3dMVM with multiple covariates?

AFNI Message Board
1

Hi AFNI gurus,

I cannot get 3dMVM running using two covariates - I am sure I am doing something wrong. I think the problem stems from attempting to use two covariates in the gltCode - can such a thing be done? I’ve tried both glt and glfCode. I realize that my gltCode tests are t-tests which could be done with 3dttest++, but I have many such tests to run and was trying to not conduct too many t-tests.

I’d appreciate any input.
-n

Command I’m attempting to run:


3dMVM -prefix MVM_FUMC_10 -jobs 10 -mask Intersection_GM_mask+tlrc \
-bsVars 'Group*Snif+Group*SPA' \
-wsVars Stim \
-qVars 'Snif,SPA' \
-num_glt 4 \
-gltLabel 1 G_Mask -gltCode 1 'Group : 1*Con -1*Aut Stim : 1*Mask Snif : SPA : ' \
-gltLabel 2 G_FBO -gltCode 2 'Group : 1*Con -1*Aut Stim : 1*FBO Snif : SPA : ' \
-gltLabel 3 G_UBO -gltCode 3 'Group : 1*Con -1*Aut Stim : 1*UBO Snif : SPA : ' \
-gltLabel 4 G_CA -gltCode 4 'Group : 1*Con -1*Aut Stim : 1*CA Snif : SPA : '  \
-dataTable \
Subj Group Stim Snif SPA InputFile \
sub-606 Aut Mask 13 46 sub-606/PPI_FUMC_10_stats_REML+tlrc'[29]' \
sub-606 Aut FBO 13 46 sub-606/PPI_FUMC_10_stats_REML+tlrc'[32]' \
sub-606 Aut UBO 13 46 sub-606/PPI_FUMC_10_stats_REML+tlrc'[35]' \
sub-606 Aut CA 13 46 sub-606/PPI_FUMC_10_stats_REML+tlrc'[38]' \
sub-608 Aut Mask 11 42 sub-608/PPI_FUMC_10_stats_REML+tlrc'[29]' \
sub-608 Aut FBO 11 42 sub-608/PPI_FUMC_10_stats_REML+tlrc'[32]' \
sub-608 Aut UBO 11 42 sub-608/PPI_FUMC_10_stats_REML+tlrc'[35]' \
sub-608 Aut CA 11 42 sub-608/PPI_FUMC_10_stats_REML+tlrc'[38]' \
...
sub-1138 Con Mask 13 40 sub-1138/PPI_FUMC_10_stats_REML+tlrc'[29]' \
sub-1138 Con FBO 13 40 sub-1138/PPI_FUMC_10_stats_REML+tlrc'[32]' \
sub-1138 Con UBO 13 40 sub-1138/PPI_FUMC_10_stats_REML+tlrc'[35]' \
sub-1138 Con CA 13 40 sub-1138/PPI_FUMC_10_stats_REML+tlrc'[38]'

Error message:


** Error:
   Inappropriate coding in test No.1!

Warning messages:
1: In if (is.na(code[[n]][QVpos + 2])) { :
  the condition has length > 1 and only the first element will be used
2: In if (QVpos == 1) outList[[1]][[n]] <- NA else outList[[1]][[n]] <- glfConstr(code[[n]][-(QVpos:(QVpos +  :
  the condition has length > 1 and only the first element will be used
3: In QVpos:(QVpos + 2) :
  numerical expression has 2 elements: only the first used
4: In QVpos:(QVpos + 2) :
  numerical expression has 2 elements: only the first used
5: In gl_Constr(lop$num_glt, lop$gltCode, lop) :
  NAs introduced by coercion

AFNI version:


afni -ver
Precompiled binary linux_openmp_64: Dec  5 2018 (Version AFNI_18.3.15 'Augustus')
2

All those 4 tests coded through -gltCode don’t make sense. Could you explain exactly what you’re looking for?

3

I have a dataset where there are two groups (Control, Autism) who were exposed to four stimuli during scanning (Mask, FBO, UBO, CA). For each participant we also have quantitative measurements from two tests (Snif, SPA) that we would like to use as covariates in the group-level analysis.

This is actually for a PPI analysis where we want to look at group differences in the correlation matrix with 8 different seeds for each of those stimuli, all while controlling for those covariates (the code in my previous post is for one of the seeds). This entails many analyses and I’m worried about multiple comparisons, so I’m trying to do this in 3dMVM rather than 3dttest++.

-n

4

we want to look at group differences in the correlation matrix with 8 different seeds
for each of those stimuli, all while controlling for those covariates

It sounds like the following is what you want if you center those two covariates properly:

-gltLabel 1 G_Mask -gltCode 1 ‘Group : 1Con -1Aut Stim : 1Mask’
-gltLabel 2 G_FBO -gltCode 2 'Group : 1Con -1Aut Stim : 1FBO’
-gltLabel 3 G_UBO -gltCode 3 ‘Group : 1Con -1Aut Stim : 1UBO’
-gltLabel 4 G_CA -gltCode 4 'Group : 1Con -1Aut Stim : 1CA’ \

If you add a covariate in the -gltCode specification, it means that you want to get the slope (covariate) effect.

5

Ah, thank you.
That resolved my problem, I misunderstood the documentation of 3dMVM, example 2.
And I found the AFNI Section 7.1 on centering covariates, so I’m sure I can figure the rest out.

Thanks again.

-n