Orthonormal decomposition of variance
This function performs the orthonormal decomposition of variance of a
quantitative variable on an orthonormal basis. It also returns the results of
five non parametric tests associated to the variance decomposition. It thus
provides tools (graphical displays and test) for analysing phylogenetic,
pattern in one quantitative trait. This implementation replace the
(deprecated) version from the ade4 package.
orthogram(x, tre = NULL, orthobas = NULL, prox = NULL, nrepet = 999,
posinega = 0, tol = 1e-07, cdot = 1.5, cfont.main = 1.5, lwd = 2,
nclass, high.scores = 0, alter = c("greater", "less", "two-sided"))x |
a numeric vector corresponding to the quantitative variable |
tre |
|
orthobas |
an object of class |
prox |
a matrix of phylogenetic proximities as returned by
|
nrepet |
an integer giving the number of permutations |
posinega |
a parameter for the ratio test. If posinega > 0, the function computes the ratio test. |
tol |
a tolerance threshold for orthonormality condition |
cdot |
a character size for points on the cumulative decomposition display |
cfont.main |
a character size for titles |
lwd |
a character size for dash lines |
nclass |
a single number giving the number of cells for the histogram |
high.scores |
a single number giving the number of vectors to return. If > 0, the function returns labels of vectors that explains the larger part of variance. |
alter |
a character string specifying the alternative hypothesis, must be one of "greater" (default), "less" or "two-sided" |
Several orthonormal bases can be used. By default, basis is constructed from
a partition of tips according to tree topology (as returned by
treePart); for this, the argument tre must be provided.
Alternatively, one can provide an orthonormal basis as returned by
orthobasis.phylo/me.phylo (argument
orthobas), or provide a proximity matrix from which an orthobasis
based on Moran's eigenvectors will be constructed (argument prox).
The function computes the variance decomposition of a quantitative vector x on an orthonormal basis B. The variable is normalized given the uniform weight to eliminate problem of scales. It plots the squared correlations R^2 between x and vectors of B (variance decomposition) and the cumulated squared correlations SR^2 (cumulative decomposition). The function also provides five non parametric tests to test the existence of autocorrelation. The tests derive from the five following statistics :
- R2Max=max(R^2). It takes high value when a high part of
the variability is explained by one score.
-
SkR2k=sum_i^(n-1) i*(R^2)_i. It compares
the part of variance explained by internal nodes to the one explained by end
nodes.
- Dmax=max_(m=1,...,n-1)(sum_(j=1)^m(R^2_j) - (m/n-1))max_(m=1,...,n-1)(sum_(j=1)^m(R^2_j) - (m/n-1)). It examines
the accumulation of variance for a sequence of scores.
-
SCE=sum_(m=1)^(n-1)(sum_(j=1)^m(R^2_j) - (m/n-1))^2sum_(m=1)^(n-1)(sum_(j=1)^m(R^2_j) - (m/n-1))^2. It
examines also the accumulation of variance for a sequence of scores.
-
ratio: depends of the parameter posinega. If posinega > 0, the statistic
ratio exists and equals sum_i (R^2)_i with
i < posinega + 1. It compares the part of variance explained by internal
nodes to the one explained by end nodes when we can define how many vectors
correspond to internal nodes.
If (high.scores = 0), returns an object of class 'krandtest'
(randomization tests) corresponding to the five non parametric tests.
If (high.scores > 0), returns a list containg :
w |
: an object of
class |
scores.order |
: a vector which terms give labels of vectors that explain the larger part of variance |
This function replaces the former version from the ade4 package, which is deprecated. Note that if ade4 is not loaded BEFORE adephylo, then the version from ade4 will erase that of adephylo, which will still be available from adephylo::orthogram. In practice, though, this should never happen, since ade4 is loaded as a dependence by adephylo.
Original code: Sebastien Ollier and Daniel Chessel.
Current maintainer: Stephane Dray <stephane.dray@univ-lyon1.fr>
Ollier, S., Chessel, D. and Couteron, P. (2005) Orthonormal Transform to Decompose the Variance of a Life-History Trait across a Phylogenetic Tree. Biometrics, 62, 471–477.
## Not run:
if(require(ape) && require(phylobase)){
## a phylogenetic example
data(ungulates)
tre <- read.tree(text=ungulates$tre)
plot(tre)
## look at two traits
afbw <- log(ungulates$tab[,1])
neonatw <- log((ungulates$tab[,2]+ungulates$tab[,3])/2)
names(afbw) <- tre$tip.label
names(neonatw) <- tre$tip.label
plot(afbw, neonatw) # relationship between traits
lm1 <- lm(neonatw~afbw)
resid <- residuals(lm1)
abline(lm1)
## plot the two traits and the residuals of lm1
x <- phylo4d(tre, cbind.data.frame(afbw, neonatw, residuals=resid))
table.phylo4d(x) # residuals are surely not independant
## default orthogram for residuals of lm1
orthogram(resid, tre)
## using another orthonormal basis (derived from Abouheif's proximity)
myOrthoBasis <- orthobasis.phylo(tre, method="oriAbouheif") # Abouheif's proximities
orthogram(resid, ortho=myOrthoBasis) # significant phylog. signal
## Abouheif's test
W <- proxTips(tre, method="oriAbouheif") # proximity matrix
abouheif.moran(resid, W)
}
## End(Not run)Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.