A defense of using resting state fMRI as null data for estimating false positive rates

TL;DR

This paper defends the use of resting state fMRI data as a valid null model for estimating false positive rates in task fMRI analysis, arguing it effectively represents noise and challenges critiques suggesting otherwise.

Contribution

It provides a defense of using resting state fMRI as null data in false positive rate estimation, addressing criticisms and clarifying its appropriateness.

Findings

Resting state fMRI data is suitable as null data for noise estimation.

Analysis software should handle the structured noise in resting state data.

Critique of Slotnick's method highlights potential issues.

Abstract

A recent Editorial by Slotnick (2017) reconsiders the findings of our paper on the accuracy of false positive rate control with cluster inference in fMRI (Eklund et al, 2016), in particular criticising our use of resting state fMRI data as a source for null data in the evaluation of task fMRI methods. We defend this use of resting fMRI data, as while there is much structure in this data, we argue it is representative of task data noise and as such analysis software should be able to accommodate this noise. We also discuss a potential problem with Slotnick's own method.