Hi, I have a qustions. In ROI-to-ROI results explorer, sometime I find the result of some regions does not correspond to that of other regions. For example, region A (source region) shows positive connectivity with region B (target region). But when I reverse the source and target region(now region A is target region & resign B is source region), it does not show positive connecitivty. Is it a bug of program? or Do I misunderstand something? Now I'm confused to interpret the results. I appreciate if you help me.
This condition might be due to the nature of the multiple-comparisons correction applied on the analysis results to determine the 'significance' of the results. If you are solely interested in the correlation between region A and region B (and this was your only test of interest) then you would determine the 'significance' of this result looking at the 'uncorrected p-value' of the ROI-to-ROI results (and this will be the same, whether you use A as source and B as target, or A as target and B as source). If, on the contrary, you are interested in the correlation between region A and *any* other region (e.g. you want to know whether there are *any* regions that are associated with a given a priori region A), then you would look at the 'FDR-corrected p-value' of the ROI-to-ROI results when using A as source and all other regions as targets. Because FDR is an adaptive method (e.g. the FDR-corrected p-value between region A and B depends not only on the association between A and B but also depends on all of the associations between region A and all other regions) it does not result in symmetric p-values (i.e. the FDR-corrected p-value between region A as source and region B as target is not the same as the FDR-corrected p-value between region B as source and region A as target, because the former depends on all of the associations between region A and all of the other regions, while the later depends on all of the associations between region B and all of the other regions). When interpreting your results just keep in mind that this FDR method (e.g. using a FDR-corrected p<.05 threshold) is set to guarantee that, among the resulting 'significant' effects for any given test, you can expect only 5% or less of these to be false positives.
Let me know if this clarifies a bit
Alfonso
Originally posted by Peter Dessen:
Hi, I have a qustions. In ROI-to-ROI results explorer, sometime I find the result of some regions does not correspond to that of other regions. For example, region A (source region) shows positive connectivity with region B (target region). But when I reverse the source and target region(now region A is target region & resign B is source region), it does not show positive connecitivty. Is it a bug of program? or Do I misunderstand something? Now I'm confused to interpret the results. I appreciate if you help me.
Thank you very much! I cleary understand it.
Kind regards,
Peter
Originally posted by Alfonso Nieto-Castanon:
This condition might be due to the nature of the multiple-comparisons correction applied on the analysis results to determine the 'significance' of the results. If you are solely interested in the correlation between region A and region B (and this was your only test of interest) then you would determine the 'significance' of this result looking at the 'uncorrected p-value' of the ROI-to-ROI results (and this will be the same, whether you use A as source and B as target, or A as target and B as source). If, on the contrary, you are interested in the correlation between region A and *any* other region (e.g. you want to know whether there are *any* regions that are associated with a given a priori region A), then you would look at the 'FDR-corrected p-value' of the ROI-to-ROI results when using A as source and all other regions as targets. Because FDR is an adaptive method (e.g. the FDR-corrected p-value between region A and B depends not only on the association between A and B but also depends on all of the associations between region A and all other regions) it does not result in symmetric p-values (i.e. the FDR-corrected p-value between region A as source and region B as target is not the same as the FDR-corrected p-value between region B as source and region A as target, because the former depends on all of the associations between region A and all of the other regions, while the later depends on all of the associations between region B and all of the other regions). When interpreting your results just keep in mind that this FDR method (e.g. using a FDR-corrected p<.05 threshold) is set to guarantee that, among the resulting 'significant' effects for any given test, you can expect only 5% or less of these to be false positives.
Let me know if this clarifies a bit
Alfonso
Originally posted by Peter Dessen:
Hi, I have a qustions. In ROI-to-ROI results explorer, sometime I find the result of some regions does not correspond to that of other regions. For example, region A (source region) shows positive connectivity with region B (target region). But when I reverse the source and target region(now region A is target region & resign B is source region), it does not show positive connecitivty. Is it a bug of program? or Do I misunderstand something? Now I'm confused to interpret the results. I appreciate if you help me.