Group Analysis: Single Subject Pthr?

AFNI Message Board
1

Hello,

I would like to run 3dttest++ and perform a 1-sample test. I might be missing it in the notes, but is there a way to indicate the single subject Pthr to run through group level analysis? That is, only include voxels that are significant at the single level in the group level analysis?

The only way I can think to do this is to save the 3dclusterize map (-pref_dat) to extract the significant voxels at the single subject level and then use that file for group analysis.

I’m not sure if I’m overthinking this, but it doesn’t seem accurate to run a group level analysis on a non-thresholded single subject file. If I did, I may see significant clusters at the group level simply because the t-stat (or beta) is consistently greater, but still not large enough to reach significance on their own.

2

is there a way to indicate the single subject Pthr to run through group level analysis? That is, only include voxels
that are significant at the single level in the group level analysis?

Would this be like cherry-picking?

it doesn’t seem accurate to run a group level analysis on a non-thresholded single subject file.

In what sense it’s not accurate?

If I did, I may see significant clusters at the group level simply because the t-stat (or beta) is
consistently greater, but still not large enough to reach significance on their own.

Significant or not, it is just an artificial line drawn in the sand that depends on many factors including experimental design, sample sizes and even randomness.

3

Here’s a hypothetical situation and please let me know if I’m thinking about it the wrong way:

Say I’m running group stats on the T-Statistical map. For one group the Average T-Stat for a cluster is 0.5 and for the other group the average T-stat for the cluster is 1.5 (very little variance). Wouldn’t that cluster be considered significant despite either cluster reaching a p value < .05 at the single subject level?

Wouldn’t it be appropriate to treat the single-subject analysis as almost binary meaning you either have significant activation or don’t? That way you would only pass the significant clusters to the group level?

The same example could be used if I did the group analysis on the beta, where I would only pass ‘significant’ beta clusters to the group level.