MANOVA & Multivariate Tests
A therapy trial tracks anxiety, depression and sleep. A memory experiment logs accuracy alongside reaction time. MANOVA (Multivariate ANOVA) tests whether groups differ on a whole set of outcomes at once, and it can detect differences that separate tests on each outcome miss.
Why not just run several ANOVAs?
Separate ANOVAs run into the familiar multiple-testing problem: five outcomes at α = .05 give you roughly a 23% chance of at least one false alarm. They also ignore the correlation structure. Outcomes usually move together, and a pattern that is unremarkable on each variable separately can be very improbable as a combination. MANOVA analyzes the outcomes jointly, so it takes those correlations into account.
🎮 Two Outcomes, One Verdict
Every dot is a person measured on two outcomes. Slide the group differences and the correlation between outcomes, then compare the three verdicts below. The arrow is the discriminant direction, the single best combination of the two outcomes for telling the groups apart. (The same people stay on screen while you drag; "New sample" redraws.)
How it works: the smartest composite
MANOVA effectively asks: is there any weighted combination of the outcomes on which the groups clearly differ? It finds the combination that maximizes the between-group difference relative to the within-group spread (the discriminant direction in the plot) and tests the groups on that. Distance is measured the multivariate way (Mahalanobis distance): not "how far apart are the means?" but "how far apart are they given the shape of the cloud?"
Mahalanobis distance explains what the ✨ button does. When two outcomes are strongly correlated, they normally move together, so a treatment that pushes them in opposite directions is extremely surprising, even if each individual shift is small. Each univariate ANOVA finds only a small, non-significant difference. MANOVA tests the combination, and given the correlation, that combination should almost never happen by chance.
Try the ✨ demo. It sets small opposite differences (+0.35 and −0.35) and a strong correlation (r ≈ 0.8). Both univariate p-values are well above .05, so neither outcome shows a difference on its own, but the MANOVA p-value is below .001. Now drag the correlation to 0 and the advantage disappears: with independent outcomes, MANOVA is roughly the two tests combined.
The statistics you'll meet in the output
Software reports four multivariate test statistics, all summarizing how much of the outcome-space variation the group factor explains:
- Wilks' Λ (lambda). The share of variance not explained by groups; smaller = stronger effect. The most commonly reported.
- Pillai's trace. The most robust to assumption violations; many texts recommend it as your default.
- Hotelling–Lawley trace (SPSS prints it as Hotelling's trace) and Roy's largest root are variations on the same idea. Roy's uses only the single best discriminant direction.
With exactly two groups they all reduce to one test, Hotelling's T², the multivariate version of the t-test. The playground above computes it. With three or more groups and several outcomes they can disagree slightly, and then Pillai's trace is the safer choice because it is the most robust.
How big is the effect?
Those four statistics test whether the groups differ, not by how much, so a multivariate result needs an effect size as much as any other result does. The standard multivariate effect size comes straight from Wilks' Λ:
η² = 1 − Λ1/s (s = the smaller of: number of outcomes, number of groups − 1)
With two groups s = 1 and the formula reduces to η² = 1 − Λ. So a Wilks' Λ of 0.62 means the grouping accounts for 38% of the variance in the outcome set taken as a whole. Statistical software will print a partial η² beside each multivariate test if you ask it to. Report one for the multivariate result, then an effect size for each outcome in the follow-ups, because those tell a reader which measures changed. The effect-size converter will turn the per-outcome numbers into d or r if your field reports those.
The wider multivariate family
- MANCOVA = MANOVA + covariates: compare groups on several outcomes at once while adjusting for baselines, exactly like ANCOVA but multivariate.
- Multivariate multiple regression is the regression counterpart: several outcome variables predicted by several predictors simultaneously; MANOVA is the special case where the predictors are group dummies.
- Discriminant analysis is MANOVA run in reverse: instead of testing whether groups differ, it uses the discriminant directions to classify new cases into groups.
- Canonical correlation is the fully symmetric version: find the linked combinations of two whole sets of variables.
After a significant MANOVA you still need to find where the difference is. The standard follow-ups are univariate ANOVAs on each outcome (with a correction) to see which outcomes differ, or a discriminant analysis to see what combination separates the groups.
Assumptions, briefly: independent observations, multivariate normality (roughly: each outcome normal, jointly well-behaved), and similar covariance matrices across groups (Box's M tests this, but it is oversensitive, and with equal group sizes MANOVA is robust to modest violations). And choose outcomes that belong together conceptually. Adding ten unrelated variables to a MANOVA dilutes real effects.
Why it matters: many studies measure several related outcomes, such as a set of symptoms, a panel of biomarkers or a battery of test scores. MANOVA and its relatives test the whole profile at once, without inflating alpha and without ignoring the correlations between the measures.
Common questions
Does a significant MANOVA protect my follow-up ANOVAs from multiple testing?
It doesn't, although many researchers believe it does. The belief comes from the omnibus F in ANOVA, where passing the overall test is treated as permission to look closer. Here the multivariate test and the univariate ones test different hypotheses, so passing the first does nothing to control the false-positive rate of the tests that follow. Correct those separately, with Bonferroni or an FDR procedure across the outcomes. The logic fails in the other direction too: MANOVA can be significant when no single outcome is, as the ✨ demo above shows.
Should I report Wilks' lambda or Pillai's trace?
Wilks' Λ is the traditional default and what most textbooks tabulate. Pillai's trace is the most robust when assumptions are violated (unequal covariance matrices, unequal group sizes), so many methodologists recommend it outright. With two groups they (and Hotelling's T²) agree exactly. When they disagree materially with 3+ groups, that itself hints at assumption trouble, and Pillai's is the safer one to report.
What should I do after a significant MANOVA?
Find out where the effect is. The standard route is a univariate ANOVA on each outcome with a multiplicity correction, to see which variables differ. The more multivariate route is descriptive discriminant analysis, which shows what combination of outcomes best separates the groups. That is often the more faithful summary, because a combination is what MANOVA actually tested.