Robust ANOVA (Welch & Trimmed Mean)
Also known as: Welch ANOVA, trimmed-mean ANOVA, heteroscedastic one-way ANOVA, Robust ANOVA (Welch & Trimmed Mean), robust ANOVA
Robust ANOVA compares the central tendency of three or more groups when the classical assumptions of normality and equal variances fail. It combines Welch's heteroscedasticity-adjusted statistic, introduced by Welch in 1951, with trimmed-mean tests advanced by Wilcox, giving reliable comparisons in the presence of outliers and unequal group spreads.
Key highlights
- Stays valid when group variances are unequal (heteroscedasticity), unlike classical ANOVA.
- The trimmed-mean version resists outliers by discarding a fixed fraction of each tail before comparing.
- Moderate difficulty and widely available, making it a practical drop-in when ANOVA assumptions are violated.
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 Robust ANOVA to compare three or more independent groups on a continuous outcome when normality or homogeneity of variance is doubtful, when outliers are present, or when group spreads clearly differ. It assumes independent observations and a sample of at least about 20. Prefer it over classical ANOVA whenever a Levene or normality check fails. Avoid it for very small samples (below about 10 per group), where a permutation test is safer, and increase the trimming fraction or move to a median-based method when more than a quarter of the observations are outliers.
Strengths & limitations
- Stays valid when group variances are unequal (heteroscedasticity), unlike classical ANOVA.
- The trimmed-mean version resists outliers by discarding a fixed fraction of each tail before comparing.
- Moderate difficulty and widely available, making it a practical drop-in when ANOVA assumptions are violated.
- Becomes unreliable for very small samples; with fewer than about 10 observations per group a permutation test is preferable.
- When the outlier ratio exceeds about 25%, a fixed trimming fraction is no longer enough and a higher trimming rate or median-based method is needed.
- Requires independent observations; it is not designed for repeated-measures or clustered data.
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 this different from ordinary one-way ANOVA?
Ordinary ANOVA assumes equal group variances and approximately normal data. Robust ANOVA relaxes both: Welch's statistic allows unequal variances, and the trimmed-mean version removes extreme observations so outliers cannot distort the comparison.
What does trimming actually do?
Trimming discards a fixed proportion (often 20%) of the smallest and largest values in each group before computing the mean. This trimmed mean is far less sensitive to outliers and heavy tails than the ordinary mean, so the test stays reliable on messy data.
When should I not use Robust ANOVA?
Avoid it for very small samples — with fewer than about 10 observations per group a permutation test is more trustworthy — and when outliers exceed roughly 25% of the data, where you should raise the trimming fraction or use a median-based method instead.
How large a sample do I need?
The method is intended for samples of at least about 20 overall, with enough observations per group that trimming still leaves a meaningful number of values to compare.
Sources
- 1.Welch, B. L. (1951). On the comparison of several mean values: an alternative approach. Biometrika, 38(3/4), 330-336.
- 2.Wilcox, R. R. (2012). Introduction to Robust Estimation and Hypothesis Testing (3rd ed.). Academic Press.ISBN 978-0123869838
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). Robust ANOVA. ScholarGate. https://scholargate.app/statistics/robust-anova