[color=#000000]Hi Sarah,[/color]
[color=#000000]That's an interesting question, and yes, that is perfectly fine and there are really many different scenarios that can lead to that behavior (absence of a bivariate correlation between regions A and B, but presence of a semipartial correlation between the same variables when controlling for the effect of a third region C), [/color][color=#000000]for example, just to name a few:[/color]
[color=#000000] 1) B may be strongly modulated by C but only weakly modulated by A (e.g. B=eps*A+C), so the A-B modulation may only be visible/significant when controlling for the effect of C[/color]
[color=#000000] 2) perhaps [/color]A and B are really independent but both have strong cumulative effects on C (e.g. C=A+B), so when controlling for C (either statistically or by design) A and B show an inverse correlation
3) or perhaps A and C are anticorrelated while both have a direct influence on B (e.g. B = A+C).
In all of those cases, the bivariate correlation between A and B may be zero, or nominally small, while the semipartial correlation between the same variables when controlling for a third related variable (C) is not.
Hope this helps
[color=#000000]Alfonso[/color]
[i]Originally posted by Sarah Kark:[/i][quote]Hello,
I am delighted to have discovered the semi-partial correlation feature but have a question:
When using bivariate correlation S2V with a given seed I obtain, for example, I do not observe FC between my seed and the amygdala (using TFCE on n=135 data). When I use semi-partial with the same seed controlling for another nearby region (my ideal analysis) I DO obtain FC between my seed of interest and the amygdala (same thresholding TCFE approach).
How is it possible to not obtain a connection between two regions using bivariate but to observe a significant connection when controlling for another area?
Thank you!
Sarah[/quote]
[color=#000000]That's an interesting question, and yes, that is perfectly fine and there are really many different scenarios that can lead to that behavior (absence of a bivariate correlation between regions A and B, but presence of a semipartial correlation between the same variables when controlling for the effect of a third region C), [/color][color=#000000]for example, just to name a few:[/color]
[color=#000000] 1) B may be strongly modulated by C but only weakly modulated by A (e.g. B=eps*A+C), so the A-B modulation may only be visible/significant when controlling for the effect of C[/color]
[color=#000000] 2) perhaps [/color]A and B are really independent but both have strong cumulative effects on C (e.g. C=A+B), so when controlling for C (either statistically or by design) A and B show an inverse correlation
3) or perhaps A and C are anticorrelated while both have a direct influence on B (e.g. B = A+C).
In all of those cases, the bivariate correlation between A and B may be zero, or nominally small, while the semipartial correlation between the same variables when controlling for a third related variable (C) is not.
Hope this helps
[color=#000000]Alfonso[/color]
[i]Originally posted by Sarah Kark:[/i][quote]Hello,
I am delighted to have discovered the semi-partial correlation feature but have a question:
When using bivariate correlation S2V with a given seed I obtain, for example, I do not observe FC between my seed and the amygdala (using TFCE on n=135 data). When I use semi-partial with the same seed controlling for another nearby region (my ideal analysis) I DO obtain FC between my seed of interest and the amygdala (same thresholding TCFE approach).
How is it possible to not obtain a connection between two regions using bivariate but to observe a significant connection when controlling for another area?
Thank you!
Sarah[/quote]