Hi,
MH is right. Your contrasts are asking two different questions, i.e. where your conditions significantly activate above baseline vs. each other.
One problem w/r to differential contrasts is that you ought to know the relative scaling of the EVs to each other, i.e. the approximate magnitude of the associated response, to get them right. In many cases, we have no prior knowledge about this. This is, in my opinion, often not properly acknowledged and may reduce the sensitivity / accuracy of the differential contrast, even though we like to model it at the first level (to get the DoFs right at the second). I would like to hear what the stats gurus think.
Cheers-
Andreas
________________________________________
Von: FSL - FMRIB's Software Library [[log in to unmask]] im Auftrag von Michael Harms [[log in to unmask]]
Gesendet: Mittwoch, 27. Oktober 2010 21:28
An: [log in to unmask]
Betreff: Re: [FSL] contrasts
Hi Aditya,
Based on what you wrote, why do you think that you did anything wrong?
Suppose that a given region shows a significant response to P1 alone,
but not P2 alone. That doesn't mean that a direct contrast of P1-P2 (or
P2-P1) will necessarily yield anything in that region. e.g., If a
region is just above threshold for significance for P1, and just below
threshold for P2, then it is quite likely that the difference of P1-P2
will be non-significant.
That distinction is why one tests contrasts directly, rather than
"guessing" about differences between two conditions based on the
appearance of their activation maps. (Dr. Smith has commented on this
previously in one of his papers, although I don't know off the top of my
head which one).
cheers,
-MH
On Wed, 2010-10-27 at 20:20 +0100, Aditya Kumar Kasinadhuni wrote:
> Hi all,
>
> I have run into a problem and request assistance. I have setup two contrasts on my model P1 and P2. Then I have gone ahead to set another two contrasts to look for regions having a a dominant effect of P1 and regions having a dominant effect of P2 indicating P1-P2 and P2-P1 as [1 -1] and [-1 1]. The results show me the regions involved with P1 and P2 solely very well, however the other two contrasts that I have mentioned above show nothing.
>
> Can anyone suggest what I could have done wrong?
|