Dear Cathy,
|1. VARIABLE EPOCH LENGTHS
|
|I acquired 80 images with a block design of 2 conditions
|times 5 block repeats. After discarding the first 3 images
|in the first block, I am left with 77 images: 5 A, 8 B,
|8 A, 8 B, 8 A, 8 B, 8 A, 8 B.
|
|I am having difficulty with defining the parameters for
|the design matrix of the fMRI analyes:
|Assuming A is the condition of interest, I have to define
|variable SOAs ([0 13:16:77]) as well as variable epoch lengths
|(5 8 8 8 8). The epoch length specification doesn't seem
|to work. The only work around I have found is to erase all
|the images of the first block. Is there another way?
My understanding is that in SPM99 each epoch-related response is defined
in terms of basis functions that define the length of that epoch type. So
within any one epoch type, the epoch length must be the same throughout.
The solution here is therefore to model epochs of different lengths as
different conditions and estimate the average effect using an appropriate
contrast. This will cost you a small penalty in terms of degrees of
freedom for modelling the extra 'short' epoch as a third condition.
|2. COMBINING DATA FROM MULTIPLE SUBJECTS
|
|I have questions regarding how to combine the data from
|multiple subjects and several sessions in a single statistical analysis.
|As far as I know, there is no option on the fMRI stats model set up
|to indicate multiple subjects-- only multiple sessions.
|
|Let's say I have 2 subjects that have 2 sessions each. Both sessions
|alternate the same 2 conditions, but the order of session 1 is ABAB and
|the order of session 2 is BABA. Using the fMRI stats model set up in
|SPM99b, when asked to input the number of sessions I have specified 2.
|When asked for the scans for session 1, I included session1 data from
|Subject 1 and session 1 data from subject 2. Is this correct? Is
|there a better way of doing this (e.g., specifying 4 sessions and
|putting each subject's images from each session into separate sessions)?
There's no distinction (in a fixed effects model) between subjects and
sessions in SPM99b. The second way you suggest (specifying four sessions)
would be better. If you combine two sessions from different subjects in
one 'megasession' then you will no longer be correctly modelling the mean
differences in activity between subjects as this is estimated on a
session-specific basis (the effects of no interest in the design matrix).
As between-subject variability in fMRI is often high compared to
within-subject variability, this is not a good idea.
More generally, if you enter four separate sessions, and then estimate the
average effect using a contrast across all four sessions, you will be
using a fixed effect model over two subjects. Whether this is most
appropriate, or whether you would prefer to use a random effects model
(which would require more subjects for reasonable power) depends on the
experimental question you're interested in- see Karl's paper in NeuroImage
I 10:1-5
best wishes,
Geraint
-----------------------------------------------------------------
Dr. Geraint Rees
Wellcome Advanced Fellow Lecturer
California Institute of Technology Institute of Neurology
Division of Biology 139-74 University College London
Pasadena 12 Queen Square
California 91125 London WC1N 3BG
voice (626) 395-2880 voice (171) 833-7472
fax (626) 796-8876 fax (171) 813-1420
http://www.klab.caltech.edu/~geraint [log in to unmask]
-----------------------------------------------------------------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|