An iterative jackknife approach for assessing reliability and power of FMRI group analyses

For functional magnetic resonance imaging (fMRI) group activation maps, so-called second-level random effect approaches are commonly used, which are intended to be generalizable to the population as a whole. However, reliability of a certain activation focus as a function of group composition or group size cannot directly be deduced from such maps. This question is of particular relevance when examining smaller groups (<20-27 subjects). The approach presented here tries to address this issue by iteratively excluding each subject from a group study and presenting the overlap of the resulting (reduced) second-level maps in a group percent overlap map. This allows to judge where activation is reliable even upon excluding one, two, or three (or more) subjects, thereby also demonstrating the inherent variability that is still present in second-level analyses. Moreover, when progressively decreasing group size, foci of activation will become smaller and/or disappear; hence, the group size at which a given activation disappears can be considered to reflect the power necessary to detect this particular activation. Systematically exploiting this effect allows to rank clusters according to their observable effect size. The approach is tested using different scenarios from a recent fMRI study (children performing a "dual-use" fMRI task, n = 39), and the implications of this approach are discussed.