Multivariate Analysis of Variance (MANOVA)
Also known as: Multivariate ANOVA, Çok Değişkenli ANOVA (MANOVA)
MANOVA is a parametric hypothesis test that simultaneously compares group means across multiple continuous dependent variables, controlling the inflation of Type I error that would result from running separate ANOVAs. Key multivariate test statistics — Wilks' Lambda, Pillai's Trace, Hotelling-Lawley Trace, and Roy's Greatest Root — were developed between the 1930s and 1950s, with Wilks' Lambda formalised by Samuel Stanley Wilks in 1932.
Key highlights
- Controls the overall Type I error rate when testing multiple related outcomes, avoiding the inflation from running separate ANOVAs.
- Detects multivariate effects — group differences in the pattern of outcomes — that no single-outcome ANOVA can reveal.
- Widely accepted in health, social, educational, and behavioural sciences with strong methodological backing (Tabachnick & Fidell, 2013).
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 MANOVA when you have two or more independent groups and two or more continuous dependent variables that are conceptually related and expected to be moderately correlated (roughly 0.30–0.90). Four assumptions must hold: (1) multivariate normality across dependent variables within each group, checked with Mardia's test or Royston's extension of Shapiro-Wilk; (2) homogeneity of variance-covariance matrices across groups, checked with Box's M test (interpret cautiously, as it is sensitive to non-normality); (3) moderate, non-redundant correlations between dependent variables — if they are uncorrelated, separate ANOVAs are equally powerful; (4) independence of observations, with a recommended minimum of 20 cases per cell.
Strengths & limitations
- Controls the overall Type I error rate when testing multiple related outcomes, avoiding the inflation from running separate ANOVAs.
- Detects multivariate effects — group differences in the pattern of outcomes — that no single-outcome ANOVA can reveal.
- Widely accepted in health, social, educational, and behavioural sciences with strong methodological backing (Tabachnick & Fidell, 2013).
- Requires substantially larger samples than ANOVA; the recommended minimum of 20 observations per cell per dependent variable is often difficult to achieve.
- Sensitive to violations of multivariate normality and to outliers, which can distort the variance-covariance matrices.
- Interpretation is more complex: a significant omnibus test must be unpacked through follow-up univariate tests, and the multivariate test statistics are not directly intuitive.
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
Which test statistic should I report — Wilks' Lambda, Pillai's Trace, or another?
Pillai's Trace is the most robust to assumption violations and is generally recommended when sample sizes are unequal or Box's M is significant. Wilks' Lambda is the most widely reported in the literature. When all assumptions are met and group sizes are equal, all four statistics give similar conclusions. Report the statistic your field conventionally uses, and always include its F approximation and p-value.
What should I do after a significant MANOVA?
Run a separate univariate ANOVA for each dependent variable, applying Bonferroni correction to control the per-comparison alpha. Only dependent variables with significant univariate tests are interpreted as driving the multivariate effect. Some researchers also perform discriminant analysis to characterise which linear combination of outcomes best separates the groups.
How many participants do I need?
A commonly cited guideline is at least 20 observations per cell (group × dependent-variable combination), and the number of cases in each group should exceed the number of dependent variables. With very small cells, the covariance matrix estimates become unstable and Box's M loses reliability.
What if my dependent variables are not correlated?
If dependent variables are uncorrelated (r near 0), MANOVA provides no power advantage over separate ANOVAs. The method is most useful when the outcomes share a common construct and have moderate positive correlations. Near-perfect correlation (r > 0.90) signals multicollinearity and makes the covariance matrices poorly conditioned — consider removing one of the redundant variables.
Sources
- 1.Tabachnick, B.G. & Fidell, L.S. (2013). Using Multivariate Statistics (6th ed.). Pearson.ISBN 978-0205849574
- 2.Field, A. (2013). Discovering Statistics Using IBM SPSS Statistics (4th ed.). SAGE.ISBN 978-1446249185
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). MANOVA. ScholarGate. https://scholargate.app/statistics/manova