fMRI replicability depends upon sufficient individual-level data

Abstract

The replicability of findings drawn from functional magnetic resonance imaging (fMRI) data have increasingly been called into question. Concerns have been raised that historically, sample sizes have been insufficient to produce adequate power, leading to unreliable results. Recently, Turner and colleagues demonstrated that even with large sample sizes, the replicability of fMRI can be worryingly low. However, their datasets featured low amounts of data at the individual-level. Here, I demonstrate that replicability depends critically on sufficient individual-level sampling. I show that fMRI can have strong replicability even at modest sample sizes when individuals are adequately sampled, but that inadequate individual-level sampling leads to poor replicability. These data indicate that fMRI replicability cannot be judged solely on sample size, and that adequate sampling at the individual-level is a critical design consideration.