Welch's Analysis of Variance
Also known as: Welch's F-test, heteroscedastic one-way ANOVA, Welch ANOVA — Heterojen Varyans ANOVA
Welch ANOVA is a parametric hypothesis test that compares the means of three or more independent groups when their variances are not equal. Introduced by B. L. Welch in 1951, it replaces classic one-way ANOVA whenever the homogeneity-of-variance assumption fails, while still requiring approximately normal data.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use Welch ANOVA to compare three or more independent groups on a single continuous outcome when group variances are unequal — for example when Levene's test is significant. The outcome should still be approximately normal in each group (check with Shapiro-Wilk per group), groups must be independent, and a sample of at least roughly 20 is advisable. If normality is clearly violated in small samples, switch to the nonparametric Kruskal-Wallis test instead.
Strengths & limitations
- Stays valid when group variances are unequal, where classic ANOVA inflates the false-positive rate.
- Together with the Brown-Forsythe test, it is among the most reliable heteroscedastic ANOVA alternatives.
- Relaxes only the equal-variance assumption, so it remains a familiar parametric mean comparison.
- Still requires approximately normal data in each group; it does not rescue non-normality.
- Only signals that some means differ — a separate post-hoc test is needed to locate the differences.
- Needs reasonable per-group sample sizes; very small or sparse groups make the variance weights unstable.
Frequently asked
When should I use Welch ANOVA instead of classic one-way ANOVA?
Use it whenever the equal-variance assumption is violated — for instance when Levene's test is significant or the group standard deviations differ markedly. Under unequal variances classic ANOVA can produce too many false positives, while Welch ANOVA keeps the error rate close to the nominal level.
Does Welch ANOVA also handle non-normal data?
No. It relaxes only the equal-variance requirement; the outcome must still be approximately normal in each group. If normality is clearly violated, especially in small samples, the nonparametric Kruskal-Wallis test is the appropriate alternative.
Which post-hoc test follows a significant Welch ANOVA?
Use the Games-Howell post-hoc test, which, like Welch ANOVA itself, does not assume equal variances. Standard Tukey comparisons assume homogeneity and are not appropriate here. Report an effect size such as eta-squared or omega-squared alongside.
How does Welch ANOVA differ from the Brown-Forsythe test?
Both are robust alternatives to one-way ANOVA under unequal variances. They use different correction formulas, and Welch ANOVA is widely regarded as one of the two most reliable heteroscedastic options; many analysts report both and treat them as confirmatory of each other.
Sources
- Welch, B.L. (1951). On the Comparison of Several Mean Values. Biometrika, 38(3/4), 330–336. link ↗
How to cite this page
ScholarGate. (2026, June 1). Welch's Analysis of Variance. ScholarGate. https://scholargate.app/en/statistics/welch-anova
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Kruskal-Wallis testStatistics↔ compare
- One-way ANOVAStatistics↔ compare
- Welch t-testStatistics↔ compare