ANOVA - Does Welch's test assume homogeneity of variances, or not?

CowboyBear

Super Moderator
#4
Got it. So a failed Levene's test would confirm the usage of Welch's ANOVA
This is common practice but actually a bad idea. It's better to just use Welch's t-test always. This is indeed the default in R.

See: Zimmerman, D. W. (2004). A note on preliminary tests of equality of variances. British Journal of Mathematical and Statistical Psychology, 57(1), 173–181. https://doi.org/10.1348/000711004849222