fMRI replicability depends upon sufficient individual-level data

T he reproducibility of task-based functional magnetic resonance imaging (fMRI), or lack thereof, has become a topic of intense scrutiny Relative to other human techniques, fMRI has high costs associated with data collection, storage, and processing. To justify these costs, the inferences gained from fMRI need to be robust and meaningful. Hence, although large, sufficiently powered data sets may be costly, this is favorable to collecting many insufficiently powered data sets from which reliable conclusions cannot be drawn. However, it can be difficult to determine a priori how much data are needed. Although power analyses can help 3 , accurately calculating power itself requires an appropriate estimate of the expected effect size, which can be hard to obtain if previous studies had insufficient data to produce reliable effect size estimates. Furthermore, mechanistic basic science explores novel phenomena with innovative paradigms such that extrapolation of effect sizes from existing data may not be appropriate.