Reliability of fMRI: Are Group Means Really Representative?

Tynan Stevens¹, Steven Beyea², Ryan D'Arcy³, David Clarke¹

The Rombouts overlap coefficient (Roverlap) is used extensively in fMRI reliability research. While it is widely recognized that this coefficient depends on analysis threshold, existing studies do not agree on the precise relationship between Roverlap and threshold, and so no clear strategy has been employed across studies. We argue that individual variability in the Roverlap-threshold relationship reduces the information available in group level Roverlap results. This variability likely contributes to the inconsistent findings in the literature to date. We thus suggest that control of reliability is best conducted at the individual-level, as group trends may not be truly representative.