Fig. 1. Methods used to estimate test-retest reliability.
A: The unit of analysis is the mean of voxels in a cortical parcel or subcortical region of interest (ROI) for the contrast for Fear > Neutral measured for each participant at each month. B: Test-retest reliability (i.e., temporal stability) is operationalized as the similarity (in rank order of participants) of the BOLD signal in each parcel from month to month. C: The data are observations for a particular parcel from each participant, for each month; we estimate an ICC for each pair of adjacent months for that parcel. This yields N × S rows of data for each parcel, where N is the sample size and S is the number of sessions (months). D: To estimate test-retest reliability, we compute an ICC that decomposes the total variance into the variance in participant means across months and error variance, after conditioning on the group means for each session. The ICC is then computed as the variance due to participant means over the total variance. We do this for each pair of adjacent months and then compute an average ICC across all pairs using meta-analysis. We compute this overall ICC as the measure of test-retest reliability for each of the 414 parcels and subcortical ROIs, and 5 regions of no-interest (see Methods for details).