MerQuR+Statistics Glossary

Statistics Glossary → Hotelling's T²

Hotelling's T²concept

The multivariate generalization of the t test; a statistic that jointly tests whether a mean vector equals a given vector, or whether two groups' mean vectors are equal.

The multivariate counterpart of the univariate t² = n(x̄ - μ₀)²/s² is T² = n(x̄ - μ₀)'S⁻¹(x̄ - μ₀), where S is the sample covariance matrix. T² sums the differences in means into a single number while accounting for the variances of the variables and the correlations between them. With suitable scaling it becomes an F statistic: for p variables and n observations, (n - p)/(p(n - 1))·T² ~ F(p, n - p).

Running a separate t test for each variable both inflates the Type I error rate and ignores the relationships between variables; T² gives one overall test. Applied to the difference scores of k conditions in repeated measures, it does not require sphericity; it is the two-group or one-sample special case of MANOVA. It assumes multivariate normality and, with two groups, equal covariance matrices, and it becomes unreliable as the number of variables approaches the number of observations.

An everyday example

A factory checks both the fill volume and the cap tightness of its bottles at the same time. Each measurement on its own looks close to target, but taken together it turns out that two measurements that normally rise together have drifted in opposite directions in this batch. Hotelling's T² captures this joint departure in a single test.

In MerQur

In repeated measures ANOVA the 'Multivariate tests' checkbox applies Hotelling's T² to the difference scores; with one within-subject factor the four tests give the same exact F and do not require sphericity. In the simple-effects output of repeated measures and mixed-design ANOVA the within-subject effect is tested with the Hotelling T² exact F; the multivariate tests of mixed-design ANOVA also include the Hotelling-Lawley trace. In Cronbach's alpha and reliability analysis the 'Additional output' option gives Hotelling's T².

Analyses where this concept appears

Repeated Measures ANOVAMixed-Design ANOVA, Split-PlotCronbach's Alpha and Reliability Analysis

Related concepts

Multivariate test statistics (Wilks' Λ, Pillai's trace, Hotelling-Lawley, Roy)