Hi Alfonso and CONN forum members:

I am having some conceptual issues with setting up my second-level analyses, and I'm hoping to receive your advice as to what the best setup would be.
If you don't mind considering my issue, I will first explain my dataset, my setup, and then my questions.

Dataset: Resting-state data of a sample (N=150) with an autism group (ASD), control group (TYP), and a personality behavioral measure (e.g., Neuroticism scored from 1 - 50). I am looking to identify significant associations between Neuroticism scores and ROI-ROI connections in the ASD group (followed by the same analysis for the TYP group afterwards).

Setup:

• AllSubjects = [1 for all 150 participants]; total of 150 values
• ASD = [1 for all 73 participants in the ASD group, 0 for all non-ASD participants]; total of 150 values
• TYP = [1 for all 77 participants in the TYP group, 0 for all non-TYP participants]; total of 150 values
• N_ASD = [Neuroticism values for all ASD participants, 0 for all non-ASD], e.g., [0 0 35 0 42 ...]; total of 150 values
• N_TYP = [Neuroticism values for all TYP participants, 0 for all non-TYP], e.g., [25 20 0 18 0 ...]; total of 150 values
• QC_MeanMotion = [mean motion values for all 150 participants]; total of 150 values

Questions:

• If I wanted to evaluate significant associations of Neuroticism with ROI-ROI connection in the ASD group only (i.e., a regression analysis for a single group), would it make more sense to have a setup of [AllSubjects 0 N_ASD 1] or [ASD 0 N_ASD 1]? Mathematically, they seem equivalent to me since N_ASD provides "0"s for the TYP group anyways, but I'm not completely sure what the difference is. There are other combinations that might make sense (for example, if I collapsed N_TYP and N_ASD into a single Neuroticism covariate with a value for each participant, I could potentially run [ASD 0 Neuroticism 1]?), but again, this is my main conceptual issue with the setup.
• To further consider this example, if I wanted to control for mean motion, I would add in the covariate, e.g., [AllSubjects/ASD 0 N_ASD 1 QC_MeanMotion 0]. However, should the QC_MeanMotion be separated into a QC_MeanMotion_ASD and a QC_MeanMotion_TYP, in which I would use QC_MeanMotion_ASD (a covariate that has "0"s for all TYP participants)?

Thank you very much for taking the time to consider my issue, I look forward to people's thoughts.

Billy

Alfonso, to partially answer my own question, it seems I found my conceptual error. In my setups using AllSubjects, the analysis includes all 150 participants, even when using a covariate that has 0 for the group not of interest. In order to use AllSubjects, those same 0s in the covariate (i.e., Neuroticism) needs to be written in as NaN. Conversely, I can use a Neuroticism covariate that includes both ASD and TYP values, but the subject variable (i.e., ASD) needs to include NaN for non-ASD participants. Put another way, only when there are 0s across all variables is a participant excluded from analyses.

To answer my question 1, the most appropriate analysis is [ASD 0 N_ASD 1]. For question 2, QC_MeanMotion needs to be split into 2 variables (QC_MeanMotion_ASD and QC_MeanMotion_TYP) to ensure that the regression analyses only include the appropriate number of participants.

I got this information from some threads back in 2015. Thank you for those old answers. If you do come across this post, could you please confirm what I've said? Thank you so much!

Billy

Hi Billy,

Yes, exactly on both accounts

a) selecting 'ASD' and 'N_ASD' and entering a [0 1] contrast would be the appropriate way to look at the association between Neuroticism and connectivity within the ASD group only (and the ASD and N_ASD variables may have 0's for the non-ASD subjects, which will be automatically excluded from this analysis)

b) selecting 'ASD', 'N_ASD', and 'QC_MeanMotion_ASD' and entering a [0 1 0] contrast would do the same analyses while controlling for the effect of motion across subjects (and the QC_MeanMotion_ASD' variable can be computed from your original QC_MeanMotion variable as 'QC_MeanMotion * ASD')

In both cases, you may check the 'n=...' text in the second-level results tab to confirm that the number of subjects included into your analysis is as expected.

Best
Alfonso
Originally posted by An Chuen Cho: