Linear model calculations are made for many random versions of data. Using residual randomization in a permutation procedure, sums of squares are calculated over many permutations to generate empirical probability distributions for evaluating model effects. This packaged is described by Collyer & Adams (2018). Additionally, coefficients, statistics, fitted values, and residuals generated over many permutations can be used for various procedures including pairwise tests, prediction, classification, and model comparison. This package should provide most tools one could need for the analysis of high-dimensional data, especially in ecology and evolutionary biology, but certainly other fields, as well.
Functions in this package allow one to evaluate linear models with residual randomization. The name, "RRPP", is an acronym for, "Randomization of Residuals in a Permutation Procedure." Through the various functions in this package, one can use randomization of residuals to generate empirical probability distributions for linear model effects, for high-dimensional data or distance matrices.
An especially useful option of this package is to fit models with either ordinary or generalized least squares estimation (OLS or GLS, respectively), using theoretic covariance matrices. Mixed linear effects can also be evaluated.
Key functions for this package:
lm.rrppFits linear models, using RRPP. plus model comparisons.
coef.lm.rrppExtract coefficients or perform test on coefficients, using RRPP.
predict.lm.rrppPredict values from lm.rrpp fits and generate bootstrapped confidence intervals.
pairwisePerform pairwise tests, based on lm.rrpp model fits.
Maintainer: Michael Collyer mlcollyer@gmail.com (ORCID)
Authors:
Dean Adams (ORCID)
Michael Collyer and Dean Adams
Useful links: