Hi Walker,
I am glad the averages are coming out to be
reasonably similar.
Regarding the LSS results, you are likely seeing
that inter-trial variance is generally very large,
relative to inter-subject variance of the averaged
betas.
Particularly with short events (as would be modeled
with GAM), the pseudo-time series of event betas is
not going to be like a flat line at the mean (which
would be wonderful, were it to happen). It is
likely going to be as noisy looking as the original
data. Plot it and see.
That high variance, when viewed as repeated measures
in your ANOVA, will drive down the significance.
This is exactly why in FMRI, one generally uses many
events (20-25?) to estimate a single beta.
- rick