Michel HOEN wrote:
> Hi dear SPM'ers,
>
> I'm currently starting running event-related fMRI experiments
> and I wondered if someone knew good references about ER-fMRI paradigm
> desiging and ER-fMRI statistical analysis.
>
> For instance, I would have several questions about the "null event",
> - what should it exactly be ?
The answer to this question depends upon the question that you want to
ask with the study. You might see recent papers by Craig Stark which
discuss the importance of picking the right baseline. Many people pick
visual fixation, which in some ways seems like the most "null" possible,
but that's probably not true.
>
> and
> - should it be modelled as a condition during statistical analysis or
> better not ?
in general people do not model the null, but rather let it be modeled
implicitly as part of the constant term in the model. My impression is
that modeling the null and comparing the modeled condition a another
condition will give you pretty much the same results as comparing
against an implicitly modeled null, but I've not tested that explicitly.
The issue that arises is that if you model the null explicitly, then
the combination of the null and other conditions will be highly
collinear with the constant (modulo truncation effects at the beginning
and end of the run). however, the way that spm performs the GLM (i.e.,
using SVD to do the design matrix inversion) should in theory be able to
deal with this.
cheers
russ
>
>
> Thanks in advance,
> Any advice or reference related to the topic would be helpfull,
>
> sincerelly,
>
>
>
> Michel Hoen.
>
> " Les lois de nos désirs sont des dés sans loisirs. "
> R. Desnos.
>
> Institut des Sciences Cognitives
> Equipe : "Cognition Séquentielle et Langage"
> UMR 5015 CNRS-UCBL
> 67, Bd Pinel
> 69675 Bron Cedex
> Tél : 04 37 91 12 65
>
> Homepage : http://miche77.free.fr/Homepro
> WEB: http://www.isc.cnrs.fr
|