Hello
I am a bit confused about the ideal way to use randomize for
a single group design.
My goal is to test for correlations between FA and % signal
change in an ROI
Based on the exchanges that I found on the forum which
matched my goals, I used the GLM gui to generate a design matrix (with EV1 as
the demeaned % signal change and group set to 1) and a contrast matrix (with 1
and -1, to check for both positive and negative correlations).
I ran the standard randomise command:
randomise -i all_FA_skeletonized -o tbss –m mean_FA_skeleton_mask
-d design.mat -t design.con -m <mask_image> -D -n 500 - -T2 –V
However, I recently came across the instructions in the TBSS
website which says that for a single group simple design, one does not have to
use a design or contrast file.
But the website also says that “randomise will
automatically select one-sample mode for appropriate design/contrast
combinations”.
So here are my questions:
1. Have I done the correct analysis? Should I use: randomise
-i OneSamp4D -o OneSampT -1 -v 5 –T (should I use - - T2 here?)
2. If the GLMgui based analysis is indeed correct, then do
the output files tstat1 and tstat 2 reflect the 1 and -1 contrasts that I
specified?
3. Which output file should I pass for cluster thresholding
for multiple comparisons?
4. How do I export the FA vs EV1 correlation data from a
cluster that shows a significant correlation?
Many thanks as always
Cibu Thomas