Hi- I’m trying to determine which group analysis to run for an ANCOVA testing for differences between 4 uneven groups of participants (dummy-coded 1, 2, 3, & 4) after controlling for several covariates (e.g., race as a categorical variable, total gray matter volume as a quantitative variable). As my dependent variable, I have stats contained in .nii.gz files (concerning warped gray matter from a VBM analysis). I’ll be masking my data to test for differences within a region.
3ttest++ using a mask & a covariates file seemed the way to go, at first. However, would this analysis treat my dummy-coded grouping variable as a quantitative, not categorical, variable?
Any guidance is appreciated! -Robie
It’s possible to analyze the model with 3dttest++, but it would involve meticulous specifications with all the variables. An easier approach would be using 3dMVM by treating all of your variables as either between-subjects or quantitative. Forget about dummy-coding, and let 3dMVM handle it for you.
Hi, Gang- Thanks for your advice. Shifting to 3dMVM seems clear, but I’m not sure about how to set up the GLT for getting the main effect of group (categorical variable “cmar”). There are 4 uneven groups, & I don’t think I want to combine or weight them in any way bc each group is unique. Consistent w/ your suggestion, “cmar” is between-subjects, & covariates, “black” & “spmtot,” are quantitative.
Or maybe I don’t need to use a GLT to get the main effect of group?
Below is my in-progress script. The “???” shows where I’m confused. Please let me know if anything else seems amiss! Thanks again! -Robie
3dMVM -prefix 3dMVM_cmar_parsOper
-gltLabel 1 cmar -gltCode 1 ‘cmar : ???’
subj cmar black spmtot InputFile \
mwp110099_t1 nevlo 1 0.51040 mwp110099_t1w_FSL_blur.nii.gz
mwp110188_t1 nevlo 0 0.50336 mwp110188_t1w_FSL_blur.nii.gz
mwp112718_t1 earlo 1 0.51168 mwp112718_t1w_FSL_blur.nii.gz
mwp110055_t1 latin 1 0.54719 mwp110055_t1w_FSL_blur.nii.gz
mwp110138_t1 latin 1 0.52185 mwp110138_t1w_FSL_blur.nii.gz
mwp112911_t1 earhi 0 0.53042 mwp112911_t1w_FSL_blur.nii.gz
mwp172133_t1 earhi 0 0.52186 mwp172133_t1w_FSL_blur.nii.gz
You labeled ‘black’ as a quantitative variable with values of 0 and 1?
-gltLabel 1 cmar -gltCode 1 ‘cmar : ???’ \
I assume that cmar is a between-subjects factor. You can compare two groups nevlo and latin as the following
-gltLabel 1 nevlo-latin -gltCode 1 ‘cmar : 1nevlo -1latin’ \
@black: Ah, perhaps I should clearly make “black” a categorical variables w/ values “black” & “white” to let 3dMVM handle it better, like you mentioned?
@cmar: Yes, “cmar” is the between-subjects factor. Thank you for the example of comparing 2 of these 4 groups to each other. I think where I get confused is, what if my interest is in determining where the main effect of group is overall? If I found a cluster (or few) where the main effect of group emerged, my plan would be to extract the values there to further probe the group differences.
Let me know if I can clarify! Thank you, -Robie
-bsVars “cmar” \
-bsVars “cmar+black+spmtot” \
-bsVars “cmar*black+spmtot” \
or even with more interactions, depending on your assumptions.
perhaps I should clearly make “black” a categorical variables w/ values “black” & “white” to let 3dMVM handle it better, like you mentioned?
At least this part "-qVars “black” in your script is troubling! You don’t have to change the coding of 0/1 for ‘black’ as long as you don’t declare it as a quantitative variable. It might be awkward if you use 0/1 to code those contrasts through -gltCode because it looks confusing with the weights.
Hi, Gang- Thanks! Your corrections for the “-bsVars” were super clarifying. & you’re right – why’d I list a dummy-coded variable under “-qVars”?!
To make sure I interpreted the output properly: Is it correct that the overall effect of group (here, “cmar,” w/ 4 different categories) is shown by setting both the OLay & Thr in the viewer as “#1 cmar F”?
Now moving on to FDR cluster correction, just want to make sure I’ve gotten this all right! Thanks again for all your help, -Robie