Dear everyone,
For FIR, onsets are rounded to the next full TR, which is then coded by a certain, fixed value (seems to be no. of microtime bins a second) and the rest is zero-padded. Thus, e.g. onsets from the middle of the previous TR, the beginning of the current TR and the middle of the current TR all result in the same regressor matrix, e.g.
1 0 0 0 0 0 ...
0 1 0 0 0 0 ...
0 0 1 0 0 0 ...
0 0 0 1 0 0 ...
0 0 0 0 1 0 ...
To make use of the temporal information we could go with separate regressor matrices for trials with different onsets within a TR, which might be interesting when plotting the time courses for ROIs, thus increasing the temporal resolution (values from regressor matrix 1 reflect TR onset, values from regressor matrix 2 middle of the TR, ...). Taking into account the different scaling it would also be possible to contrast these time points on whole-brain level. However, the regressors would be based on different trials, thus the differences would not be unbiased.
After all, the idea is to capture certain time windows with a series of vectors. As shifted onsets result in shifted time windows, would it be advantageous/preferable from a theoretical POV to go with one regressor matrix a condition, but instead of the binary on-off pattern let the matrix reflect the different onsets by coding the "amount" of a TR being part of that time window? E.g. for some onset occuring in the middle of an TR this would result in a matrix like
0.5 0.5 0 0 0 0 ...
0 0.5 0.5 0 0 0 ...
0 0 0.5 0.5 0 0 ...
0 0 0 0.5 0.5 0 ...
0 0 0 0 0.5 0.5 ...
I'm interested to hear your opinions. Very likely this has already been discussed in the past, but a quick search on the list didn't provide anything related (or possibly, I just missed it).
Best
Helmut
|