Nichols, T.; Ridgway, G.; Webster, M.; Smith, S.; (2008) GLM permutation - nonparametric inference for arbitrary general linear models. Presented at: 14th Annual Meeting of the Organization for Human Brain Mapping, Melbourne, Australia.

Preview

PDF 13848.pdf Download (67kB)

Abstract

Introduction: Permutation methods are finding growing use in neuroimag- ing data analyses (e.g. randomise in FSL, SnPM in SPM, XBAMM/BAMM/CAMBA, etc). These methods provide ex- act control of false positives, make only weak assumptions, and allow nonstandard types of statistics (e.g. smoothed variance t- test). With fast and inexpensive computing, there would seem few reasons not to use nonparametric methods. A significant limitation of these methods, however, is the lack of flexibility with respect to the experimental design and nuisance variables. Each specific design dictates the type of exchange- ability of null data, and hence how to permute. Nuisance effects (e.g. age) render data non-exchangeable even when the effect of interest is null. Hence, even something as simple as ANCOVA has no exact permutation test. Recently there has been an active literature on approximate– but accurate–permutation tests for 2-variable regression, one effect of interest, one nuisance (see review by Anderson & Robinson [1]). Here we extend and evaluate these methods for use with an arbitrary General Linear Model (GLM).

Download activity - last month

Export as JSONXMLCSV

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item