2012/1/10 eloy Martínez de las Heras
<[log in to unmask]>
Dear FSL list users and staff,
We are working in the analysis of TBSS and we designed two differents models:
1. Single-Group Average with 1 Additional Covariate of interest and 3 additional covariates.
Contrast 1. Where does the data correlate with the effect of interest?
Correct.
Contrast 2. Where does the data correlate negatively with the effect of interest?
Correct.
2. Unpaired Two-Group mean comparison difference with the 3 additional covariates.
Contrast 1. Where is the first group's mean greater than the second's?
Contrast 2. Where is the second group's mean higher?
The contrasts here will tell you where the covariate-adjusted mean for group1>group2 and group2>group1, respectively. This does not mean that mean of group1>group2. It means that at the mean value of the covariate (across all subjects), that group1>group2 or group2>group1. Imagine a positive slope of 2 for the covariate. If group1 has a covariate mean of 20 and group 2 has a covariate mean of 30 (assuming equal sample sizes), the difference of group1 and group2 is made for a covariate value of 25 in both groups. Since the slope is positive, the group2 mean is lowered and the group1 mean is increased. Thus, the groups could have the same mean response, but after adjusting for the covariate the responses could be significantly different.
All covariates has been demeaned"
demeaning does not change the slopes, only the interpretation of the group intercept terms.
Could you check the design of matrices? Are there any mistakes?
Nope. Designs are fine, just be careful about the interpretation.
Thanks a lot
Eloy
