help
485 Subscribers 
help > RFT inference reports FDR-corr instead of FWE-corr
Showing 1-2 of 2 posts
May 3, 2022 10:05 PM | Marta Ceko - University of Colorado Boulder
RFT inference reports FDR-corr instead of FWE-corr
Hi,
in Conn 20b, 2nd-level analysis results summary, the Worsley RFT stats (under standard settings for cluster-based inference #1) are reported as p <0.05, cluster-size p-FDR corrected, p < 0.001 unc. voxel-wise.
Should the cluster-size correction used in this case not be FWE?
Example: k = 11, cluster-size p-FWE .059, FDR 0.047 --- > is reported as not-significant using RFT.
Thank you,
Marta
in Conn 20b, 2nd-level analysis results summary, the Worsley RFT stats (under standard settings for cluster-based inference #1) are reported as p <0.05, cluster-size p-FDR corrected, p < 0.001 unc. voxel-wise.
Should the cluster-size correction used in this case not be FWE?
Example: k = 11, cluster-size p-FWE .059, FDR 0.047 --- > is reported as not-significant using RFT.
Thank you,
Marta
May 4, 2022 03:05 PM | Alfonso Nieto-Castanon - Boston University
RE: RFT inference reports FDR-corr instead of FWE-corr
Hi Marta,
Sorry, I am not exactly sure to understand what you mean, could you please clarify how exactly you generated that example? In CONN's result explorer window when using the RFT default settings that cluster would be labeled as significant, as its cluster-level p-FDR value is below .05. If you are referring to how people use RFT generally in the literature, then yes, you are perfectly right that both approaches (using cluster-level p-FWE and cluster-level p-FDR) are perfectly valid/common approaches, and you can simply use your preferred method clicking on the 'show details' button and selecting there the option 'cluster-size p-FWE' or 'cluster-size p-FDR'.
Best
Alfonso
Originally posted by Marta Ceko:
Sorry, I am not exactly sure to understand what you mean, could you please clarify how exactly you generated that example? In CONN's result explorer window when using the RFT default settings that cluster would be labeled as significant, as its cluster-level p-FDR value is below .05. If you are referring to how people use RFT generally in the literature, then yes, you are perfectly right that both approaches (using cluster-level p-FWE and cluster-level p-FDR) are perfectly valid/common approaches, and you can simply use your preferred method clicking on the 'show details' button and selecting there the option 'cluster-size p-FWE' or 'cluster-size p-FDR'.
Best
Alfonso
Originally posted by Marta Ceko:
Hi,
in Conn 20b, 2nd-level analysis results summary, the Worsley RFT stats (under standard settings for cluster-based inference #1) are reported as p <0.05, cluster-size p-FDR corrected, p < 0.001 unc. voxel-wise.
Should the cluster-size correction used in this case not be FWE?
Example: k = 11, cluster-size p-FWE .059, FDR 0.047 --- > is reported as not-significant using RFT.
Thank you,
Marta
in Conn 20b, 2nd-level analysis results summary, the Worsley RFT stats (under standard settings for cluster-based inference #1) are reported as p <0.05, cluster-size p-FDR corrected, p < 0.001 unc. voxel-wise.
Should the cluster-size correction used in this case not be FWE?
Example: k = 11, cluster-size p-FWE .059, FDR 0.047 --- > is reported as not-significant using RFT.
Thank you,
Marta