I have some basic questions about the resulting file after using 3dLME or 3dMVM on my resting state data (DMN correlation maps). I know that it is possible to add covariates (within subj vars, between sub vars and quantitative vars) and labels to specify interactions . You can expect main and interaction effects from those covariates when you look at the sub-briks of 3dLME/3dMVM output.
My question is, what is the meaning of these sub-briks? Are these the correlations (main/interaction effects) between the covariates and my resting state maps (data which I am working)? or are these the results with the covariates removed?
I think that if I am looking for correlations but I previously established glt and glf labels, then I should convert those F and tstats to r scores. Is that correct?
My confusion arose because I was using 3dtest++ with covariates before. It seems that covariates in 3dtest++ and covariates in 3dLME/3dMVM work different. Am I right?
I attach an example. I have two groups (young vs old_adults). I am interested to know how to treat factors/covariates/interactions as effects of interest and also as a confounds. Is that possible?
The followings are the effects of interest/confounds that I would like to clarify:
a. Factor (gender)
b. Covariate (age)
c. Interaction between factor_covariate (genotype_and_some_measure or gender_and_age)
d. Interaction between factor-factor(genotype_and_gender)
e. Interaction between covariate-covariate (age_and_some_measure)
Are the glt Labels correct (see code bellow) basing on the effects that I established? Then, If I want to treat those as confounds, should I add more labels? If so, how to specify that in the syntax?
My last questions are:
Is -bsVars correct basing on the presented idea?
When I have two groups, should I put qVarsCenters for each one?
After looking at 3dMVM examples I am still not sure what is num_glt.
3dMVM -prefix Example -jobs 6 \
-bsVars "group*gender*genotype*age*some_measure" \
-qVars "age,some_measure"
-qVarsCenters '28.3 61.9, 1.5 1.1' (for both groups?) \
-num_glt ? \
# effects of interest:
# a. Factor (gender)
# b. Covariate (age)
# c. Interaction between factor_covariate (genotype_and_some_measure or gender_and_age)
# d. Interaction between factor-factor(genotype_and_gender)
# e. Interaction between covariate-covariate (age_and_some_measure)
-gltLabel 1 a_gender -gltCode 1 'group : 1*young -1*old gender : 1*male -1*female' \
-gltLabel 2 b_age -gltCode 2 'group : 1*young -1*old age : ' \
-gltLabel 3 c_Interaction_genotype_and_some_measure -gltCode 3 'group : 1*young -1*old genotype : 1*TT -1*NN some_measure : ' \
-gltLabel 4 d_Interaction_gender_and_genotype -gltCode 4 'group : 1*young -1*old gender: 1*male -1*female genotype: 1*TT -1*NN ' \
-gltLabel 5 e_Interaction_age_and_some_measure -gltCode 5 'group : 1*young -1*old age: some_measure: ' \
-dataTable \
Subj group gender genotype age some_measure InputFile \
Sub1 young male TT 35 1.2 suj_1_melodic_resamp.nii.gz'[0]' \
Sub2 young female NN 27 1.4 suj_2_melodic_resamp.nii.gz'[0]' \
Sub3 young female TT 25 1.3 suj_3_melodic_resamp.nii.gz'[7]' \
Sub4 young male TT 35 1.44 suj_4_melodic_resamp.nii.gz'[17]' \
Sub5 young female NN 25 1.3 suj_5_melodic_resamp.nii.gz'[8]' \
Sub6 young male NN 35 1.3 suj_6_melodic_copia_resamp.nii.gz'[0]' \
Sub7 young male TT 20 1.4 suj_7_melodic_copia_resamp.nii.gz'[0]' \
Sub8 young male NN 25 1.9 suj_8_melodic_copia_resamp.nii.gz'[7]' \
Sub9 young male TT 25 1.8 suj_9_melodic_copia_resamp.nii.gz'[17]' \
Sub10 young female NN 35 1.6 suj_10_melodic_copia_resamp.nii.gz'[8]' \
Sub11 young female TT 27 1.9 suj_11_melodic_otracopia_resamp.nii.gz'[0]' \
Sub12 young female TT 25 1.4 suj_12_melodic_otracopia_resamp.nii.gz'[0]' \
Sub13 young male TT 35 1.5 suj_13_melodic_otracopia_resamp.nii.gz'[7]' \
Sub14 young male NN 28 1.5 suj_14_melodic_otracopia_resamp.nii.gz'[17]' \
Sub15 young female TT 25 1.6 suj_15_melodic_otracopia_resamp.nii.gz'[8]' \
Sub16 old male NN 80 1.0 suj_16_melodic_ND_resamp.nii.gz'[3]' \
Sub17 old female NN 75 1.1 suj_17_melodic_ND_resamp.nii.gz'[16]' \
Sub18 old female NN 68 1.1 suj_18_melodic_ND_resamp.nii.gz'[5]' \
Sub19 old male TT 68 1.1 suj_19_melodic_ND_resamp.nii.gz'[18]' \
Sub20 old female NN 72 0.6 suj_20_melodic_ND_resamp.nii.gz'[1]' \
Sub21 old male NN 68 0.7 suj_21_melodic_ND_copia_resamp.nii.gz'[3]' \
Sub22 old female TT 80 1.1 suj_22_melodic_ND_copia_resamp.nii.gz'[16]' \
Sub23 old male NN 75 1.2 suj_23_melodic_ND_copia_resamp.nii.gz'[5]' \
Sub24 old male TT 72 1.23 suj_24_melodic_ND_copia_resamp.nii.gz'[18]' \
Sub25 old female NN 80 1.4 suj_25_melodic_ND_copia_resamp.nii.gz'[1]' \
Sub26 old female TT 51 1.0 suj_26_melodic_ND_otracopia_resamp.nii.gz'[3]' \
Sub27 old female NN 75 1.1 suj_27_melodic_ND_otracopia_resamp.nii.gz'[16]' \
Sub28 old male TT 75 1.1 suj_28_melodic_ND_otracopia_resamp.nii.gz'[5]' \
Sub29 old male NN 68 1.2 suj_29_melodic_ND_otracopia_resamp.nii.gz'[18]' \
Sub30 old female TT 72 1.3 suj_30_melodic_ND_otracopia_resamp.nii.gz'[1]'
I am interested to know how to treat factors/covariates/interactions as effects of interest and also as a confounds. Is that possible?
This is a too often a misconception. The model does not care which variables (or effects) you want to focus on when reporting the results. In other words, it is the way you formulate/construct the model, not your interest, that matters.
Are the glt Labels correct (see code bellow) basing on the effects that I established?
A label is just a name. It is the way you specify each comparison that matter. For example,
Dear Gang,
thank you very much for your reply.
So, every time I formulate a contrast, the model will provide me with the result of that particular contrast after eliminating the effect of the remaining factors/variables included in the model. Is that correct?
every time I formulate a contrast, the model will provide me with the result of that particular contrast after
eliminating the effect of the remaining factors/variables included in the model. Is that correct?
Karel, the word “eliminating” is inaccurate to describe the situation to say the least. In fact, you cannot “eliminate” the effect of any variable in the model; instead, you can control that variable (e.g. in those GLT specifications in 3dMVM or 3dLME as you’re asking here) at a particular value (e.g., center or mean if it’s a quantitative variable) or the average across all levels (if it’s a factor). I hope this clarifies the situation.
Thank you so much Gang! after clarifying the inaccurate term “eliminating”, the main idea is starting to make sense for me.
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.