I am using 3dDeconvolve with the -stim_base option to include trials with incorrect responses to the baseline (as suggested in another message board post).
I use the -stim_times format to define the rest of my regressors, but it seems that in order to use the -stim_base option I will have to make a -stim_files style file for the incorrect trials. Problem is, that my stimulus does not start at the start of the TR (as would be modeled in -stim_files). So is there a way to define a regressor using timing files (as in -stim_times) and include that in the baseline? Or is it best to use -stim_files in conjunction with the -stim_minlag and -stim_maxlag options?
I was not aware of such a limitation, it probably has to do with
ease of programming relative to its importance.
In any case, there are 2 options:
Run 3dDeconvolve twice, once to generate a full X-matrix,
once to do the regression. The first time, use -stim_times to
generate the regressor for the X-matrix. The second time, just
grab that regressor for a -stim_file option.
Do you really care? This has no effect at all on the other
betas or t-stats, nor does it affect the non-trivial contrasts.
The effects that it does have are:
a. It changes the Full-F, putting that regressor in the baseline.
b. That beta (and stats) will not be output (without -bout).
Since it has no real effect on the result, I would suggest not
worrying about putting it into the baseline model. Just use it
Thanks a lot for your response!
With regards to option 2, just to make sure I understand: does the inclusion of this factor in the baseline not make a difference because AFNI calculates the partial statistics of the Betas for the regressors? Such that the null hypothesis for EACH regressor contains this “error trials” regressor anyway?
If so, in what scenarios is the -stim_base option useful? Only when one is interested in the full model?
Yes. Individual stats are all partial.
The main use of -stim_base is just to prevent a bunch of
unimportant volumes from being added to the output stats
dataset, such as stats for motion parameters.
And yes, once in a while people want to be picky about
what applies to the full F-stat.
Rick is correct – it would be tricky in the code to put a -stim_times regressor in the baseline model, so I didn’t make that possible. As it really only affects the Full-F statistic (again, Rick is correct – imagine that!), I decided it wasn’t worth the programming effort.