Hi.
> When looking at the output from the FEAT report, we are left with a
> little ambiguity on how to interpret the full vs. th partial model fits.
> In a paradigm involving two interleaved off-on-off tasks, we set up the
> design matrix for each task individually, then select contrasts of:
>
> EV1 EV2
> C1 1 0
> C2 0 1
>
> expecting to see maps for each EV independently. The maps, in fact, seem
> to be concordant with this design. The time course plots, however, are
> confusing in that that "partial model fit" seems to conform to our
> expectations in setting this up, but the full model fit seems to include
> parameter weights from both EVs. If this is the case:
that's right - the full model fit is indeed the full model - i.e. ALL EVs
* by their parameter estimates, versus the data. thus you will see the
total fit.
> 1. why would this be the FSL default behavior
the partial model fit is data-full_model+partial_model vs partial_model
and does indeed make nice simple sense on such straightforward designs,
but is _very_ hard to meanignfully interpret with more complex contrasts
(even [1 -1] contrasts end up confusing), so the partial model fit is not
the main timeplot provided.
> 2. when FEAT selects best pixel and cluster averages, is it doing so on
> the basis of both EV's or only one (the latter, we hope and presume)
the best pixel is the max Z for the given contrast ie yes in your case for
contrast 1 uses only EV1 etc. likewise the cluster average is using voxels
for the given contrast that survived thresholding.
hope this makes sense!
Thanks, Steve.
|