Statistics Glossary → Bonferroni correction
Bonferroni correctionconcept
A correction that keeps the probability of at least one false significant result below α when m tests are run, by judging each test at α/m (or multiplying the p-values by m).
In 20 independent tests each run at the 5% level, the chance that at least one comes out significant is about 64% even if there is no real effect at all. The Bonferroni correction controls this familywise error rate (FWER): the significance threshold is lowered to α/m or, equivalently, each p-value is multiplied by m (and capped at 1). It is valid whatever the dependence between the tests, which makes it simple and universal.
Its drawback is conservatism: as the number of tests grows, or when the tests are strongly correlated, the chance of missing real effects (Type II error) rises markedly. Holm's step-down method gives the same familywise control and is always at least as powerful as Bonferroni; the Šidák correction is slightly less conservative for independent tests. When many exploratory tests are run, the Benjamini-Hochberg procedure, which controls the false discovery rate (FDR) rather than the familywise error, is often more appropriate.
An everyday example
A researcher tests separately whether each of 10 foods is linked to headaches. Even if none of them actually causes headaches, judging each test at 5% gives about a 40% chance that at least one food will look 'guilty'. With Bonferroni, each test is judged at 0.05 / 10 = 0.005, so chocolate, with p = 0.02, no longer counts as significant.
In MerQur
It is offered under 'Post-hoc test' in one- and two-way ANOVA, ANCOVA and Repeated Measures ANOVA, under 'EMM / simple-effect adjustment' in Mixed-Design ANOVA, and as the EMMEANS p adjustment in the mixed models and Generalized Estimating Equations. It also appears in the Kruskal-Wallis (Dunn) and Friedman post-hoc step, in the column-proportion and cell tests of the chi-square and crosstab analyses, in Correlation Analysis, in the Kaplan-Meier pairwise comparisons and in the local tests of Moran's I and Getis-Ord Gi*; MANOVA gives a Bonferroni p for the F test of each dependent variable. Multiple Comparison Correction applies Bonferroni to raw p-values alongside methods such as Šidák and Holm.
Analyses where this concept appears
One-Way ANOVATwo-Way ANOVARepeated Measures ANOVAMixed-Design ANOVA, Split-PlotMANOVAANCOVAMultiple Comparison CorrectionKruskal-WallisFriedmanChi-Square IndependenceChi-Square Goodness of FitCrosstabMR × Categorical CrosstabMR × MR CrosstabCochran's Q TestCorrelation AnalysisMultiple Linear RegressionLinear Mixed Models (LMM)Nested Mixed Model (Nested LMM, Multi-Level Hierarchical Mixed Model)Crossed Mixed Model (Crossed LMM, Crossed Random Effects Mixed Model)Generalized Estimating Equations (GEE)Kaplan-Meier Survival AnalysisMoran's I (Spatial Autocorrelation)Getis-Ord Gi* Hotspot
Related concepts
Multiple comparisons and the familywise error rate (FWER)Holm (step-down Bonferroni) methodŠidák correction