ANOVA estimated marginal means
Posted: Wed Mar 17, 2021 5:43 pm
I have some fake data comparing two age groups (young vs old) with one within-subjects factor (condition: A vs B ). The marginal means reported by the RM ANOVA in jamovi don't match what I expect and I'm trying to understand how they are calculated.
I've attached a jamovi file showing all the analyses described here.
The means are
young A = -0.257
young B = 0.693
old A = 0.729
old B = 1.564
I expect the marginal means for the main effects to be:
young = (-0.257 + 0.693)/2 = 0.218
old = (0.729 + 1.564)/2 = 1.146
A = (-0.257 + 0.729)/2 = 0.236
B = (0.693 + 1.564)/2 = 1.129
And I expect the marginal means for the interaction to match the means in the descriptives as given above.
The reported estimated marginal means report exactly the expected pattern, except that every reported value (across main effects and interactions) is 0.232 less than expected. I don't understand where this contstant of 0.232 is coming from.
If I instead calculated the model as a linear mixed model in the GAMLj module (again, see attached jamovi file), I get the exact same ANOVA results, but the estimated marginal means are now the values I expect.
I've attached a jamovi file showing all the analyses described here.
The means are
young A = -0.257
young B = 0.693
old A = 0.729
old B = 1.564
I expect the marginal means for the main effects to be:
young = (-0.257 + 0.693)/2 = 0.218
old = (0.729 + 1.564)/2 = 1.146
A = (-0.257 + 0.729)/2 = 0.236
B = (0.693 + 1.564)/2 = 1.129
And I expect the marginal means for the interaction to match the means in the descriptives as given above.
The reported estimated marginal means report exactly the expected pattern, except that every reported value (across main effects and interactions) is 0.232 less than expected. I don't understand where this contstant of 0.232 is coming from.
If I instead calculated the model as a linear mixed model in the GAMLj module (again, see attached jamovi file), I get the exact same ANOVA results, but the estimated marginal means are now the values I expect.