Statistics Glossary → Šidák correction
Šidák correctionconcept
A multiple-comparison correction that lowers the significance level of each test to 1 − (1 − α)^(1/m) so that the probability of at least one false positive across all tests stays at α.
When m independent tests are run and H₀ is true for all of them, the probability that at least one comes out significant by mistake is 1 − (1 − α)^m. Šidák inverts this expression to find the threshold for each test; equivalently, the adjusted p value is 1 − (1 − p)^m. For α = 0.05 and m = 5, for instance, the per-test threshold is 0.0102, compared with 0.05/5 = 0.0100 for Bonferroni.
The Šidák threshold is always slightly larger than Bonferroni's, so it is slightly less conservative, but the difference is small in practice. Keeping the familywise error rate exactly at α depends on the tests being independent; it generally remains valid when the tests are positively correlated. The step-down Holm-Šidák version provides the same guarantee with more power. Both Šidák and Bonferroni lose power quickly as the number of tests grows; in exploratory work, false discovery rate (FDR) methods may be more suitable.
An everyday example
A hospital tests separately whether a new drug increases each of 6 different side effects. If each test is run at the 5% level, there is about a 26% chance of at least one 'significant' result even if the drug increases none of them. With the Šidák correction, each test's threshold drops to 0.0085 and the overall probability is brought back to 5%.
In MerQur
It is offered in Multiple Comparison Correction alongside methods such as Bonferroni, Holm and Holm-Šidák. It is also one of the p-value adjustment options for pairwise comparisons in one-way and two-way ANOVA, repeated measures and mixed-design ANOVA, ANCOVA, the Kruskal-Wallis and Friedman tests, linear mixed models and GEE.
Analyses where this concept appears
One-Way ANOVATwo-Way ANOVARepeated Measures ANOVAMixed-Design ANOVA, Split-PlotANCOVAMultiple Comparison CorrectionKruskal-WallisFriedmanLinear 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)
Related concepts
Bonferroni correctionMultiple comparisons and the familywise error rate (FWER)