Bonferroni Correction Calculator
Find the corrected significance level when you run several tests at once, or paste your p-values and see which results survive the Bonferroni, Šidák, and Holm corrections.
Bonferroni threshold = α ÷ number of tests
—
—
—
—
—
—
—
—
—
—
—
—
—
—
—
Chance of a False Positive
—
P-values Against the Thresholds
—
How it was calculated
Summary
Notes on this result
—
How to Use This Calculator
- Pick Find the corrected α if you only need the new threshold, or Adjust my p-values if you already have the results of your tests.
- Choose the method. Bonferroni divides α by the number of tests. Šidák is slightly less strict and assumes the tests are independent. Holm (p-value mode only) is never less powerful than Bonferroni.
- Enter the number of tests and your significance level as a percentage, usually 5. In p-value mode, paste the p-values instead. The number of tests is counted for you, or you can enter a larger number if some tests were run but not listed.
- Press Calculate. A test counts as significant only if its p-value is below the corrected threshold, or its adjusted p-value is below α.
- Decide the number of tests before looking at the results. Every comparison you considered counts, including the ones you did not report.
What Is the Bonferroni Correction?
A Bonferroni correction calculator adjusts statistical significance thresholds to control the family-wise error rate when performing multiple hypothesis tests. The Bonferroni correction formula is α/m, where α represents the original significance level and m represents the number of tests. For 10 tests at α = 0.05, the adjusted significance threshold is 0.005.
Each test carries its own chance of a Type I error, and those chances add up. The correction tightens the threshold for each individual test so that the chance of at least one false positive across the whole family stays at or below your chosen α.
The chance of at least one false positive is the family-wise error rate (FWER). With 20 independent tests at α = 0.05 and no correction, the chance of at least one false positive is about 64%. Dividing α by 20 brings it back to roughly 5%. The price is lower power: each single test now needs stronger evidence to count.
Problem
More tests mean more chances of a false positive.
Fix
Test each hypothesis at α ÷ m instead of α.
Cost
Less power, so real effects are easier to miss.
Definition reference: Wikipedia — Bonferroni correction.
Bonferroni Correction Formulas
Bonferroni threshold
αcorrected = α ÷ m
m is the number of tests. Reject a null hypothesis only if its p-value is below this value.
Šidák threshold
αcorrected = 1 − (1 − α)1/m
Exact for independent tests. Always slightly larger than α ÷ m.
Family-wise error rate (no correction)
FWER = 1 − (1 − α)m
The chance of at least one false positive across m independent true null hypotheses.
Adjusted p-values
Bonferroni: min(1, m · p)
Šidák: 1 − (1 − p)m
Compare the adjusted p-value with the original α. The decision is the same as comparing p with the corrected threshold.
Holm–Bonferroni (step-down)
Sort p(1) ≤ … ≤ p(m). Reject p(i) while p(i) < α ÷ (m − i + 1); stop at the first failure.
The smallest p-value faces the same bar as Bonferroni, and each later one faces an easier bar. It controls the same FWER and never rejects fewer hypotheses.
Method explained: BMJ — Bland & Altman, Multiple significance tests: the Bonferroni method.
Worked Example 1: Four Tests at α = 0.05
A study compares four outcomes between a treatment and a control group. What threshold should each test use?
- Bonferroni: 0.05 ÷ 4 = 0.0125.
- Šidák: 1 − (1 − 0.05)1/4 = 1 − 0.98726 = 0.01274.
- Without correction, the chance of at least one false positive across the four tests is 1 − 0.954 = 18.55%.
- A p-value of 0.02 would count as significant on its own, but not after correction.
Worked Example 2: Ten Tests with Šidák
With 10 tests at α = 0.05, the Šidák threshold is 1 − 0.951/10 = 0.005116, a little above the Bonferroni value of 0.005. The gap is small because both are close to α ÷ m when α is small.
Worked Example 3: Adjusting a List of P-values
Six tests give p = 0.001, 0.004, 0.012, 0.018, 0.027, and 0.048. All six are below 0.05 without correction.
- Bonferroni: threshold 0.05 ÷ 6 = 0.00833, so two results (0.001 and 0.004) stay significant. Adjusted p-values are 0.006, 0.024, 0.072, 0.108, 0.162, and 0.288.
- Šidák: threshold 0.00851, also two results.
- Holm: 0.001 < 0.05/6, 0.004 < 0.05/5, 0.012 < 0.05/4, but 0.018 is not below 0.05/3 = 0.0167, so the procedure stops. Three results stay significant.
Corrected Thresholds by Number of Tests
At α = 0.05. The last column is the chance of at least one false positive if you do not correct, assuming independent tests and no real effects.
| Tests (m) | Bonferroni | Šidák | Risk without correction |
|---|---|---|---|
| 1 | 0.05000 | 0.05000 | 5.00% |
| 2 | 0.02500 | 0.02532 | 9.75% |
| 3 | 0.01667 | 0.01695 | 14.26% |
| 4 | 0.01250 | 0.01274 | 18.55% |
| 5 | 0.01000 | 0.01021 | 22.62% |
| 10 | 0.00500 | 0.00512 | 40.13% |
| 20 | 0.00250 | 0.00256 | 64.15% |
| 50 | 0.00100 | 0.00103 | 92.31% |
| 100 | 0.00050 | 0.00051 | 99.41% |
The risk climbs fast: by 14 tests it passes 50%, and by 100 tests a false positive is almost certain if you do not correct.
Error-rate definition: Wikipedia — Family-wise error rate.
Bonferroni, Šidák, Holm, and Benjamini–Hochberg
The four most common ways to handle multiple comparisons.
| Method | Controls | Assumption | Power |
|---|---|---|---|
| Bonferroni | Family-wise error rate | None, any dependence | Lowest |
| Šidák | Family-wise error rate | Independent tests (it also holds under some positive dependence) | Slightly above Bonferroni |
| Holm | Family-wise error rate | None, any dependence | Never below Bonferroni |
| Benjamini–Hochberg | False discovery rate | Independent or positively dependent tests | Highest, but a different guarantee |
Use a family-wise method when even one false positive is costly, such as a confirmatory clinical trial. Use false-discovery-rate control when you screen thousands of candidates, such as genes, and can accept a small share of false leads. This calculator covers the first three.
Frequently Asked Questions
What is the Bonferroni correction?
The Bonferroni correction is a method for multiple comparisons, also called multiple testing, that lowers the significance level of each hypothesis test so that the family-wise error rate across all tests stays at or below α. Each test is run at an adjusted alpha, which is α divided by the number of tests. It guards against a Type I error, or false positive, because the probability of at least one false positive grows with every extra test on the same dataset.
How do I calculate the Bonferroni correction?
Count the number of tests, choose α (usually 0.05), and divide: corrected α = α ÷ m. For 4 tests, 0.05 ÷ 4 = 0.0125, and a result is significant only if its p-value is below that threshold. Alternatively, multiply each p-value by m, cap it at 1, and compare the adjusted p-value with the original α. The decision is the same either way, and a Bonferroni correction calculator like this one shows both, along with the critical value that matches the corrected threshold.
What is the Šidák correction for 10 tests at p = 0.05?
About 0.0051. The formula is 1 − (1 − 0.05)1/10 = 0.005116. The Bonferroni value for the same case is exactly 0.005, so the Šidák threshold is slightly more lenient. Šidák is an alternative correction method that holds the error rate at α exactly when the statistical tests are independent.
When should I use the Bonferroni correction?
Use it when you run several hypothesis tests on the same data and want to avoid claiming statistical significance for a chance result. It works best with a small number of planned comparisons. It is common for a post hoc test after an ANOVA, where each pairwise comparison between group means is checked, as well as for subgroup analyses and for several outcomes in a clinical trial or other research study.
Is Bonferroni the only method for multiple comparisons?
No. Šidák and Holm–Bonferroni control the same family-wise error rate with a little more power. Benjamini–Hochberg controls the false discovery rate instead, which suits large screens. Tukey's test and Dunnett's test are built for comparing group means, and Tukey's test takes its critical value from the studentized range distribution. Whichever method you pick, say so in your report so readers can judge the inference.
Is the Bonferroni correction too conservative?
It can be. With many tests, or tests that are strongly correlated, the threshold becomes very small, and real effects, where the alternative hypothesis is true, are missed more often. Holm gives more power with the same guarantee against rejecting a true null hypothesis, and false-discovery-rate methods give more still if you can accept a few false leads. Planning a larger sample can offset the loss of power.
Does the correction also apply to confidence intervals?
Yes. To make sure all m intervals contain their population values at the same time at a 95% confidence level, build each confidence interval at a level of 1 − α ÷ m. For 4 intervals that is 98.75%, so each interval is wider than a single 95% interval.
Which tests count towards the number of tests?
Every test in the family you are drawing conclusions from, including ones whose results you do not report. Define the family before you see the data from the experiment. Counting only the tests that looked interesting makes the correction too weak and lets the real error rate creep above α.
Other Useful Calculators
Power Analysis Calculator
Sample size, power, and detectable effect.
Correlation Coefficient Calculator
Pearson, Spearman, and Kendall with a p-value.
Empirical Rule Calculator
The 68-95-99.7 rule for normal data.
Raw Score Calculator
Convert a z-score or percentile back to a raw score.
Poisson Distribution Calculator
Probability of a number of events in an interval.
Quadratic Regression Calculator
Fit a parabola to curved data.