Dear Darren
After thinking this through I would like to ask a further question about
which model I should use. If I understand you correctly I estimate my
group*condition and condition effects using the flexible factorial and the
main group effects using an anova or t-test with difference contrasts.
For the flexible factorial part I specified my factors like this:
1. Subject with independence yes and variance equal
2. Group with independence yes and variance unequal
3. Condition with independence no and variance equal
I then create a design matrix by specifying group*condition as interaction
effect of interest or by creating a design matrix based on the main effect
of condition and the interaction effect of group*condition (not sure which
would be the best).
When I create my design matrix by just specifying group*condition it looks
exactly like the design matrix I could have build with the full factorial
(only I didn't specify the factor subject). What would be the addition of
entering the factor subjects in the flexible factorial if they are not used
in the design matrix later. Because if I would build my model in the full
factorial (with group and condition as factors, specifying independence and
variance as I did with the flexible factorial) I would be able to estimate
the effect of group in the same model, since this model is apparently not
hampered by the incorrect error term.
Many thanks in advance,
Harma
|