Hi Rick,
I do understand that the loss of DoF is a problem, I was just wondering why it was used in example 9 but not 11 and whether the more elaborate nuisance regression would make it either problematic or not needed. After reading all those bp discussions now I don't feel too keen on using it at all any more, but I have been advised to use it and I also read here somewhere that it is still the norm and regression can not remove all the noise. Maybe a compromise could be to use a less stringent threshold than 0.01-0.1?
I actually already tried with 1d_tool.py since I found this suggestion from you in another thread. If I go with the threshold of less than 25% censored, I am losing 30 (out of 212) of my participants with enorm .3 and 17 with enorm .5 (see below). I actually thought it would be much worse. I have a mixed group 8-25 year olds and participants with stronger impairments usually didn't want to or couldn't do the scan at all so maybe that is why I have less strong movers than expected. I didn't know how to include the outlier threshold in 1d_tool.py though, but that should only be a couple of volumes extra that are outliers, right?
Enorm 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Lost part. 45 37 30 27 24 21 17
Ah, I understand. I guess I rather skip the global WM regressor then.
Thank you ever so much!
Janina
