Hi Philipp,
a) to make it more interpretable, you may want to flip the sign of the S-mode value AND the extracted time course OR the associated spatial map. The S-mode is essentially the strength of the response (i.e. you can multiply the timeocurse amplitude by the S-mode, for example, if I got that right). See previous posts on this.
b) it depends: if the component is close to zero and has a wide SD it may simply be a quite inconsistent network. On the other hand, if you would compare a particular RSN both awake and under anaethesia you may expect it to show up for the formaer but possibly not the latter condition.
Hope that helps,
Cheers-
Andreas
________________________________
Von: FSL - FMRIB's Software Library im Auftrag von Philipp G. Saemann
Gesendet: Do 20.11.2008 15:43
An: [log in to unmask]
Betreff: [FSL] Group ICA: meaning of significance of F test
Hello FSL team,
we are performing group ICA on resting network data across different
physiological states, using the time concatenated version of group ICA.
We pooled subjects of all four conditions into the analysis, as we expect some
changes of component strength across the four conditions.
We are wondering if components need to be significant across the whole group
if the different conditions should be compared in the end.
The more specific questions are:
(a) What exactly do negative mode values (in the boxplot) indicate in
comparison to positive values? Does ANY deviation from zero indicate
contribution of that subject to the network?
(b) If the component is explaining variance to a non-significant degree, what
does this actually indicate? Can this component still be comared across our
four conditions (given, it is a meaningful component and multiple testing is
corrected for etc.)
Thanks a lot for any hints on this,
Philipp
|