본페로니 검정법

The Bonferroni test is a multiple comparison test that identifies which sample means differ significantly from others. It minimizes Type 1 error by reducing the significance level alpha, which otherwise increases with multiple sample pairs. Named after Italian mathematician Carlo Emilio Bonferroni, this test is essential when comparing means across multiple samples simultaneously.

The Bonferroni test divides the original alpha value by the total number of pairwise comparisons to calculate an adjusted alpha. For example, with three sample pairs, the adjusted alpha equals the original alpha divided by three. This adjustment controls the overall Type 1 error rate across all comparisons, making the test more conservative.

First, pair all samples in every possible combination and state null hypotheses assuming equal means. Calculate the t-statistic and P-value for each pair. Compare each P-value against the adjusted alpha value. If P-value is less than adjusted alpha, reject the null hypothesis; otherwise, fail to reject it. Finally, identify which sample pairs have significantly different means.

Compare each pair's P-value to the adjusted alpha threshold. A P-value less than the adjusted alpha indicates the sample pair has significantly different means, so you reject the null hypothesis. A P-value greater than the adjusted alpha suggests the means are not significantly different, so you fail to reject the null hypothesis for that pair.

The Bonferroni test assumes that the means in each sample pair are equal. This null hypothesis is tested for every pairwise comparison in the dataset. If statistical evidence contradicts this assumption for a particular pair, the null hypothesis is rejected, indicating significantly different means between those samples.

Use the Bonferroni test when conducting multiple pairwise comparisons and needing strict control over Type 1 error rates. It is particularly useful after one-way ANOVA with unequal sample sizes, where you need to identify which specific sample means differ. The test's conservative nature makes it ideal for studies requiring high statistical rigor.

The Bonferroni test controls Type 1 error by reducing the significance level alpha proportionally to the number of comparisons. By dividing alpha by the number of pairs, each individual comparison uses a stricter threshold, reducing the cumulative probability of false positives across all comparisons. This conservative approach protects against incorrectly rejecting true null hypotheses.