Learn R Programming

compositions (version 2.0-8)

pwlr: Pairwise log ratio transform

Description

Compute the pairwise log ratio transform of a (dataset of) composition(s), and its inverse.

Usage

pwlr( x, as.rmult=FALSE, as.data.frame=!as.rmult, ...)
      pwlrInv( z, orig=gsi.orig(z))

Value

pwlr gives the pairwise log ratio transform; accepts a compositional dataset

pwlrInv gives a closed composition with the given wplr-transform; accepts a dataset

Arguments

x

a composition, not necessarily closed

z

the pwlr-transform of a composition, thus a [D(D-1)/2]-dimensional real vector, or a matrix with such many columns

as.rmult

logical; should the output be produced as an rmult object?

as.data.frame

logical; should be as a data.frame? if both are false, rmult will be taken

...

currently unused

orig

the original composition, to check consistency and recover component names

Author

R. Tolosana-Delgado http://www.stat.boogaart.de

Details

The pwlr-transform maps a composition in the $D$-part Aitchison-simplex isometrically to a $D(D-1)/2$ dimensonal euclidian vector, computing each possible logratio (accounting for the fact that $log(A/B)=-log(B/A)$, and therefore only one of them is necessary). The data can then be analysed in this transformation by multivariate analysis tools not relying on the invertibility of the covariance function. The interpretation of the results is relatively simple, since each component captures the behaviour of the simple ratio between two party. However redundance between them is extremely high, and any of alr, clr or ilr transformations may be preferred in most applications.

The pairwise logratio transform is given by $$ pwlr(x)_{ij} := \ln\frac{x_i}{x_j} $$.

The inverse requires some explanation, because of the redundance between pwlr scores. Note that for any three components $A,B,C$ it holds that $log(A/C)=log(A/B)+log(B/C)$. So, any vector of $D(D-1)/2$ coefficients will not be necessarily a valid pwlr-transformed composition: if these coefficients do not satisfy that kind of relations, the vector is, strictly speaking, not a pwlr and should not be inverted. Nevertheless, the function gives a least-squares inversion, as proposed by Tolosana-Delgado and von Eynatten (2009).

References

Aitchison, J. (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and Applied Probability. Chapman & Hall Ltd., London (UK). 416p.

Tolosana-Delgado, R. and H. von Eynatten (2009); Grain-size control on petrographic composition of sediments: compositional regression and rounded zeroes. Mathematical Geosciences: 41(8): 869-886. tools:::Rd_expr_doi("10.1007/s11004-009-9216-6").

See Also

Examples

Run this code
(tmp <- pwlr(c(1,2,3)))
pwlrInv(tmp)

Run the code above in your browser using DataLab