Print

Print
Dear Falk,

Please see below:


On 2 January 2015 at 16:20, Falk Mancke <[log in to unmask]> wrote:
Dear FSL-experts,

I am currently trying to do an analysis using FSL FIRST. I followed the recommendations of the UserGuide respectively the mailing list. However, I am not sure whether I set up the design matrices/contrasts correctly and would really appreciated it if you could take a look at them:

1. to analyze shape differences between a patient and a control group (EV1), controlling for age (EV2), sex (EV3) and IQ (EV4, using scores from Raven's Progressive Matrices) the matrix/contrast of the attached file shape.fsf was used. I did not demean the EVs, since I used the –D option of randomise

2. to analyze a correlation (EV1), again controlling for age, sex and IQ the matrix/contrast of the attached file corr.fsf was used. Again scores were not demeaned.

Are these correct?

Yes, both are fine as long as you use -D. Otherwise, you must include an intercept in the model, that is, a column full of a non-zero constant value.

 

Besides it would be really great, if you could answer the following questions:

- setting up the design matrix, I was confused that group was specified using one EV1 only instead of two, as described here http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Randomise_details-1 - is that because FIRST treats matrices different then FEAT?

These are equivalent. It's possible to use a single EV to code for group, plus an intercept, or two EVs for group, without the intercept, or one EV for group without the intercept with the -D.
 

- groups do not have the same size, does this matter?

It's fine. They don't need to be of the same size.

All the best,

Anderson