Problems running 3dttest++ with ETAC and covariates

AFNI version info (afni -ver): Version AFNI_23.3.05 'Septimius Severus'

I ran a 3dttest++ with ETAC and the ETAC masks reflect the activity in the subbrick corresponding to the t-statistic map for the mean effect, not the t-statistic map for the covariate. I want to apply ETAC to the map that shows me the voxels where there is a relationship between task activity and a clinical correlate (i.e., voxels where there is a relationship between alcohol cue reactivity and subjects' AUDIT scores). Can someone please tell me how to get ETAC to apply to the covariate map?

Thanks in advance for any help!

If I remember correctly, the ETAC method is not fully implemented for the case with a quantitative variable. On the other hand, the ETAC approach is part of a broader landscape of multiple testing adjustment methods. These methods often lean toward excessive conservatism. When one rigidly applies a dichotomous threshold to continuous data, it’s akin to drawing a line in shifting sands—a nuanced challenge for robust scientific investigations. Consequently, the common practice of strict thresholding has several inherent limitations and may hinder reproducibility in neuroimaging.

As an alternative, we advocate for a "highlight-but-don’t-hide" strategy. This approach strikes a delicate balance about evidence strength, preserving information integrity while promoting transparency. Gradually, the field is embracing this perspective, as evidenced by studies like this paper published in the American Journal of Psychiatry.

Gang Chen

1 Like

Hi Gang,

Thank you for the prompt response. I plan on using the highlight-but-don't-hide strategy when displaying my figures. As it applies to testing that my clusters have survived multiple comparisons while using a covariate, should I average the smoothing parameters from the preprocessing steps and use 3dClustsim instead?

Using 3dClustSim to account for spatial smoothness is a common practice. Alternatively, you could employ a voxel-level p-value threshold (e.g., 0.01) in conjunction with a cluster-level threshold (e.g., 20 voxels), and then apply the highlight-but-don’t-hide strategy.

Gang Chen

1 Like

Excellent! Thank you for the advice!