Statistics Glossary → Brown-Forsythe F* test
Brown-Forsythe F* testconcept
An F test of the equality of means that can replace one-way ANOVA when the group variances are not equal.
The classical ANOVA F test assumes equal group variances; when the variances differ and the group sizes are unbalanced, the Type I error rate can drift noticeably away from the nominal α. In the Brown and Forsythe (1974) F* statistic the between-groups sum of squares is unchanged, but the denominator is built from a weighted sum of each group's variance, with weights (1 - nᵢ/N), instead of the pooled variance, and the denominator degrees of freedom are estimated with the Satterthwaite approximation. The null hypothesis is the same as in ANOVA: all group means are equal.
This test is often confused with the Brown-Forsythe test for equal variances, which is the version of Levene's test based on deviations from the median rather than the mean and tests variances; F* tests means. Welch's ANOVA answers the same question; in simulation studies Welch is usually slightly more powerful, although F* can do better under some conditions with extreme values. Reporting both lets you check whether they agree.
An everyday example
You are comparing delivery times for three courier companies. One always delivers in 2 to 3 days, another sometimes delivers the next day and sometimes two weeks later, and you have far fewer records for the third. With spreads this different, it is safer to test whether the mean delivery times are equal with Brown-Forsythe F* or Welch's test rather than with classical ANOVA.
In MerQur
It is requested in One-Way ANOVA with the 'Extra output: Brown-Forsythe F*' checkbox; F*, the Satterthwaite denominator degrees of freedom and p are given as an extra row, and in the classical variant a Welch F row is added as well.