help
488 Subscribers 
help > Controlling for motion differences between two groups in a sample
Showing 1-4 of 4 posts
Jun 22, 2020 08:06 PM | An Chuen Cho
Controlling for motion differences between two groups in a sample
Hello CONN Forum and Alfonso,
Apologies if this has been asked before (couldn't find this topic after a thorough search).
In my ROI-ROI analysis, I have two groups in the sample - an autism diagnosis group and a control group. After removing cases with excessive outlier volumes (>50% outlier volumes), these two groups still have a significant difference (t-test) in both outlier volumes and mean motion (CONN's composite motion). I've seen only one way to control for this difference, which is Saad et al., 2013's global correlation regressor. What are some possible/recommended approaches in handling this difference in mean motion? Looking forward to everyone's answer, thank you in advance!
Sincerely,
Billy
Apologies if this has been asked before (couldn't find this topic after a thorough search).
In my ROI-ROI analysis, I have two groups in the sample - an autism diagnosis group and a control group. After removing cases with excessive outlier volumes (>50% outlier volumes), these two groups still have a significant difference (t-test) in both outlier volumes and mean motion (CONN's composite motion). I've seen only one way to control for this difference, which is Saad et al., 2013's global correlation regressor. What are some possible/recommended approaches in handling this difference in mean motion? Looking forward to everyone's answer, thank you in advance!
Sincerely,
Billy
Jun 27, 2020 12:06 PM | Alfonso Nieto-Castanon - Boston University
RE: Controlling for motion differences between two groups in a sample
Hi Billy,
Yes, in addition to including Saad's Global Correlation measure (in CONN this is available in the QC_GCOR second-level covariate created during denoising), you may also include any additional motion-related covariates (e.g. QC_MeanMotion, QC_MaxMotion) in your second-level analyses in order to further remove group-differences in connectivity that may be better explained by between-subject differences in any of these motion-related metrics
Hope this helps
Alfonso
Originally posted by An Chuen Cho:
Yes, in addition to including Saad's Global Correlation measure (in CONN this is available in the QC_GCOR second-level covariate created during denoising), you may also include any additional motion-related covariates (e.g. QC_MeanMotion, QC_MaxMotion) in your second-level analyses in order to further remove group-differences in connectivity that may be better explained by between-subject differences in any of these motion-related metrics
Hope this helps
Alfonso
Originally posted by An Chuen Cho:
Hello CONN Forum and Alfonso,
Apologies if this has been asked before (couldn't find this topic after a thorough search).
In my ROI-ROI analysis, I have two groups in the sample - an autism diagnosis group and a control group. After removing cases with excessive outlier volumes (>50% outlier volumes), these two groups still have a significant difference (t-test) in both outlier volumes and mean motion (CONN's composite motion). I've seen only one way to control for this difference, which is Saad et al., 2013's global correlation regressor. What are some possible/recommended approaches in handling this difference in mean motion? Looking forward to everyone's answer, thank you in advance!
Sincerely,
Billy
Apologies if this has been asked before (couldn't find this topic after a thorough search).
In my ROI-ROI analysis, I have two groups in the sample - an autism diagnosis group and a control group. After removing cases with excessive outlier volumes (>50% outlier volumes), these two groups still have a significant difference (t-test) in both outlier volumes and mean motion (CONN's composite motion). I've seen only one way to control for this difference, which is Saad et al., 2013's global correlation regressor. What are some possible/recommended approaches in handling this difference in mean motion? Looking forward to everyone's answer, thank you in advance!
Sincerely,
Billy
Jul 7, 2020 05:07 AM | An Chuen Cho
RE: Controlling for motion differences between two groups in a sample
Hi Alfonso,
Thank you so much for taking the time to answer my question. As a follow-up, I am not sure where or how I can 'include' the QC_GCOR second-level covariate into the model. I did confirm that this 2nd-level covariate was created, as it can now be found on the Setup tab under "Covariates (2nd Level). Does that mean it's already included in the model, or do I have to select it somewhere in the Analyses (1st-level) or Results (2nd-level) tabs in order to include it into the model? For reference, I am using CONN 18.a.
Thank you so much in advance.
Sincerely,
Billy
Thank you so much for taking the time to answer my question. As a follow-up, I am not sure where or how I can 'include' the QC_GCOR second-level covariate into the model. I did confirm that this 2nd-level covariate was created, as it can now be found on the Setup tab under "Covariates (2nd Level). Does that mean it's already included in the model, or do I have to select it somewhere in the Analyses (1st-level) or Results (2nd-level) tabs in order to include it into the model? For reference, I am using CONN 18.a.
Thank you so much in advance.
Sincerely,
Billy
Jul 7, 2020 05:07 PM | Alfonso Nieto-Castanon - Boston University
RE: Controlling for motion differences between two groups in a sample
Hi Billy,
Do not recall exactly in 18a but generally if the QC_MeanMotion covariate does not show up directly in the 'subject-effects' list in the Results (2nd-level) tab, simply right-click on that same 'subject-effects' list and select the 'show secondary variables' option there to make that variable appear.
After that, in order to evaluate the differences between the two groups while controlling for the effect of QC_MeanMotion, simply select the 'Controls', 'Autism', and 'QC_MeanMotion' covariates in the 'subject-effects' list, and enter a [-1 1 0] between-subjects contrast
Best
Alfonso
Originally posted by An Chuen Cho:
Do not recall exactly in 18a but generally if the QC_MeanMotion covariate does not show up directly in the 'subject-effects' list in the Results (2nd-level) tab, simply right-click on that same 'subject-effects' list and select the 'show secondary variables' option there to make that variable appear.
After that, in order to evaluate the differences between the two groups while controlling for the effect of QC_MeanMotion, simply select the 'Controls', 'Autism', and 'QC_MeanMotion' covariates in the 'subject-effects' list, and enter a [-1 1 0] between-subjects contrast
Best
Alfonso
Originally posted by An Chuen Cho:
Hi Alfonso,
Thank you so much for taking the time to answer my question. As a follow-up, I am not sure where or how I can 'include' the QC_GCOR second-level covariate into the model. I did confirm that this 2nd-level covariate was created, as it can now be found on the Setup tab under "Covariates (2nd Level). Does that mean it's already included in the model, or do I have to select it somewhere in the Analyses (1st-level) or Results (2nd-level) tabs in order to include it into the model? For reference, I am using CONN 18.a.
Thank you so much in advance.
Sincerely,
Billy
Thank you so much for taking the time to answer my question. As a follow-up, I am not sure where or how I can 'include' the QC_GCOR second-level covariate into the model. I did confirm that this 2nd-level covariate was created, as it can now be found on the Setup tab under "Covariates (2nd Level). Does that mean it's already included in the model, or do I have to select it somewhere in the Analyses (1st-level) or Results (2nd-level) tabs in order to include it into the model? For reference, I am using CONN 18.a.
Thank you so much in advance.
Sincerely,
Billy