rcomp(X,parts=1:NCOL(oneOrDataset(X)),total=1,warn.na=FALSE,
detectionlimit=NULL,BDL=NULL,MAR=NULL,MNAR=NULL,SZ=NULL)"rcomp" representing a closed composition
or a matrix of class "rcomp" representing
multiple closed compositions, by rows."rcomp" follow the traditional
(often statistically inconsistent) approach regarding compositions simply
as a multivariate vector of positive numbers summing up to 1. This space of
D positive numbers summing to 1 is traditionally called the D-1-dimensional
simplex. The compositional scale was in-depth analysed by Aitchison
(1986) and he found serious reasons why compositional data should be
analysed with a different geometry. The functions around the class
"acomp" follow his
approach. However the Aitchison approach based on log-ratios is
sometimes criticized (e.g. Rehder and Zier, 2002). It cannot deal with
absent parts (i.e. zeros). It is sensitive to large measurement errors
in small amounts. The Aitchison operations cannot represent simple
mixture of different compositions. The used transformations
are not uniformly continuous. Straight lines and ellipses in Aitchison
space look strangely in ternary diagrams. As all uncritical statistical
analysis, blind application of logratio-based analysis is sometimes
misleading. Therefore it is sometimes useful to analyse
compositional data directly as a multivariate dataset of portions
summing to 1. However a clear warning must be given that the
utilisation of almost any kind of
classical multivariate analysis introduce some kinds of artifacts
(e.g. Chayes 1960) when applied to compositional data. So, extra care
and considerable expert knowlegde is needed for the proper
interpretation of results achieved in this non-Aitchison approach. The
package tries to lead the user around these artifacts as much as
possible and gives hints to major pitfalls in the help. However
meaningless results cannot be fully avoided in this (rather inconsistent) approach.
A side effect of the procedure is to force the compositions to sum to
one, which is done by the closure operation clo .
The classes rcomp, acomp, aplus, and rplus are designed in a fashion as similar as
possible, in order to allow direct comparison between results achieved
by the different approaches. Especially the acomp logistic transforms
clr, alr, ilr are mirrored
by analogous linear transforms cpt, apt,
ipt in the rcomp class framework.
cpt, apt, ipt,
acomp, rplus,
princomp.rcomp,
plot.rcomp, boxplot.rcomp,
barplot.rcomp, mean.rcomp,
var.rcomp, variation.rcomp,
cov.rcomp, msd,
convex.rcomp, +.rcompdata(SimulatedAmounts)
plot(rcomp(sa.tnormals))Run the code above in your browser using DataLab