Dear all,
shifting regressors in time may be viable for exploring data; but in the
case of EEG spectra this procedure seems to make sense only with reasonable
short events reflected in the regressors. EEG power anyway cumulates
information of extended EEG sweeps, and TR again introduces some temporal
blurring.
However, depending on the EEG data basis and its temporal relationship to
image acquisition (and on reviewer demand) data analysis might benefit from
shifting regressors.
The supposedly high correlation can be overcome using spm_orth, that allows
to create matching regressors (the second column vector is orthogonalized
with respect to the first one).
The problems arising with such correlations, of which design matrices to
use and their correct interpretation is explicitly described in:
Andrade A, Paradis AL, Rouquette S, Poline JB. Ambiguous results in
functional neuroimaging data analysis due to covariate correlation.
Neuroimage. 1999 Oct;10(4):483-6.
I hope this is of some help,
best regards and best wishes for 2006!
Renate Wehrle
At 12:35 02.01.2006 -0000, Helmut Laufs wrote:
Dear List,
When correlating EEG [power] data with EPI time series, we used the
canonical HRF for convolution of thtat data to specify a GLM.
A reviewer would like to see different offsets (e.g. shift same regressor
by +/- n seconds).
My questions now are:
Q0: Do I have to put them all into the same design, i.e. e.g. column 1:
'original regressor', column n: 'orig shifted by + or - n sec' etc?
Q1: They would be highly autocorrelated, wouldn't they? (By the way: TR is
about 4 seconds)
Q2: But running separate analyses, it will be difficult to compare (in
fact, at a different spot in the ms, the reviewer criticized just that)?
[Although this has been done in the past and made it into HIF {clinical}
journal...]
Q3: I could try and orthogonalize (spm_orth ?), but what would that mean,
or in other words, in what order would I have to do this?
Q4: Am I right in thinking that the AR(1) procedure only looks at the data
(Y), not the design (X) and so this would have nothing to do with my problem?
Q5: What's the way forward?
Again, any help greatly appreciated, and again best wishes for 2006,
Helmut Laufs
|