Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session C: Friday, August 15, 2:00 – 5:00 pm, de Brug & E‑Hall

Comparing Variance Partitioning and the Residual Method for Interpreting Brain Recordings

Leo Schultheiß¹, SUBBA REDDY OOTA², Anwar O Nunez-Elizalde³, Fatma Deniz⁴; ¹Technical University of Munich, ²INRIA, ³Independent, ⁴Technische Universität Berlin

Presenter: Fatma Deniz

A shift toward more naturalistic experiments in computational cognitive neuroscience has enabled a richer analysis of brain recordings. These naturalistic experiments often allow for the extraction of multiple feature spaces from stimuli, helping better explain variance in voxelwise encoding models. Two key methods for determining the unique contribution of each feature space to the variance explained are variance partitioning and the residual method. However, no systematic comparison has been conducted to assess their suitability and properties. To address that gap, this work compares both methods by evaluating them in simulated and real-world experiments and comparing their results. Our findings reveal that both variance partitioning and the residual method can effectively determine the unique variance a feature space explains. However, the residual method requires careful verification of the linear dependence between feature spaces, a step that variance partitioning does not need.

Topic Area: Methods & Computational Tools

Extended Abstract: Full Text PDF