Statistical methods for intersubject analysis of neuroscience data

The statistical analysis of neuroscience data poses several challenges due to the data's typically high dimensionality and its complex spatiotemporal structure. In this work we address statistical issues arising from two types of neuroscience data: magnetic resonance images (MRI) and electrocorticographic (ECoG) signals. Permutation tests are commonly applied to MRIs in the neuroscience literature. A methodology to control for potentially confounding covariates in permutation tests using a Markov chain Monte Carlo algorithm is proposed. Temporal relationships in ECoG data are often estimated using Granger temporality. We propose a methodology to control for potentially mediating or confounding associations between time series. Finally, an extension to estimate Granger temporality across multiple subjects is developed using covariance smoothing.