Hi,
I recently saw the thread below regarding correlations between groups and I was wanting to run this approach with my own data.
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=fsl;dc11d1db.1305
However, I got something that confused me. For my design I have Age as a covariate and behavioral scores as the variable of interest making my design look like this.
Con Patient Age_Con Age_Patient Behav_Con Behav_Pat
1 0 Age 0 Behav 0
1 0 Age 0 Behav 0
0 1 0 Age 0 Behav
0 1 0 Age 0 Behav
Age and Behavioral performances were both demeaned.
I previously ran a correlation analysis within my patient group (essentially same design but only EV2, EV4, EV6, and no Controls) and since this is a new approach for me I wanted to double check it with my previous result. Therefore, I set up my contrast as:
0 0 0 0 0 1
0 0 0 0 -1 1
0 0 0 0 1 -1
Now the results I got from the group comparisons (aka. contrast 2 and three) agreed with my hypothesis. However, the first contrast that I believed would replicate my within group comparison yielded 2613 more significant voxels than previously seen. I would like to chalk this up to variability with permutations but I feel its a bit much to assume that. I am using 5000 permutations in my randomise script.
Thank you,
Jeremy