Dear Andrew,

I have three group (9 subjects for group A, 12 subjects for group B, and 8 subjects for group C) with two times (Pre and Post) measurement.
My example design matrix is:
-1 -1 0 1 0 0 
-1 -1 0 0 1 0
-1 -1 0 0 0 1
-1 0 -1 1 0 0 
-1 0 -1 0 1 0
-1 0 -1 0 0 1
-1 0  0 1 0 0 
-1 0  0 0 1 0
-1 0  0 0 0 1
 1 1  0 1 0 0 
 1 1  0 0 1 0
 1 1  0 0 0 1
 1 0  1 1 0 0
 1 0  1 0 1 0
 1 0  1 0 0 1
 1 0  0 1 0 0
 1 0  0 0 1 0
 1 0  0 0 0 1

First column represents the times effect. Second and Third represents the group effect with time effect. Others are subjects.
My Contrast is [0 1 1 zeros(1,29)]
I also set the exchange blocks [1:29 1:29]


Then I tried the post-hoc analysis. I separately did two ANOVA analysis (group A,B and C in two times) and three paired t-test (Pre and Post in three groups).
However, no matter how small the threshold (even 1) for F-test and t-test, I can not see any significant results.
I was wondering is my procedure wrong? Or is there any alternative way to approach the post-hoc?

Hi Hsu,

Your null hypothesis is not clear. What are you trying to test?

The contrast that you suggest seems to be testing whether the interaction differs between the two groups.I doubt that this is what you really want to test.

Your design matrix also looks a bit strange. Having -1, 0 and 1 in the same column will test for a linear ordering.

My suggestion is to simplify your design.

For example, can you average the connectivity matrices across the two time points?

Can you simplify your design by only testing two groups?

Andrew


Originally posted by Li-Ming Hsu:

Dear Andrew,

Thanks for your reply.

My purpose is trying to see the significant changes in the group a and group b in the Post compared to the Pre.
No significant of the group c between Pre and Post. And also no significant difference among group a, b, and c in the Pre.

Can I just simplify my connectivity matrix by subtracting Post and Pre, and then apply the one way ANOVA among the group a, b, and c?

Best,
Li-Ming

Hi Li Ming

If you are only interested in groups A and B, I suggest using a simple within-subjects t-test. There is an example of such a test in the Help section.

There is no need to include group C in this analysis.

To test for difference between group a, b and c in Pre, an F-test can be used. The Post data is not necessary for this contrast. 

In summary, I suggest that you consider generating a separate design matrix for each of your specific hypotheses and only using the data that is specifically needed to test your hypotheses.This will make things easier and also assist with the interpretation.

Yes - It is possible to subtract Pre and Post and perform testing on the resulting matrix.

Andrew
Originally posted by Li-Ming Hsu: