Dear SPM99 user,
I have an fMRI study in which subjects are presented with 3 different types
of words (unpleasant, pleasant and neutral words), presented event-related
in two sessions. Each word was presented for 1 second with variable SOAs.
Subjects had to passively view the words. After the experiment subjects had
to remember as many of the presented words in a free recall. Free recall
was a paper-pencil version, intended to control that subjects attended
during passive word viewing. Nevertheless, now we want to use the
behavioural data as an additional regressor/parameter in our design. We
want to test whether the activation pattern found for passive viewing of
unpleasant and pleasant words compared to neutral words in our study is
associated with better recall of emotional words relative to neutral words
after scanning.
What I would like to do is correlate the activation elicited during viewing
pleasant, unpleasant, and neutral words with the number of correctly
remembered words (hits) across subjects.
Thus, each subject has the same vector, containing the hit rates for each
word as it was remembered across subjects (e.g [10 4 0 0 ….]. 10 means that
this word was remembered by 10 out of 15 subjects. 0 represents words that
have not been remembered by any subject at all.
I would hope that this can show me that activation of emotional encoding
during word viewing correlates with increased memory for emotional words.
Individual memory performance varied al lot such that we would get many
zero entries if we used the individual performance data.
Now I am a little bit concerned about the appropriate model to choose in
SPM99.
Does anybody of you have been confronted with a similar problem and have a
good idea how to model the data correctly in SPM99?
What I did so far to solve my problems:
What I did first was to modulate each word in a single trial analysis for
each participant separately. Then, I calculated a simple regression
including the ß-images of each word (without the ß-images for the globals)
and the appropriate regressor [vector containing the behavioural data;see
above]. However I am interested in second level analyis. Thus my question
concerns the correct processing of the single subject simple regression
data in a second level analysis?
Alternativley, I have modulated the words as one condition represented as
one column in the design matrix and one parameter (a vector containing the
behavioural data [(e.g [10 4 0 0 ….]. Again my question is what is the
correct processing of the single subject parametric data in a second level
analysis?
And last but not least, what is the difference between the two proposed
models, single trial analysis and parametric analysis in this case?
I would appreciate your help.
Thanks for your effort.
Cornelia
|