Hi
When filtering out some components the GLM analysis would not be on against the
remaining components - the filtered data is not created by reconstructing the data from
the components you selected not to be removed. Instead, the ones you have selected for
removal are regressed out of the data first. In a way this means that the time courses of
those components serve as nuisance regressors for the GLM. There is an issue, however,
wrt the loss in degrees of freedom (the GLM does not know that the data was filtered
previously). This, however, should not bias the later statistics.
hth
Christian
On Fri, 8 May 2009 21:36:12 +0100, Adam Jacks <[log in to unmask]> wrote:
>Dear FSL users,
>
>I am currently analysing a data set from a speech production paradigm
>collected with block design fMRI. Although the participants maintained their
>jaws clenched during speech, you can imagine there was still a fair amount
>of head motion. As I try to make sense of the data, one of the approaches I
>have attempted for motion correction is using MELODIC to derive ICs, as
>suggested on previous messages. However, nearly all components (all but 3 of
>36) seem to be clearly associated with motion. My question is this: is it
>valid for me to filter out almost all of the components and run GLM analyses
>based on the few remaining? As I move to looking at comparing data from
>different individuals, is there a problem with removing more components from
>some participants than others?
>
>I appreciate any reflections you may have!
>
>Adam Jacks, Ph.D.
>
>Assistant Professor
>Department of Communication Disorders
>Texas State University- San Marcos
>
>Faculty Associate
>Research Imaging Center
>The University of Texas Health Science Center at San Antonio
|