Hi,
Thank you very much for looking into this! I'm sorry to bother you again, but
I'm not sure if I have done the analysis correctly. I would be very helpful if you
or someone else could have a quick look and comment on what I did.
Let me first remind you of the problem:
I have 19 subjects, and each subject was scanned under 6 conditions. The
conditions correspond to three factors A, B, C. I want to analyze these data
with a 3-factor (2x2x2) repeated measures ANOVA. I’m mainly interested in
comparing the effect [B1C1>B2C1] across the two levels of factor A.
PROBLEM: In two of my experimental conditions (cells), there are very few
observations, and in some subjects none at all. To run the 3-factor ANOVA, I
would have to exclude these subjects from the analysis. My question was if
there was another way of testing the above interaction effect.
WHAT I DID: I set up a 2x2 design instead. The design matrix has inputs only
from conditions where C=1. This means I don’t have any empty conditions,
since these only occur when C=2 (conditions A2B1C2 and A2B2C2). And the
interaction effect AxB should equal “AxB given CV=1"), exactly what I’m after.
I followed the 2x2 ANOVA example on the website, just adding 19 EVs with 1s
to indicate inputs from a given subjects. I use four inputs from each subject.
These are the first level copes that correspond to the effect of each condition
vs. baseline. I set up three F-tests, just like in the example, and F3 is for the
interaction effect.
QUESTIONS:
1. Any caveats against testing a “reduced” design in the higher-level analysis?
2. Did I implement the analysis correctly?
3. I get F with df1=1, df2=54 – would that be correct here?
Thank you very much in advance,
Hanne
|