>
Hi
> I have 2 questions regarding FEAT analysis.
>
> a) I have a situation where in study of 12 subjects I want to
> compare 2
> modeling strategies (with and without single trial psychophysiological
> information). When modeled entirely seperately using the quantitative
> information (3rd column in 3-column log file containing
> physiological data
> instead of fixed stimulus intensity) does not give "more"
> activation in any
> of the 12 subjects. However, in the group level analysis, all of a
> sudden
> the model containing psychophysiological information gives more (also
> reasonable) activation, but only in the mixed effects flame.
>
It is because
1) the pp information is now accounting for some relevant information
that was previously being assigned to your cope of interest -
therefore the cope of interest is more likely to reflect the same
thing across subjects.
2) The pp information accounts for new information that was
previously regarded as noise, so that when you take it up to the
group level, the 1st level variances are smaller, and more suitably
balanced across subjects.
> How is that possible?
>
> b) I want to compare these 2 models more directly, asking which one
> is more
> informative or even, what is the quantitative information adding to
> the
> "regular" model. What is the way to go? Contrasting seems to give
> unreasonable results. Orthogonalisation? If itīs orthogonalisation how
> should it be set up? Again: I have to identical time courses with and
> without quantitative information in the 3 column of the custom
> logfile.
>
This is exactly what you have already done, by including the pp to
complete with your effect of interest. The fact that your effect fo
interest has changed significantly due to the pp information tells
you that the pp information can account for variance in the FMRI data.
Do not orthogonalise - you want to let the two regressors compete to
explain the signal.
You can tell how much ambiguity there is between the two signals, by
i) correlating them
ii) looking at the deisgn efficiency in 1st level feat - if you need
a ridiculous signal change (>3%?) to detect your effect, then
probably the regressors are ambiguous.
If you want to go further, an F-test at the first level in each
subject can tell you whether your pp is significantly reducing the
variance, but, from what you have told us, it almost certainly is -
It very much sounds to me as though including the pp is the right
thing to do.
T
> Thanks!
>
> Arian
>
>
>
|
