We also noticed the residual non zero values in the FIR design matrix.
This only seems to happen when there is overlap in the bins of separate
trials of the same trial type, so I believe that the orthogonalisation of the
regressors may indeed be doing this (as Kaper suggested). Could this
orthogonalisation cause problems when you want to use the betas that
come out of the analysis for making PSTH's??
We used the betas of a FIR model to calculate PSTH's of overlapping
responses of visual events. We noticed that the PSTH's that came out of
this were not as expected. Approximately the first two TR's seemed to be
missing (BOLD responses started half way up).
To check that this was not caused by some anomalies in the data, we
used the FIR design matrix on self-generated 'perfect' data with exactly
the same result; again the first two TR's of the response formed by the
betas was missing. The FIR model combined could however perfectly
explain the data (R-squared was 100%).
On Thu, 7 May 2009 17:24:36 -0700, Kasper Winther Jørgensen
<[log in to unmask]> wrote:
>We have been struggling with this issue as well and it seems to come
>that SPM does orthogonalisation of the regressors within trial type
>line 283 in spm_fMRI_design.m) after the design is set up. As far as I
>see there is no way to avoid this (without commenting out these lines).
>My question is then: is it appropriate to orthonalise the regressors in a
>FIR model? If I want to estimate the HRF for a given condition is it valid
>to concatenate the estimated beta images into a 4D file and view the
>resulting timeseries as the response to my condition?
>2009/5/5 Joshua Goh <[log in to unmask]>
>> Thanks Karsten,
>> I tried turning of the high-pass (set to Inf), but as you can see in the
>> attached figures and the design matrix values, there are still residual
>> non-zero off-diagonal values other than the diagonal values.
>> Any other possible modifications?