Gang,
Just to follow-up, I noticed something with the model I posted for 3dLME. I think the model "time*score" problem is having the same values for the 1st level of time in my model. If I change subject 1's value to 1 and keep the remaining values 0 for level one of Time, then 3dLME executes.
For example:
This works
3dLME -prefix LME_output -jobs 7 \
model "time*score" \
-qVars "score" \
-SS_type 3 \
-ranEff '~1' \
-mask mask+tlrc \
-dataTable \
Subj time score InputFile \
s1 one 1 Time-sub01-a+tlrc \
s2 one 0 Time-sub02-a+tlrc \
s3 one 0 Time-sub03-a+tlrc \
s4 one 0 Time-sub04-a+tlrc \
s5 one 0 Time-sub05-a+tlrc \
s6 one 0 Time-sub06-a+tlrc \
s7 one 0 Time-sub07-a+tlrc \
s8 one 0 Time-sub08-a+tlrc \
s9 one 0 Time-sub09-a+tlrc \
s10 one 0 Time-sub10-a+tlrc \
s11 one 0 Time-sub11-a+tlrc \
s12 one 0 Time-sub13-a+tlrc \
s13 one 0 Time-sub14-a+tlrc \
s14 one 0 Time-sub16-a+tlrc \
s1 two 0.5914 Time-sub01-b+tlrc \
s2 two 0.3432 Time-sub02-b+tlrc \
s3 two 0.632 Time-sub03-b+tlrc \
s4 two 0.0671 Time-sub04-b+tlrc \
s5 two 0.2778 Time-sub05-b+tlrc \
s6 two 0.3944 Time-sub06-b+tlrc \
s7 two 0.8301 Time-sub07-b+tlrc \
s8 two 0.7662 Time-sub08-b+tlrc \
s9 two 0.6289 Time-sub09-b+tlrc \
s10 two 0.9613 Time-sub10-b+tlrc \
s11 two 0.8316 Time-sub11-b+tlrc \
s12 two 0.2091 Time-sub13-b+tlrc \
s13 two 0.8233 Time-sub14-b+tlrc \
s14 two 0.8609 Time-sub16-b+tlrc \
s1 three 0.7056 Time-sub01-c+tlrc \
s2 three 0.2488 Time-sub02-c+tlrc \
s3 three 0.6336 Time-sub03-c+tlrc \
s4 three 0.4584 Time-sub04-c+tlrc \
s5 three 5.9936 Time-sub05-c+tlrc \
s6 three 7.1744 Time-sub06-c+tlrc \
s7 three 3.908 Time-sub07-c+tlrc \
s8 three 1.208 Time-sub08-c+tlrc \
s9 three 5.1216 Time-sub09-c+tlrc \
s10 three 1.136 Time-sub10-c+tlrc \
s11 three 7.9072 Time-sub11-c+tlrc \
s12 three 4.968 Time-sub13-c+tlrc \
s13 three 1.2136 Time-sub14-c+tlrc \
s14 three 3.4784 Time-sub16-c+tlrc
So, is this a bug or a statistical issue of the model? If I pull data from one voxel and run the same model in SPSS I can get main effects for score, time, and score*time interaction.
Big picture wise, the concern is not having the full model vs. only having the interaction term in the model. With ANOVA's you usually get main effect and interaction terms from your result.
Thanks.
Michael
