Dear Jack,
>
>... but considering the 'Okam knife'
>principle, one could suggest that correlation + PPI would be more simply
>explained by a change in effective connectivity (in the meaning of
>efficiency)
>from A to B. Would you reject that as a reviewer (keeping this very
>hypothetical formulation) ?
>
I agree. If area B appears in the PPI with area A, the case for an
effective connection would be strengthened by also demonstrating an
increase in correlation. Its only if an effective connection was inferred
from a correlation alone that I would object; in this case Occam's Razor
might tend to make me reject the hypothesis of an effective connection in
favour of the more likely explanation of stimulus-driven correlations.
>>The main effect of the psychological factor is
>> modelled out as an effect of no interest in PPI studies. With this gone,
>> it is still perfectly possible to imagine variance in area A explaining
>> more of the variance in area B in a task-dependent way without any change
>> in the overall activity in the task vs control conditions. An
>> over-simplistic example would be if area B receives two alternative inputs,
>> one from area A (during test conditions) and the other from area C (during
>> control conditions) with no net change in the activity in any of these
>> three areas between conditions. This scenario is possible whether the
>> connection is 'direct' or 'modulatory'.
(Sorry, you are right, I meant 'C'; above is the correct version of the
paragraph.)
>Could you clarify why this symmetrical proposition (Activation by the psychol
>factor + PPI => stable modulatory effective connectivity as the simplest
>explanation) doesn't sounds as reasonable as the former one ? I may have
>missed
>something but, it seems that your argument could apply to both conditions.
I agree that it is as reasonable. I was just pointing out that a net
change in activity in an area in response to the psychological factor is
not a precondition for inferring effective connectivity, and therefore
masking by the psychological factor (which you suggested earlier) is
unnecessary. In most circumstances I imagine that there is likely to be a
net change in activity, but I have seen demonstration of a highly
significant PPI in an area which doesn't show up in the original contrasts
(in other peoples' data).
>Say that A is correlated with the factor F1
>Say that the PPI of A*F2 (a second factor) gives B.
>The question is wouldn't we have to check that F2*F1 does not also fit B ?
This seems to me to be correct. Not being very mathematically minded, I
had to illustrate this for myself with a trivial example, as follows.
Lets give F1 two levels, X and Y, and F2 two levels, P and Q. Now let's
imagine that there are two unconnected areas, area A which shows up in the
main effect of F1 (X vs Y), and area B which shows up in the interaction
between F1 and F2. But there is no main effect of F2 (P vs Q) in either
area. For example, the levels of signal (in arbitrary units) might be:
Area A:
XP = 1; YP = 0; XQ = 1 and YQ = 0
Area B:
XP = 1; YP = 0; XQ = 0 and YQ = 1
Because of these response patterns, area B must show up in the PPI with
area A using F2 (P vs Q) as the psychological factor. Thus in the context
of P, area A appears to have a positive influence over area B (when A gives
1, B gives 1, and when A gives 0, B gives 0), but in the context of Q, area
A appears to have a negative influence over area B (when A gives 1, B gives
0, and when A gives 0, B gives 1). Yet this need not imply any sort of
connection between them.
I must admit that I wasn't aware of this, so I guess I'll have to go back
to Friston et al., 1997 ( NeuroImage 6, 218-229) and try to get to grips
with that paper properly. Perhaps if this conclusion is incorrect, though,
someone could put us right!
Best wishes,
Richard.
from: Dr Richard Perry,
Clinical Research Fellow, Wellcome Department of Cognitive Neurology,
Darwin Building, University College London, Gower Street, London WC1E 6BT.
Tel: 0171 504 2187; e mail: [log in to unmask]
Pager: 04325 253 566.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|