Dear Josselin,
You shouldn't model this in this way.
Instead put in a single EV for each scanner where that EV is
1 for scans from that scanner and 0 for all other scans.
In this way it can model any consistent change between
scanners and effectively remove it.
As for the "group" column, you can either put a separate
value for each scanner, which will model a different variance
from images in that scanner, or you can just have them all
in the same group (1 for all). It depends partly on whether
you expect any interesting difference in variance between
scanners *and* whether you have enough data from each
scanner to get a good variance estimate. I would say that
you need at least 10 scans (ideally 20+) from each scanner
to treat them separately in the group column.
I hope this helps.
All the best,
Mark
On 2 Jun 2011, at 21:00, Josselin Houenou wrote:
> Hi,
> I am analyzing a huge bunch of data (>300 T1 MRI) coming from three different scanners in a VBM analysis. I am not interested in studying the effect of the scanner by itself and I would like to remove this effect. I am not sure what to do. If I state this as an EV, I will have to set values to e.g. 1, 2 and 3 (before demeaning). It would mean that scanners 1 and 2 are more resemblant than 1 and 3, which is not true.
>
> Or should I model them as 3 "groups" in Glm such as :
>
> Group EV1 EV2 EV3
> 0
> 0
> 0
> 0
> 1
> 1
> 1
> 1
> 2
> 2
> 2
>
> ????
>
> Thanks
> Josselin Houenou
>
|