cross validated MANOVA via 3dMVM

AFNI Message Board
1

Dear afni experts,
Can 3dmvm perform the cross-validated MANOVA as described in this paper https://www.sciencedirect.com/science/article/pii/S1053811913011920 ?
Thanks,
Best
Rujing

2

Rujing,

I assume that the cross-validated MANOVA discussed in that paper is in spatial context (i.e., across voxels in the brain). The multivariate modeling implemented in 3dMVM is meant for the levels of a within-subject (or repeated-measures) factor, so it would not work for cross-validated MANOVA in its current form.

3

Hi Gang
I see. However, in https://afni.nimh.nih.gov/pub/dist/edu/latest/afni_handouts/FATCAT_04_netw_stats_mvm.pdf, Page 9, it shows

  1. Network-level test:
    multivariate model (MVM)
    {FA1 , FA2, FA3, …}~ var1 + var2 + var3 …
    ======
    Here is my idea about this test: multi-X was used to test the effect on the multi-Y. And the ys are brain connectivity acquired from the DTI data. It should be similar to voxel by voxel activity.
    If this idea was correct, then i can use 3dmvm to perform {voxels by voxels activity} ~ var1 +var2 +var3. If var1 showed significant effect on the voxels-by-voxels, then I can say multi-voxels contained information to encode the var1.
    If it is still OK, then I need to know how to calculate the cross validation for multi-variate ANOVA?
    Maybe I am crazy, as I really want to perform cvMANOVA via AFNI program.
    rujing
4

Rujing,

In the case of DTI data, the number of white matter properties are quite limited. However, the number of voxels is quite substantial; in addition, 3dMVM in its current form is not suited for handling the spatial structures among the huge number of voxels in the brain. The underlying modeling mechanism may work for cvMANOVA, but it needs some thinking and coding work.

5

Hi Gang
Ok, I see.
Looking forward some update in the machine learning, as 3dsvm is not enough.
Thanks,
Best.
Rujing