3dMVM omnibus F-test with 3 groups

AFNI Message Board
1

Hi,

Im trying to compare FC of 3 groups using 3dVMV. I tried using this glfLabel below for the F-test, but get the following error: Error in glfRes[[ii]]Df : operator is invalid for atomic vectors.

Heres a portion of my script. Could you let me know where the error is?

Thank you.

-bsVars 'group+age+sex+FD' \
-qVars 'age,FD' \
-SS_type 2 \
-num_glf 1 \
-glfLabel 1 'omnibusF' -glfCode 1 'group : 1*short -1*intermediate & 1*short -1*long & 1*intermediate -1*long' \
-resid /resid_3MVM_DMN.nii.gz \
-dataTable \
Subj group age sex FD InputFile \
Subject124 short 30 2 0.195394 /BETA_Subject124_Condition001_Source004.nii \
........
Subject127 long 22 2 0.113489 /BETA_Subject127_Condition001_Source004.nii \
........
Subject130 intermediate	19 1	0.1608 /BETA_Subject130_Condition001_Source004.nii \
2

3dMVM automatically provides an omnibus F-statistic for the comparisons among the three groups. In addition, your -glfCode specification is rank deficient - only two of the three are needed.

3

I see. So, If I remove one of the three contrasts it would show me a similar omnibus test? I could use gltCode to compare two groups as well like a post-hoc test, right?

Thank you!

Gang Wrote:

3dMVM automatically provides an omnibus
F-statistic for the comparisons among the three
groups. In addition, your -glfCode specification
is rank deficient - only two of the three are
needed.

4

If I remove one of the three contrasts it would show me a similar omnibus test?

Yes, it should.

I could use gltCode to compare two groups as well like a post-hoc test, right?

You can specify the omnibus test for the three groups using glfCode:

-glfLabel 1 'omnibusF' -glfCode 1 'group : 1*short -1*intermediate & 1*short -1*long' \
5

Hi, I didn't want to post a new message so just piggy backing from my older post:
If I have 4 groups, e.g., HC, exp1, exp2, and exp3 and wanted to do an omnibus test then I would do

-gltLabel 1 'omnibusF' -gltCode 1 'group : 1*HC -1*exp1 & 1*HC -1*exp2 & 1*HC -1*exp3' \

Is this correct?

6

The test should be specified using the -glfCode option as follows:

-glfLabel 1 'omnibusF' -glfCode 1 'group : 1*HC -1*exp1 & 1*HC -1*exp2 & 1*HC -1*exp3' \

Gang Chen

7

oh yes, that was a typo. So, this orthogonal contrast will give me an estimated overall difference among all groups then?
Thank you.

8

Such an omnibus test evaluates the overall differences across all groups, but it does not specify between which groups those differences occur.

Gang Chen