[color=#000000]Hi Jeff,[/color]
[color=#000000]You are right, the MVPA omnibus test is a multivariate test so there is no implicit directionality there. If you find a significant cluster A there you know that [/color]"some aspect" of the functional connectivity pattern between A and the rest of the brain correlates across subjects with your biomarker values. That "aspect" is basically a pattern of differences in the connectivity between the seed A and the rest of the brain per unit change in biomarker values (e.g. the connectivity between A and each voxel increase/decrease across subjects by ### per unit-change in biomarker values - and the ### value/sign may be different for each voxel-). That same pattern will be shown when you perform post hoc regression SBC analyses using the significant cluster A as seed. Strictly speaking the thresholding of that pattern, part of the standard cluster-based inferences of your post hoc SBC, will be somewhat arbitrary (i.e. not really part of our confirmatory test), but it can help you nevertheless describe that pattern in terms of simpler parts (e.g. increased connectivity between A and B with higher biomarker values and/or decreased connectivity between A and C with higher biomarker values)
Let me know if that makes sense and/or your thoughts/comments/suggestions
Alfonso
[i]Originally posted by Jeff Browndyke:[/i][quote]Hi, Alfonso.
You were right about the between-condition contrast. My model for the MVPA regression is as follows:
Between-subjects contrast [0 1] -- all subjects and biomarker value
Between-conditions contrast [-1 1] -- reflecting differences from baseline to follow-up
Voxel-to-voxel measures [1 0 0 0; 0 1 0 0; 0 0 1 0; 0 0 0 1] -- 4 MVPA components
I was hoping that examining the MVPA output x regression in the calculator would tell me something about the directionality of the association, as this model is an F-test. Or, is it a simple matter of reversing sign on the between-subjects contrast weight to say examine for an inverse association between the regressor and MVPA?
If it is not possible to determine the directionality of the association at the omnibus level and we can only look at post-hoc associations, how does one know if any significant MVPA-derived blobs reflect a positive or inverse association with the regressor? Can't the post-hoc seed-to-voxels reflect something different (e.g., inverse association MVPA blob seed, which then is used in post-hoc seed to voxel reflecting a combination of positive and negative associations with that seed)?
Warm regards,
Jeff[/quote]
[color=#000000]You are right, the MVPA omnibus test is a multivariate test so there is no implicit directionality there. If you find a significant cluster A there you know that [/color]"some aspect" of the functional connectivity pattern between A and the rest of the brain correlates across subjects with your biomarker values. That "aspect" is basically a pattern of differences in the connectivity between the seed A and the rest of the brain per unit change in biomarker values (e.g. the connectivity between A and each voxel increase/decrease across subjects by ### per unit-change in biomarker values - and the ### value/sign may be different for each voxel-). That same pattern will be shown when you perform post hoc regression SBC analyses using the significant cluster A as seed. Strictly speaking the thresholding of that pattern, part of the standard cluster-based inferences of your post hoc SBC, will be somewhat arbitrary (i.e. not really part of our confirmatory test), but it can help you nevertheless describe that pattern in terms of simpler parts (e.g. increased connectivity between A and B with higher biomarker values and/or decreased connectivity between A and C with higher biomarker values)
Let me know if that makes sense and/or your thoughts/comments/suggestions
Alfonso
[i]Originally posted by Jeff Browndyke:[/i][quote]Hi, Alfonso.
You were right about the between-condition contrast. My model for the MVPA regression is as follows:
Between-subjects contrast [0 1] -- all subjects and biomarker value
Between-conditions contrast [-1 1] -- reflecting differences from baseline to follow-up
Voxel-to-voxel measures [1 0 0 0; 0 1 0 0; 0 0 1 0; 0 0 0 1] -- 4 MVPA components
I was hoping that examining the MVPA output x regression in the calculator would tell me something about the directionality of the association, as this model is an F-test. Or, is it a simple matter of reversing sign on the between-subjects contrast weight to say examine for an inverse association between the regressor and MVPA?
If it is not possible to determine the directionality of the association at the omnibus level and we can only look at post-hoc associations, how does one know if any significant MVPA-derived blobs reflect a positive or inverse association with the regressor? Can't the post-hoc seed-to-voxels reflect something different (e.g., inverse association MVPA blob seed, which then is used in post-hoc seed to voxel reflecting a combination of positive and negative associations with that seed)?
Warm regards,
Jeff[/quote]