> Is there another way to construct the linear trend without excluding the middle term? Could using 3dRegAna work?
No, forget about 3dRegAna. 3dLME might be a better solution. First, code "condition" as a quantitative variable:
one -> -1
two -> 0
three -> 1
Then run 3dLME:
3dLME -prefix ... -jobs 12 -model "condition" -qVars "condition" -ranEff '~condition' \
-gltLabel ... -gltCode ... \
-dataTable \
Subj condition InputFile \
s1 -1 s1_A+tlrc \
s1 0 s1_B+tlrc \
s1 -1 s1_C+tlrc \
...
Gang
