Repeated measures ANOVA Posthoc ES

gvt · Post by **gvt** » Fri Dec 22, 2023 1:12 pm

I´m trying to do ANOVA repeated measures in JAMOVI but I can´t see effect size in posthocs. Will be correct to do separatlety paired T-student? In t-student I can only see Cohens, is possible to show eta partial?

reason180 · Post by **reason180** » Fri Dec 22, 2023 6:26 pm

Personally, I would use the ANOVA posthocs output table to hand-calculate each Cohen's d effect size.

Specifically, I would convert each 'SE of differences' to an 'SD of differences' by by multiplying the 'SE of differences' by the square root of the number of differences. I would then divide each mean difference by the corresponding SD of differences to get Cohen's d for each posthoc test.

gvt · Post by **gvt** » Sat Dec 23, 2023 9:12 am

thank you. But do you find the approach of the RM ANOVA + Tstudents correct?

reason180 · Post by **reason180** » Sat Dec 23, 2023 3:27 pm

There is controversy as to whether it is correct or not. On grounds that the ANOVA assumes equal variances, some stats programs use the same pooled error term for each test in the set of post hoc tests. In such a situation, individual t tests would usually not produce the same result as ANOVA post hoc tests would. jamovi used to use a pooled error term for repeated-measures post hoc tests, Now it does not.

See:
viewtopic.php?t=741
viewtopic.php?t=114
viewtopic.php?t=1924

Therefore, now, individual t tests should produce the same result as the repeated-measures ANOVA (I have only verified this for one-factor, repeated-measures models; and I suggest you check all of this yourself by actually running all of the analyses).

A caveat: If there are any missing values, individual t tests will give the same results as repeated measures ANOVA, if you first delete all the data for any individual who has missing data for ANY of the possible individual t tests.

Note that other stats programs still use a pooled error term, while still others let the user choose.

jamovi