visitors is a red flag. Common causes are bot traffic, redirect delays, caching, or a tag that fails to fire on one version.

Can I test three or more versions with this tool?

Not at once. This calculator compares two groups. If you have three or more versions, run them two at a time against the control, but know that more comparisons raise the risk of a false win. For many groups at once, a chi-square or ANOVA test fits better.

Can I use this for revenue, order value, or time on page?

No. This tool is built for rates, where each visitor either converts or does not. For averages like revenue per visitor or minutes on page, use a t-test instead. Those numbers need a different formula.

Does a significant result mean Group B will always win?

No. It means the gap is unlikely to be random chance. At 95% confidence, you still accept about a 5% chance of a false alarm. Results can also shift if your audience, season, or offer changes. Big wins are worth a re-test before a full rollout.

Can I check my results while the test is still running?

You can look, but do not stop the test the moment it turns green. Checking again and again and stopping at the first win is called peeking, and it creates fake winners. Set your sample size and end date first, then judge the result at the end.

What is the confidence meter showing?

The filled bar is your achieved confidence, which is (1 − p-value) × 100. The dashed red line is the level you picked. When the bar passes the line, your result is significant and the bar turns green.

Why did my conversion rate box go blank?

The rate only fills in when both boxes hold valid numbers. It clears if visitors are empty or zero, or if conversions are larger than visitors. Fix the red error message under the field and the rate comes right back.

Is this calculator free to use?

Yes. It is free, needs no sign-up, and runs in your browser. Your numbers are not sent anywhere or saved.