Hi John,

The 3dUndump program reads in the data as float, and stores it
based on any '-datum' option provided. If none was given, it
defaults to short.

NaN values are read in as that. If you save your dataset as
floats, I belive they are zero'd out. But if you save it as
shorts, I belive the conversion from float to short translates
them to -32767 and they are stored as such (which is presumably
not what you want).

In any case, it is probably not good to have NaN in your input
file. If you want to remove them, you can do so by:

sed 's/NaN/0.0/g' INPUT.txt > OUTPUT.txt

Note that AFNI programs do not tend to treat NaN in any special
way, except for where it either complains about them, or it
zeros them out.

----

The ANOVA program do not check for cases where say, "only 6 of
the 8 input datasets have a non-zero value". Degrees of freedom
are assigned for entire tests, not voxel-by-voxel. So these
programs do not have the ability to do what you are suggesting,
as far as I know.

- rick