Data-Analytical Stability of Second-Level Inference in fMRI Data Analysis

• Activity Number: 360
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Sanne P. Roels* and Beatrijs Moerkerke and Tom Loeys
• Companies: Ghent University and Ghent University and Ghent University
• Keywords: fmri ; reproducibility ; inference ; second-level model

Abstract

We investigate the impact of decisions in the second level (i.e. over subjects) inferential process in fMRI on the data-analytical stability of results, a proxy for reproducibility of results. Combining information over subjects is crucial for the localisation of brain activation, however, no golden standard exist on how to achieve this. A second level analysis typically proceeds via a two-step general linear model in which information is first pooled within a subject and in a second step over subjects (Beckmann et al., 2003). We compare data-analytical stability of models that take into account first level (within subjects) variability and models that do not take this into account. In all models, we consider 3 commonly used procedures to correct for the multiplicity of testing (Liebermann et al.; 2009; Bennett et al., 2009). We additionally contrast permutation based inference and inference based on parametrical assumptions (Nichols et al., 2002). Based on a simulation study and on real data we advice on which strategies result in the most stable (i.e. least variable) results.