Best practices for 3dLME - two questions for Gang :)

AFNI Message Board
1

Hi, I have another couple questions for Gang (thank you so much for answering my previous question by the way!!!)

  1. What is the best practice perform cluster correction when using 3dLME? I ran a model in 3dLME that cannot be done using the other programs, and got the residuals output using the -resid option. Should I get the ACF values from each subject’s first-level errts_REML, average those, and then feed those into 3dClustSim? Or is it okay if I just get the ACF values from the 3dLME residuals and use those for 3dClustSim?

  2. I am analyzing a fMRI dataset of twins, and was hoping to account for twin-ness by doing a random intercept for each pair. My goal is to test the significance of the relationship between a contrast (cue_negative-cue_neutral) and a factor score (quant var). I was wondering - would it be ok to make my outcome variable the first-level contrast of interest? (i.e., cue_negative-cue_neutral#Coef)? This way, I have one row per twin, and two rows for each twin pair that share the same Subj code. Or should I model the task conditions separately (e.g., cue_negative and cue_neutral) as a within-subject variable? (so, feed it cue_negative#Coef and cue_netural#Coef for each subject)? I am confused as to what are the best practices here; I know other programs do the within- and between-subjects analyses separately, but looking at the 3dLME documentation examples it seems like it’s all done in one go here in AFNI’s 3dLME.

Thank you so much!!

2

Should I get the ACF values from each subject’s first-level errts_REML, average those, and then feed those into 3dClustSim?
Or is it okay if I just get the ACF values from the 3dLME residuals and use those for 3dClustSim?

Theoretically speaking the ACF estimates from the 3dLME residuals should be fine for 3dClustSim. It would be interesting to compare the ACF estimates at the group level and the ones at the individual level.

I was wondering - would it be ok to make my outcome variable the first-level contrast of interest? (i.e., cue_negative-cue_neutral#Coef)?

If the contrast is what you’re looking for, then running the group analysis directly with the contrast is preferred, and makes everything more straightforward.

3

Thank you very much Gang!!!