help
help > RE: Correlation values and 2nd level effect size
Jun 8, 2022 05:06 PM | omaomae - Georgetown University
RE: Correlation values and 2nd level effect size
Dear Alfonso,
Thank you for clarifying this confusion about the ROI.h values that I had as well. I am also conducting ROI-to-ROI analyses. I've been looking at these values and comparing them to the method to obtain the raw Fisher's transformed correlation coefficients described here, but I've found that the values are typically the same after I average the exported values across subjects.
Is this an error and should I not be reporting the ROI.h values as the correlation coefficients? Or is it possible for the values to be the same under certain circumstances? If so, is there an indication as to whether a certain dataset will result in a greater difference between the ROI.h values and the correlation coefficient values?
Best,
Lillian
Originally posted by Alfonso Nieto-Castanon:
Thank you for clarifying this confusion about the ROI.h values that I had as well. I am also conducting ROI-to-ROI analyses. I've been looking at these values and comparing them to the method to obtain the raw Fisher's transformed correlation coefficients described here, but I've found that the values are typically the same after I average the exported values across subjects.
Is this an error and should I not be reporting the ROI.h values as the correlation coefficients? Or is it possible for the values to be the same under certain circumstances? If so, is there an indication as to whether a certain dataset will result in a greater difference between the ROI.h values and the correlation coefficient values?
Best,
Lillian
Originally posted by Alfonso Nieto-Castanon:
Hi
Suneel,
Sorry this thread is a bit confusing because it refers to two different sorts of "correlations" (one referring to Fisher-transformed correlations -across time- representing connectivity strength, and another referring to correlations -across subjects- between behavioral variables and functional connectivity strength) and it is not always clear which one we are referring to.
In any case, if this helps clarify, when using the "import values" button you will get a new second-level covariate that contains the connectivity values for each subject (these values are typically Fisher-transformed correlation coefficients, but it depends on the details of the particular first-level analysis that you are evaluating, e.g. was this a RRC or seed-to-voxel analysis? did you choose correlation or regression coefficients? etc.). These values are a different number per subject and per cluster/ROI
The values in the ROI.h field of the ROI.mat file represent "contrast values", the are computed as:
h = C' * B
where B are the estimated regression coefficients from your 2nd-level model (of the form Y = X*B + noise, where X is your design matrix and Y is your functional connectivity data) and C is your between-subjects contrast vector or matrix. These values are a different number per contrast and per cluster/ROI, so they are never "the same" as the values extracted using the "import values" button (in particular the contrast values (h) are only indirectly related to the original connectivity-strength connectivity values (Y), through the equation:
h = C' *pinv(X)*Y
Hope this helps
Alfonso
Originally posted by Suneel Banerjee:
Sorry this thread is a bit confusing because it refers to two different sorts of "correlations" (one referring to Fisher-transformed correlations -across time- representing connectivity strength, and another referring to correlations -across subjects- between behavioral variables and functional connectivity strength) and it is not always clear which one we are referring to.
In any case, if this helps clarify, when using the "import values" button you will get a new second-level covariate that contains the connectivity values for each subject (these values are typically Fisher-transformed correlation coefficients, but it depends on the details of the particular first-level analysis that you are evaluating, e.g. was this a RRC or seed-to-voxel analysis? did you choose correlation or regression coefficients? etc.). These values are a different number per subject and per cluster/ROI
The values in the ROI.h field of the ROI.mat file represent "contrast values", the are computed as:
h = C' * B
where B are the estimated regression coefficients from your 2nd-level model (of the form Y = X*B + noise, where X is your design matrix and Y is your functional connectivity data) and C is your between-subjects contrast vector or matrix. These values are a different number per contrast and per cluster/ROI, so they are never "the same" as the values extracted using the "import values" button (in particular the contrast values (h) are only indirectly related to the original connectivity-strength connectivity values (Y), through the equation:
h = C' *pinv(X)*Y
Hope this helps
Alfonso
Originally posted by Suneel Banerjee:
Hi Alfonso,
I have a simple follow-up question based on your response. You suggest to use the 'import values' button so arrive at pairwise correlations (r-values). In what instance are these the same as the values in ROI.h? Or does ROI.h contain the Fisher-transformed version of the pairwise correlations? Thanks!
I have a simple follow-up question based on your response. You suggest to use the 'import values' button so arrive at pairwise correlations (r-values). In what instance are these the same as the values in ROI.h? Or does ROI.h contain the Fisher-transformed version of the pairwise correlations? Thanks!
Threaded View
| Title | Author | Date |
|---|---|---|
| Nicola Toschi | Nov 27, 2015 | |
| Pravesh Parekh | Nov 27, 2015 | |
| Pravesh Parekh | Nov 28, 2015 | |
| Alfonso Nieto-Castanon | Nov 28, 2015 | |
| Suneel Banerjee | Jan 28, 2022 | |
| Alfonso Nieto-Castanon | Feb 3, 2022 | |
| omaomae | Jun 8, 2022 | |
| Nicola Toschi | Feb 21, 2016 | |