> -bsVars 'GROUP+age' -qVars 'age' \
> -num_glt 1 \
> -gltLabel 1 MRNvsPSU -gltCode 1 'GROUP : 1*MRN -1*PSU age :' \

Two issues here:

1) The two groups in the GLT coding are labeled as MRN and PSU, but they are called Group1 and Group2 in the data table.

2) With {-bsVars 'GROUP+age' -qVars 'age'}, you assume that the two groups have the *same* age effect. However, under such an assumption it does not make sense to test the age effect difference between the two groups with {-gltLabel 1 MRNvsPSU -gltCode 1 'GROUP : 1*MRN -1*PSU age :'}. Change the between-subjects variable specification to

-bsVars 'GROUP*age' -qVars 'age'

> the large discrepancy in variance between the two groups (group1 ages 18-22; group2 ages 18-50ish)

By the default 3dMVM centers a covariate around the overall mean. If the overall mean age is something like 32, it could be problematic because you don't even have any subjects around that age in Group1. If you prefer to interpret the group effect (e.g., group difference) at a different age (e..g, 22), use option -qVarCenters

Gang



Edited 1 time(s). Last edit at 08/14/2014 04:13PM by Gang.