Sphericity violation in RM Anova

gvt · Post by **gvt** » Thu Mar 07, 2024 5:45 pm

I am doing an RM anova, to compare repeated measures of 3 subjects in 4 different conditions. In the sphericity check, only the 4th condition is violated. I understand then that I must do a Welch anova.

Would it be correct in this case to perform only the Welch Anova for condition 4 and report the results of the other 3 conditions according to the RM Anova analysis?

Would there be any particularity in this case in writing the results? I'm wondering if I should put any specific mention or simply describe how the tests that have been carried out are normally done.

reason180 · Post by **reason180** » Thu Mar 07, 2024 10:06 pm

Hi.

It is not possible to calculate sphericity separately for each of the levels of your four-level repeated-measures ANOVA. Sphericity is a characteristic of the entire model, not of any particular level of the repeated-measures factor.

Also, Welch's ANOVA is for between-subjects factors only. it cannot be applied to a repeated-measures factor.

Regarding assumption-violation more generally, I would see the following:
https://forum.cogsci.nl/discussion/8822 ... and-levene
viewtopic.php?t=2731

gvt · Post by **gvt** » Thu Mar 07, 2024 10:41 pm

I confused the concept, I wanted to refer to the homogeneity of variances when I talked about sphericity...

reason180 · Post by **reason180** » Thu Mar 07, 2024 11:14 pm

OK. But the 'homogenous variance' assumption doesn't apply to repeated-measures factors (it only applies to between-subject factors). And even if you had a between-subject design rather than a repeated-measures design, it would make no sense to ask whether there is a significant difference between the variance in Condition A and the variance in Condition A--they're the same condition. Thus, equal (homogeneous) variance can only occur across multiple conditions.

jamovi