I had to run a similar analysis (linear correlation of FA with some behavioural score already demeaned).I wanted to know how important the addition of EV1 is,to this design.

Can you kindly elaborate on that?

Regards,

Archith

On Tue, May 24, 2016 at 12:34 PM, Anderson M. Winkler <[log in to unmask]> wrote:

Hi Jeff,

Please, see below:

On 24 May 2016 at 01:01, Jeff Brooks <[log in to unmask]> wrote:

Hello,
I'm attempting to use TBSS and Randomise to run a very simple analysis where I look at which regions of white matter have FA values that correlate with a behavioral measure. I checked the archives and saw some similar questions but nothing corresponding to a straightforward, single group case like this. Moreover, I couldn't find much info on the Randomise and GLM user guide pages on how to run a simple correlation like this. I'd just like to check and see if I'm implementing this properly so that I can confidently interpret my results. Here are my design.mat and design.con files:

design.mat:
/NumWaves 2
/NumPoints 16
/Matrix
1 0.434234212
1 0.545979109
1 0.477247646
1 0.099429575
1 0.395724822
1 0.324461103
1 0.449161679
1 0.512661832
1 0.170029416
1 0.309619646
1 0.567401353
1 0.125277039
1 0.473249147
1 0.323318238
1 0.500022694
1 -0.375590874

The above design is fine.

 

design.con:
/NumWaves 2
/NumPoints 16
/Matrix
0 1
0 1
0 1
0 1
0 1
0 1
0 1
0 1
0 1
0 1
0 1
0 1
0 1
0 1
0 1
0 1

These contrasts above aren't good. Try instead:

/NumWaves 2

/NumPoints 2

/Matrix

0 1

0 -1

 

And here is the command I used to run the analysis. I used the -D option to demean the data since I entered raw values: randomise -i all_FA_skeletonised -o tbss -m mean_FA_skeleton_mask -d design.mat -t design.con -n 500 -D --T2

All good, except that the -D shouldn't really be used, as the design above already includes an intercept (randomise deals with it, but it doesn't look right to have an intercept then demean, as that creates internally an all-zeroes EV).

 

Currently when I run this command it does several rounds of the 500 permutations instead of just one, which doesn't seem correct. Any insight is greatly appreciated.

The reason is that the original contrast file contained 16 instances of the same positive contrast. With the new one there should be only 2 runs of 500 permutations, one for the positive, another for the negative.

All the best,

Anderson

 

--

Project Assistant,
Language,Literacy and Music Lab

National Brain Research Centre

Gurgaon-122051,India