help
help > RE: ROI-to-ROI analysis with two covariates
Nov 14, 2014 02:11 AM | Alfonso Nieto-Castanon - Boston University
RE: ROI-to-ROI analysis with two covariates
Hi Charles,
First, if you want to look at what ROI-to-ROI connections might show associations between connectivity strength and both covariates A and B, then this is simply a conjunction analyses, which you can easily obtain from the intersection of the results in the two independent analyses (associations with covariate A, and associations with covariate B; both corrected at the same FDR- or FWE- corrected level). The tricky part is when you want to look at what ROI-to-ROI connections might show differential associations between connectivity strength and the two covariates A and B (e.g. associated with one covariate but not the other, or more strongly associated with one covariate compared to the other). If your two covariates are in the same "units" (e.g. same measures in two time-points) then you would typically accomplish this selecting the AllSubjects, CovariateA, and CovariateB in subject effects and entering a contrast [0 1 -1] (and performing a two-sided test). When the two covariates are not in the same "units" I would probably recommend to normalize the two covariates (divide each covariate by their in-sample standard deviation) and then using the same [0 1 -1] contrast on the transformed covariates/model. Alternatively you could (outside CONN, unfortunately) perform tests that compared the two dependent correlation coefficients between the connectivity values and each of the two covariates. Let me know your thoughts.
Best
Alfonso
Originally posted by Charles Malpas:
First, if you want to look at what ROI-to-ROI connections might show associations between connectivity strength and both covariates A and B, then this is simply a conjunction analyses, which you can easily obtain from the intersection of the results in the two independent analyses (associations with covariate A, and associations with covariate B; both corrected at the same FDR- or FWE- corrected level). The tricky part is when you want to look at what ROI-to-ROI connections might show differential associations between connectivity strength and the two covariates A and B (e.g. associated with one covariate but not the other, or more strongly associated with one covariate compared to the other). If your two covariates are in the same "units" (e.g. same measures in two time-points) then you would typically accomplish this selecting the AllSubjects, CovariateA, and CovariateB in subject effects and entering a contrast [0 1 -1] (and performing a two-sided test). When the two covariates are not in the same "units" I would probably recommend to normalize the two covariates (divide each covariate by their in-sample standard deviation) and then using the same [0 1 -1] contrast on the transformed covariates/model. Alternatively you could (outside CONN, unfortunately) perform tests that compared the two dependent correlation coefficients between the connectivity values and each of the two covariates. Let me know your thoughts.
Best
Alfonso
Originally posted by Charles Malpas:
Hi Alfonso,
I would like to run an ROI-to-ROI analysis (eventually using NBS) to compare two covariates. Specifically, I would like to identity which subnetworks are associated with covariate A and which are associated with covariate B. I could simply run two regression analyses and then look at where the two resulting networks differ, but I think some sort of conjunction/disjunction analysis would be required. Do you have any ideas how this might be accomplished in Conn?
Many thanks, and sorry for the left-of-field question. I have looked everywhere and can't seem to find any references to running this sort of analysis on a connectivity matrix.
Kind regards,
Charles
I would like to run an ROI-to-ROI analysis (eventually using NBS) to compare two covariates. Specifically, I would like to identity which subnetworks are associated with covariate A and which are associated with covariate B. I could simply run two regression analyses and then look at where the two resulting networks differ, but I think some sort of conjunction/disjunction analysis would be required. Do you have any ideas how this might be accomplished in Conn?
Many thanks, and sorry for the left-of-field question. I have looked everywhere and can't seem to find any references to running this sort of analysis on a connectivity matrix.
Kind regards,
Charles
Threaded View
| Title | Author | Date |
|---|---|---|
| Charles Malpas | Nov 1, 2014 | |
| Alfonso Nieto-Castanon | Nov 14, 2014 | |