Conjunction Inference Using the Bayesian Interpretation of the Positive False Discovery Rate

Many neuroimaging studies use conjunction analyses to find brain regions with effects consistent across tasks. We find that SPM’s minimum statistic inferences [1] are often misinterpreted and, in other work, we describe a simple method to make inferences on the “conjunction null” using the minimum statistic ([2] & Poster WE137). In this work we describe a new method for conjunction inference that doesn’t use the minimum statistic. We propose using Bayesian methods to compute the posterior probability of the conjunction null. In lieu of a full Bayesian model, we show how the Positive False Discovery Rate (pFDR) can be used to compute such a posterior probability. We describe the method and apply it to a conjunction analysis from an fMRI study of inhibition.