Permutation tests for phylogenetic comparative analyses of high-dimensional shape data: What you shuffle matters

• DOI: 10.1111/evo.12596
• Publication Date: 2015-01-06T16:51:24Z
• Authors: Dean C. Adams; Michael L. Collyer
• Keywords: Resampling; Shuffling; Trait

Evaluating statistical trends in high-dimensional phenotypes poses challenges for comparative biologists, because the high-dimensionality of trait data relative to number species can prohibit parametric tests from being computed. Recently, two methods were proposed circumvent this difficulty. One obtains phylogenetic independent contrasts all variables, and statistically evaluates linear model by permuting phylogenetically (PICs) response data. The other uses a distance-based approach obtain coefficients generalized least squares models (D-PGLS), subsequently permutes original evaluate effects. Here, we show that PICs is not equivalent prior analyses as D-PGLS. We further explain why are correct exchangeable units under null hypothesis, demonstrate misspecification permutable leads inflated type I error rates tests. then simply shuffling recalculating with each iteration yields significance levels correspond those found using Thus, while summary statistics based on PGLS same, lead strikingly different inferential outcomes respect biological inferences.