Jonathan Rosenblatt, Ben Gurion University of the Negev, Israel

On the different replicability notions implied by different statistical tools---advocating prevalence estimation for neuroimaging

Different replicability notions require different statistical inference devices. In this talk we argue that random effects inference (RFX), which is the gold standard in neuroimaging, is not compatible with the notion of replicability that interests most neuroscientists (or psychologists). This explains why RFX significant findings may not replicate on different sized groups, and particularly, with single subjects.
We make the same argument also for for multivariate pattern analysis (MVPA), where no statistical gold standard exists. We try to identify what is the replicability notion that interests neuroimaging practitioners, and argue that prevalence estimation is the only statistical inference compatible with it.

Organized by

John-Dylan Haynes / Carsten Allefeld