Dear Marlou,
Yes, this corresponds to the partitioned error approach as described in
the document you refer to.
The pooled error approach would have consisted in bringing six contrast
images per subject (for each cell of the 2x3 design) in a flexible
factorial design (three factors: subject, stimuli, session; main effect:
1, interaction: [2 3]; [1 -1 0 -1 1 0;0 1 -1 0 -1 1] F-contrast).
With the three contrast images you initially created (A1-A2 per
session), you could have entered them in a one-way ANOVA within-subject
and used an F-contrast [1 -1 0;0 1 -1]. This would, I think, match the
approach described by Martyn McFarquhar in his article mentioned at the
end of the wiki entry.
Best regards,
Guillaume.
On 11/11/2019 16:12, Marlou Lasschuijt wrote:
> Dear Reader,
>
> We have collected data on subjects receiving sips of chocolate milk [A1] and sips of water [A2] over 3 sessions [ B1 B2 B3].
> First I made a [A1 - A2] contrast per session, than on second level I analysed the main effect in a 1x3 full factorial model.
> However, as the contrast was already made on first level, the second level was a 1x3 model rather than 2x3 and no interaction effects could be calculated, i.e. difference between drinking chocolate milk and water over the 3 sessions.
>
> Than I considered making an interaction model on first level, because of the problem described above, and because I understood this is better concerning the degrees of freedom. I am not sure how to do this correctly as I have 3 sessions.
>
> I found this explanation example 2 - 3 factors: https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fen.wikibooks.org%2Fwiki%2FSPM%2FGroup_Analysis&data=02%7C01%7C%7C4c290c1d824e467abe4e08d766c35b3c%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C1%7C637090861541438858&sdata=9Nt8Eyz0l2Wjl35vyw4Y380uLz3ma%2FmPKjB6R2hkfZQ%3D&reserved=0
> there they describe to:
>
> make two contrasts on first level: ( I am not sure wheter this is a t-contrast or f-contrast, I am assuming it should be a t contrast)
>
> A1B1 A1B2 A1B3 A2B1 A2B2 A2B3 contrast1 [ 1 -1 0 -1 1 0 ] and contrast2 [0 1 -1 0 -1 1].
>
> Than they analyse this using a two-sample t test with F contrast [ 1 0 ; 0 1]
>
> This should than give the interaction effect between stimuli [A1 and A2] and session [ B1, B2 B3] .
>
> Do you think this is the right way of analyzing this?
>
> Thank you in advance,
> Marlou
>
--
Guillaume Flandin, PhD
Wellcome Centre for Human Neuroimaging
UCL Queen Square Institute of Neurology
London WC1N 3BG
|