More on ANOVA

Unequal sample sizes cause confounding.

Comparing unweighted means eliminates confounding.

Type III sums of squares test an effect while controlling for other effects, thus eliminating confounding.