Steve:
==============Original message text===============
On Mon, 02 May 2005 1:23:43 pm CDT "Stephen J. Fromm" wrote:
>Perhaps this regressor should also be weighted as an exponential with two
> tails but I'm not sure what to use.
> What's your motivation behind using such weightings? Are you
> assuming the artifact somehow "spreads" throughout the time series?
We don't use such weightings. I'm assuming though that different causes of bad timepoints may have
various time courses. Perhaps a spike would be relatively instantaneous and just affect the one
scan, while a movement may have a buildup and tapering type of course. Perhaps without knowing no
weighting is appropriate. I guess another way would be a type of FIR set- perhaps 3-5 columns of a
single 1 corresponding to scans around the bad scan. That may be too fancy but would allow modeling
each of the scans surrounding the bad scan as a separate effect. The illustration below is what the
design would look like for those particular scans, with 0's above and below. This is not a contrast
but part of the design, i.e., 5 user defined effects of no interest. having a single 1 for the
scans to model.
0 0 0 0 0
1 0 0 0 0
0 1 0 0 0
0 0 1 0 0
0 0 0 1 0
0 0 0 0 1
0 0 0 0 0
Darren
|