Hi
On Tuesday, Oct 28, 2003, at 13:14 Europe/London, Johanna Pekkola wrote:
> Dear FSL people,
>
> I am having problems to understand a contrasting & design matrix
> question.
>
> My experiment is basic: a block design with one active condition and
> a baseline. I have run the GLM-analyses by having one EV (EV1, for the
> active condition) and one contrast with value 1 for EV 1. So, the
> resulting statistical maps show the regions where voxels time-courses
> follow the model in the specified level of signifigance. These areas
> are
> then considered "activated".
>
That sound right
> Iīve read articles where baseline > active condition contrast is
> represented (in addition of active condition > baseline contrast), with
> same kind of block design. How is this done?
> I canīt feed in both EV 1 and baseline in the design matrix and have a
> 1 -1
> contrast since the two EVs would be a linear combination of each other.
There is a difference between multiple EV's and multiple contrast. With
multiple EV's you have to make sure that the design matrix is not rank
deficient (i.e. one of the EVs is a linear combination of the other
ones - this will make the model unidentifiable)
You can, however, have as many contrasts as you wish (contrasts are
used to combine beta estimates - i.e. they can't interfere with the
estimation anymore). In your case, the design only contains your single
EV1 and a contrast of 1 asks for regions in the brain where
'activation>baseline'. A contrast of -1 asks for 'baseline >
activation'.
There are more details at
http://www.fmrib.ox.ac.uk/fslcourse/fmri/s_2722_con.htm
or in chapter 14 of 'Functional MRI - an introduction to methods'
ta
christian
> Could I use orthogonalizing of the contrasts? Or is this kind of
> thinking at all sensible since I do not have a baseline for the
> baseline?
> And what would be represented if I ran the EV 1 contrast with value -1
> instead of 1?
>
> Sorry for these stupid questions...
> And thanks for help the previous times.
>
> Best wishes,
> Johanna Pekkola
> TKK/LCE
> Espoo, Finland
|