Der FSL'ers,
I tried to estimate a VBM multiple regression design with randomise, which
gave me unexpectedly high p values. As I am skeptical about my results, I am
wondering if I made a mistake at some point:
The following design with 1 covariate of interest (1st regressor) and two
nuisance variables (2nd column: age & 3d column: global GM volume) was
created with the GLM tool:
%! VEST-Waveform File
/NumWaves 3
/NumPoints 28
/PPheights 9.800000e-01 3.810000e+01 8.607700e+02
/Matrix
7.900000e-01 2.824000e+01 6.572900e+02
8.800000e-01 2.600000e+01 6.204000e+02
7.400000e-01 2.942000e+01 7.207500e+02
8.100000e-01 2.494000e+01 7.028600e+02
9.800000e-01 2.460000e+01 6.524600e+02
9.300000e-01 2.275000e+01 5.986000e+02
...
I was interested in positive and negative correlation between focal GM
volume and regressor 1. Therefore, I created the following contrasts in Glm:
%! VEST-Waveform File
/ContrastName1 pos_pA
/ContrastName2 neg_pA
/NumWaves 3
/NumContrasts 2
/PPheights 8.410671e-01 8.410671e-01
/RequiredEffect 4.138 4.138
/Matrix
1.000000e+00 0.000000e+00 0.000000e+00
-1.000000e+00 0.000000e+00 0.000000e+00
I started randomise with the following parameters:
* -c 2 and -C 2
* 5000 permutations
* Demean the data first
I checked the order of my covariate values and the smoothed (10 mm FWHM),
modulated gray matter partitions. They are ok. The cluster-level analyses
result in a cluster that covers the entire cortex. Did I something wrong?
Any help would be greatly appreciated...
Cheers,
Nikos
|