Print

Print
Hi Vasilis,

In this case, you can put both EVs in the same model, as well as an intercept (so, 3 EVs overall) or demean data and design (option -D) without the intercept then.

If either or both EVs explain some of the variance in the data, they will help to reduce the variance of the error, and this should increase the value of the statistic. If they are correlated to some extent, it won't be possible to disambiguate the effects of each, and having both in the model is a safer approach.

All the best,

Anderson


On 10 November 2014 14:54, Vasilis M. Karlaftis <[log in to unmask]> wrote:
Dear Anderson,

I have one measurement per subject, so my input image has the subtractions of pre FA skeletons from post FA skeletons. This is why I am using the single group design with ones in the Group column in the design matrix.

Best regards,
Vasilis

On 10 November 2014 14:29, Anderson M. Winkler <[log in to unmask]> wrote:
Hi Vasilis,

A question: for the two behavioural measurements (learning rate and strategy score): do you have one measurement per subject, or one measurement per scan (that is, two measurements per subject)? Depending on the answer, the design changes.

All the best,

Anderson


On 10 November 2014 11:18, Vasilis M. Karlaftis <[log in to unmask]> wrote:
Dear FSL experts,

I have DTI data in two timepoints, one before and one after training. I am interested in the FA differences between these two timepoints, so I followed the TBSS analysis along with this http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Randomise_details-11 for randomise. What we want to investigate next is how the FA changes are correlated with the behavioural measurements. Thus I followed these instructions http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Randomise_details-4.

My question is how many EVs should I include in my design? I am interested in two behavioural measurements, learning rate and strategy score. Should I create a design matrix with two EVs, one for each, and contrasts as follows?

1    0
-1   0
0    1
0   -1

Or would it be ok if I use two different designs, each behavioural measure separately?
By reading previous emails, I understood that having more than one EVs in the design can affect the variance of the data, even if you create a contrast for only one of them. So if I assume that both measures affect the data, does this mean that it is better to use one design file for both?

Best regards,
Vasilis