3dMVM: glt vs. glf

AFNI Message Board
1

Hello,

I have a question about running 3dMVM. I would like to run a 3 Group x 2 Phase x 2 Reinforcement repeated measures ANOVA and I am not sure whether I should specify “glt” or “glf” -what is the difference between the two? Are “glf” only for variables that have more than 2 factors?

I have provided my code and a breakdown of my ANOVA below.

Thank you,
Tamara

ANOVA:
3 Group (control vs. temporal vs. frontal) x 2 Phase (acquisition vs. reinforcement) x 2 Reinforcement (correct vs. incorrect)

CODE:

3dMVM -prefix group -jobs 8
-bsVars group
-wsVars "phasereinf"
-SS_type 3
-num_glt 3
-gltLabel 1 phase -gltCode 1 'phase : 1acq -1rev’
-gltLabel 2 reinf -gltCode 2 'reinf : 1corr -1incor’
-gltLabel 3 phase_reinf -gltCode 3 'phase : 1acq -1rev reinf : 1corr -1incor’
-num_glf 4
-glfLabel 1 group -glfCode 1 'group : 1ctrl -1temp & 1ctrl -1front’
-glfLabel 2 group_phase -glfCode 2 'group : 1ctrl -1temp & 1ctrl -1front phase : 1acq -1rev’
-glfLabel 3 group_reinf -glfCode 3 'group : 1ctrl -1temp & 1ctrl -1front reinf : 1corr -1incor’
-glfLabel 4 3way_inter -glfCode 4 'group : 1ctrl -1temp & 1ctrl -1front phase : 1acq -1rev reinf : 1corr -1*incor’
-dataTable
Subj group phase reinf InputFile
8225 temp acq corr 8225/Proc_FIRST_8225_April2019.results/8225_RevP_acq_corr_fed+tlrc
…
…
…

2

Are “glf” only for variables that have more than 2 factors?

Yes, that’s usually the case. However, you don’t really have to perform a GLF test. For example, the following

-glfLabel 1 group -glfCode 1 ‘group : 1ctrl -1temp & 1ctrl -1front’ \

is actually redundant because 3dMVM automatically provide the statistic for it, and it can be more informatively replaced by two separate GLTs:

-gltLabel ? group -gltCode 2 ‘group : 1ctrl -1temp’
-gltLabel ? group -gltCode 2 ‘group : 1ctrl -1front’ \

3

great, thank you very much.

In this case, would it be incorrect to use glt with 3 factors? The example for “-glfCode K CODING” section in the AFNI 3dMVM -help contains an example for a main effect of group.

Thank you again!

4

would it be incorrect to use glt with 3 factors?

No, there is nothing wrong with your specification because it’s indeed a t-test. For example,

-glfLabel 4 3way_inter -glfCode 4 ‘group : 1ctrl -1temp & 1ctrl -1front phase : 1acq -1rev reinf : 1corr -1incor’ \

would test the same hypothesis as the three-way interaction F=test that 3dMVM automatically provides except that the F-stat is the corresponding t-stat squared.