Dear Lucas,
On 20/01/17 17:38, Lucas L. wrote:
> Dear Guillaume
>
> Your answer is much appreciated.
>
> 1. Could I equally use a flexible factorial design?
Almost universally, the answer is yes as any of the other designs
available in SPM can be specified with a flexible factorial design (with
appropriate non-sphericity modelling):
* one-sample t-test: one factor with one level.
* two sample t-test: one (between subject) factor with two levels.
* paired t-test: one (within subject) factor with two levels and a
subject factor.
* one-way ANOVA: one (between subject) factor with N levels.
* one-way ANOVA within subject: one (within subject) factor with N
levels and a subject factor.
* full factorial: one or more (between subject) factors with N levels.
Now, I guess your question is whether to use a flexible factorial design
with a between subject factor (group), a within subject factor
(pre/post) and a subject factor, then use a contrast [1 -1 -1 1 0 ... 0]
to test for a group by intervention interaction. The answer is yes; it
will be a pooled error model while the previous suggestion (two sample
t-test of the differential within subject effect) corresponds to a
partitioned error model.
> 2. What complicates the things a bit is that the analysis is done on spatial maps from ICA GIFT. I have several components and in the middle of that I need to somehow mask with one-sample t-test maps.
If you have multiple components from an ICA then maybe this can be
modelled as another within subject factor. I don't use ICA so hopefully
others have further suggestions on a typical way to analyse them.
Masking can be done at the inference stage.
Best regards,
Guillaume.
> Would you have any further thoughts on that?
> Thank you again,
> Lucas
>
>
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|