Statistics Glossary → Tukey's HSD test
Tukey's HSD testconcept
A multiple-comparison test that, after a significant ANOVA, compares the means of every pair of groups while keeping the familywise Type I error rate at α.
The HSD (Honestly Significant Difference) test compares all k(k - 1)/2 pairs of k groups at once. The difference between two means is divided by a standard error taken from the ANOVA's error mean square, and the result is judged against the studentized range (q) distribution rather than the t distribution. Because that distribution accounts for the gap between the largest and smallest of k means, the probability of at least one false positive across all pairs stays at α.
Tukey's HSD assumes equal variances and the other ANOVA assumptions; with unequal group sizes the Tukey-Kramer adjustment is used. If the variances differ markedly, the Games-Howell test is more appropriate. When all pairs are of interest it is usually more powerful than the Bonferroni correction; when only a few pre-planned comparisons are needed, planned contrasts or Bonferroni may be more suitable. For comparisons against a single control group only, Dunnett's test is the designed tool.
An everyday example
You compare the effect of four fertilisers on tomato yield, and ANOVA shows that they differ. But which fertiliser differs from which? With four fertilisers there are 6 pairs. If you tested each pair separately at the 5% level, your chance of declaring at least one difference when none exists would approach 26%. Tukey's HSD keeps the overall false-alarm risk of all six comparisons at 5%.
In MerQur
Tukey HSD is among the 'Post-hoc test' options in One-Way and Two-Way ANOVA, Tukey-Kramer in ANCOVA, and Tukey is one of the 'Follow-up ANOVA post hoc' options in MANOVA. Tukey HSD is offered for the upper-group comparisons in the Nested Mixed Model and the fixed-factor comparisons in the Crossed Mixed Model; in Linear Mixed Models, GEE and GLMM Tukey can be chosen as the EMMEANS p adjustment.
Analyses where this concept appears
One-Way ANOVATwo-Way ANOVAMANOVAANCOVALinear 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)Generalized Linear Mixed Models (GLMM)
Related concepts
Post hoc tests (pairwise comparisons)Games-Howell testCompact letter display (homogeneous subsets)