Course contentsShow
← Data Science
Lesson 458 of 2,14512. ANOVA and Multiple ComparisonsFree lesson

Bonferroni Correction

A conservative approach that adjusts the significance level by dividing alpha by the number of comparisons.

Bonferroni Correction

You'll learn: How to control the risk of false positives when performing multiple statistical comparisons.

The Multiple Comparisons Problem Revisited

In the previous lesson, you learned that running multiple statistical tests inflates your risk of making a Type I error (false positive). If you test 20 comparisons at α = 0.05, you'd expect at least one "significant" result by chance alone, even if nothing is truly different.

What is the Bonferroni Correction?

The Bonferroni Correction is a conservative, straightforward method to control this risk. The idea is simple: divide your significance level (α) by the number of comparisons you're making.

If you're conducting m comparisons and want to maintain an overall significance level of α = 0.05, you test each individual comparison at:

α_adjusted = α / m

For example, if you're comparing 5 groups (requiring 10 pairwise comparisons), you'd use:

α_adjusted = 0.05 / 10 = 0.005

Now, a comparison is only considered significant if its p-value is less than 0.005 instead of 0.05.

The Trade-Off

Analogy: Think of your α budget like money. You have $0.05 to spend on risk. The Bonferroni method splits that budget equally among all your comparisons, making each test much more stringent.

This approach is conservative—it effectively protects against false positives but increases the risk of false negatives (missing real differences), especially when you have many comparisons.

When to Use It

The Bonferroni Correction works well when:

  • You have a small to moderate number of comparisons
  • False positives are particularly costly in your context
  • You want a simple, easy-to-explain adjustment

Key Takeaway: The Bonferroni Correction divides your significance level by the number of comparisons to control family-wise error rate, providing strong but conservative protection against false positives.