ANOVA when unequal variance

[Bibek Sapkota]

Hi
I used the Kruskal-Wallis H test instead of One-way ANOVA to compare income across three groups because Bartlett's test for equal variances was significant. But, my supervisor is not happy because I used a non-parametric test. Is there any way to do ANOVA when variances are heterogeneous across groups?

Regards,
Bibek

[Rich Goldstein]

us -regress- with the "vce(robust)" option and follow this with a post-hoc test for the omnibus test; see

help test

[Joseph Coveney]

Is there any way to do ANOVA when variances are heterogeneous across groups?

Yes.

See Option #3 at this post.

And, because you have only three groups, you can just as easily use

ttest <outcome variable> if inlist(<group variable>, <two of three group IDs, comma-separated>), by(<group variable>) welch

for pairwise comparisons. To see why, when you go to the post that I referred to immediately above, scroll down and see especially the posts by Bruce Weaver later in that thread.