The Algonauts Project 2021 Challenge - Full Track (Whole Brain) Forum

Go back to competition Back to thread list Post in this thread

> Reliability threshold

Dear Experts,
I have a question on how reliable voxels were chosen. In your website, you say that:
"The split-half reliability was then calculated using Spearman-Brown formula (2ρ/(1+ρ)) and by averaging the reliability across all combinations of splits. The remaining reliable beta values were used as the encoding model prediction data for the Full Track."
However, I cannot find anywhere what the threshold of the Spearman-Brown reliability was for a voxel to be considered "reliable". Would you be able to provide this information?
Thank you very much,

Leonardo Tozzi

Posted by: leo.tozzi @ Aug. 3, 2021, 5:21 p.m.

Hi Leonardo,
Sorry for the late reply. The threshold for split-half reliability varied for each ROI. We performed a permutation test by permuting the conditions in one of the split to calculate the p-values. The voxels with p values >0.05 were discarded as non-reliable voxels.
Hope this helps with your question!

Posted by: alexlascelles @ Aug. 9, 2021, 2:33 p.m.

Thank you very much!

Leonardo Tozzi

Posted by: leo.tozzi @ Aug. 9, 2021, 2:35 p.m.
Post in this thread