I was interested in finding out how to save out the residuals of my 3dMVM model. It this possible to do (similar to the -resid in 3dttest++)? I am performing a seed based resting state correlation analysis using age, gender, group and MOCA as between group covariates and do no have any within subject covariates. If it is not possible within 3dMVM, is there another model that is more appropriate for this analysis (ie 3dLME since 3dttest++ won’t work in this situation)? Finally, if no residuals can be output by these models, what would be the best “noise” estimate for 3dFWHMx (ie the average of each individuals epits/errts.anaticor files even though it is technically not noise or something else)?

Ideally I want to use this noise/residual term from 3dMVM as an input to 3dFWHMx with the -acf option in order to properly estimate my smoothness, as underlying spatial stucture from 3dMVM output may incorrectly estimate my smoothness. This output would be fed into 3dClustSim in order to estimate the clusterwise extent for given P values.

I recently saw an article by Nichols and colleagues which evaluates the validity of clusterwise estimates in FMRI. The conclusion was that alpha <0.05 is not correctly assessed for any p>0.01. Due to limited number of subjects in a study, I could see where using a more liberal p<0.01 and larger cluster extent is appealing. Does the -acf option in 3dClustSim/3dFWHMx address issues raised in this article so that cluster extents calculated at p<0.01 are indeed accurate or do you recommend going to a p<0.005/0.001 in order to address these limitations for limited N (ie 20 per group).

am performing a seed based resting state correlation analysis using age, gender, group and MOCA
as between group covariates and do no have any within subject covariates.

In that case, you can directly use 3dttest++ instead of 3dMVM by coding all those between-subjects variables as covariates. See the second half of the following page for coding categorical variables in 3dttest++:

Does the -acf option in 3dClustSim/3dFWHMx address issues raised in this article so that cluster
extents calculated at p<0.01 are indeed accurate or do you recommend going to a p<0.005/0.001
in order to address these limitations for limited N (ie 20 per group).

Use the new option -Clustsim in 3dttest++, and it should properly handle FWE correction. Make sure that you have the most recent version of AFNI.

while we are on this topic, is there currently a robust way to handle FWE correction on the results of a one-way ANOVA? That is, I cannot reformulate the contrast to be handled by 3dttest++.

Such a one-way within-subject ANOVA boils down to three pairwise comparisons. For each of three pairwise comparisons, you can run a paired t-test with 3dttest++ -Clustsim.

Hi Gang,
Thanks for the help, I’ll try that out! In a different study (part of which is surface based and part volumetric) I needed to use 3dMVM because I have interaction terms that I cannot model in 3dttest++. Additionally I wanted the f-test.

Is there any easy way to get residuals from 3dMVM or would the code need to be modified?

Is there any issues with slow_surf_clustsim.py or has this been updated in the most recent version?

Is the most recent version of 3dClustSim fixed or is it only fixed in the ClustSim option within 3dttest++

Hi Gang,
Thanks for the update. As I see it, currently I have three options (though I’m sure this will change as this topic evolves)

I was able to use the ACF option in afni_proc.py by passing it through the 3dBlurToFWHM options for a resting state analysis. In regards to final cluster size (3dClustSim simulation) is it better to

a) estimate to ACF parameters (3dFWHMx) by calculating residuals from the 3dMVM output stats (or in one analysis I can run 3dttest++), one for each comparison.

b) estimate the 3dclustsim -acf parameters from the errts.fanaticor+orig file (technically not noise since its resting state output) using 3dFWHMx -acf?

c) estimate the 3dclustsim -acf parameters from all.runs -errts.fanaticor+orig which should represent first level residuals if it is too difficult to obtain from 3dMVM?

I guess I am not sure if estimates from input files are valid for clusterwise thresholding or if it HAS to be from the output stats files (ie 3dMVM or 3dttest++ residuals).

If you can fit your group model with 3dttest++, the option -Clustsim is currently the most rigorous FWE correction approach. 3dMVM does not output the residuals yet. In regard to the ACF parameters, the average values across all runs/subjects seem to be a reasonable option.

Hi Gang, I found your replies to this post from several years ago. I am wondering if you could shed some light on my problem using 3dMVM -resid in the post below.

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.