Hotelling's T² Test
Also known as: Hotelling T² Testi — Çok Değişkenli t-Testi, multivariate t-test, Hotelling T-squared
Hotelling's T² test is a multivariate parametric hypothesis test that simultaneously compares the mean vectors of two independent groups across multiple continuous outcome variables. It was introduced by Harold Hotelling in 1931 as the direct multivariate generalization of Student's t-test, replacing the scalar mean difference with a vector difference scaled by the pooled variance-covariance matrix.
Key highlights
- Controls the family-wise Type I error rate when testing multiple correlated outcomes simultaneously, unlike running separate t-tests.
- Exploits the correlational structure among outcomes — if variables are correlated, T² can be more powerful than Bonferroni-corrected univariate tests.
- Converts exactly to an F distribution, giving precise rather than approximate p-values.
- Directly generalises Student's t-test, making it conceptually intuitive for researchers already familiar with the univariate case.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use Hotelling's T² when you want to compare exactly two independent groups on two or more correlated continuous outcome variables simultaneously. Three assumptions must hold: multivariate normality of the outcome variables within each group (check with Mardia's test or Royston's multivariate extension of Shapiro-Wilk), homogeneity of the variance-covariance matrices across groups (check with Box's M test, interpreted cautiously because it is sensitive to non-normality), and independence of observations both within and between groups. A minimum total sample of around 30 is advisable; the test becomes unreliable when sample size is smaller than the number of outcome variables. When three or more groups are involved, use MANOVA instead.
Strengths & limitations
- Controls the family-wise Type I error rate when testing multiple correlated outcomes simultaneously, unlike running separate t-tests.
- Exploits the correlational structure among outcomes — if variables are correlated, T² can be more powerful than Bonferroni-corrected univariate tests.
- Converts exactly to an F distribution, giving precise rather than approximate p-values.
- Directly generalises Student's t-test, making it conceptually intuitive for researchers already familiar with the univariate case.
- Assumes multivariate normality, which is harder to verify and achieve than univariate normality, especially in small samples.
- The homogeneity of covariance matrices assumption (Box's M) is often violated in practice and Box's M test itself is sensitive to non-normality.
- Restricted to exactly two groups — three or more groups require MANOVA.
- Statistical power decreases as the number of outcome variables grows relative to sample size; including weakly discriminating variables can reduce power.
- A significant omnibus T² does not identify which individual variables contribute to the group difference.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How is Hotelling's T² different from running separate t-tests on each outcome?
Running k separate t-tests inflates the family-wise Type I error rate — for example, five tests each at α = 0.05 give roughly a 23% chance of at least one false positive under the null. Hotelling's T² tests all outcomes simultaneously at a single α level, controlling that error rate. It also accounts for the correlations among variables, which separate tests ignore.
What is the relationship between Hotelling's T² and MANOVA?
MANOVA is the multivariate generalisation for three or more groups. Hotelling's T² is mathematically identical to the two-group special case of MANOVA: the MANOVA F statistic equals the T²-derived F when there are exactly two groups. For two groups, all four MANOVA test criteria (Wilks' Λ, Pillai's trace, Hotelling-Lawley trace, Roy's largest root) yield the same F value.
My T² is significant — how do I find out which variables are responsible?
The omnibus T² only signals that the mean vectors differ overall. To identify contributing variables, perform follow-up univariate t-tests with a Bonferroni or Holm correction, or use discriminant function analysis to find the linear combination of variables that best separates the groups. Report both approaches together for a complete picture.
What should I do if Box's M test is significant?
Box's M tests equality of covariance matrices, but it is highly sensitive to non-normality and is often significant even when the practical violation is minor. As a rule of thumb, if the covariance matrices look broadly similar in magnitude and the groups have similar sizes, the T² result is relatively robust. If matrices differ substantially, consult robust multivariate procedures or consider reporting results under both homogeneous and heterogeneous covariance assumptions.
Sources
- 1.Hotelling, H. (1931). The Generalization of Student's Ratio. Annals of Mathematical Statistics, 2(3), 360–378.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). Hotelling's T² Test. ScholarGate. https://scholargate.app/statistics/hotelling-t2