Dear Joe,
It is a very interesting approach to use RT as covariate in the first level
analysis: I have been thinking about this too, but I am not sure how to do
this.
My first level model then has 1 EV (EV1) for my events, which are all of the
same type, and EV1 is convolved with an HRF. The timings of my events are
contained in timings.txt (three columns, third column set to 1)
I can think of the following to include RT as covariate in EV2:
Do exactly the same for EV2 as for EV1, but now with the third column of
timings.txt set to 'RT' of each event. Then probably orthogonalize EV2 wrt
to EV1. Then any variance explained by the orthogonalized EV2 shows the
voxels that correlate with RT rather than being constant across RTs.
Do you think this makes sense, or is there perhaps a simpler way to do this?
Thanks very much,
Serge.
Serge ARB Rombouts, PhD
Dept Neurology / FMT / Alzheimer Center
VU University Medical Center
Amsterdam
The Netherlands
Tel +31 20 4440316
-----Original Message-----
From: Joseph Devlin [mailto:[log in to unmask]]
Sent: Tuesday, September 21, 2004 10:23 PM
To: [log in to unmask]
Subject: Re: [FSL] copes or zstats and reaction tIme data?
Dear Anna,
This is just my $0.02 but I would think that you should be using copes
(i.e. effect sizes) rather than zstats (statistics) in your
correlations. The reason is that zstats reflect both the effect sizes *and
the noise* in a given ROI while copes are effectively just a measure of the
evoked effect according to your contrast of interest.
It may not be very surprising that your RTs don't correlate well with ROI
cope values for a couple of reasons. First, if they are copes, they are
summary statistics across the entire time series per subject so the
correlation is the mean signal in a region across individuals with the
individual's mean RT. Second, The mean RT may not be very meaningful
across subjects, unless it first reflects a mean contrast between relevant
conditions and is also normalized to remove baseline differences in RTs
across subjects. Even so, you're looking for a single brain region to
explain the variance across subjects for a global behavioural measure...
It may be more convincing to enter RTs as a covariate in your individual
subject first level analyses and look for brain regions which correlate
with RTs and then determine whether these are consistent across
subjects. That way, you avoid all of the above issues and simply look for
a consistent brain region which varied with the individual's response times
across trials/blocks.
Hope this is helpful. Good luck!
Joe
--------------------
Joseph T. Devlin, Ph. D.
FMRIB Centre, Dept. of Clinical Neurology
University of Oxford
John Radcliffe Hospital
Headley Way, Headington
Oxford OX3 9DU
Phone: 01865 222 738
Email: [log in to unmask]
|