|
Fully
compositional data arise when the rows of a data matrix sum
to some constant, 100% or 1 being common. A subset of the
columns of such a data matrix constitutes sub-compositional
data. Such data arise commonly in archaeology, in studies
involving assemblage comparison, spatial clustering, and artefact
compositional analysis, for example.
The paper has two main aims:
(a) To review and illustrate some of the statistical approaches,
some very recent, that have been proposed for the analysis
of such data. We discuss the tension that can arise in choosing
between methods that have theoretically desirable properties
and those that produce practically interpretable results.
This can amount to a choice between different methods of data
transformation and weighting, and the question of how to deal
with zero values is sometimes an issue.
(b) To illustrate analyses using the R language, an open source
package allowing analyses similar to those possible in the
commercial S-Plus package we have previously advocated. R
is yet to be widely used by archaeologists, but makes available
very powerful statistical facilities at no cost, other than
the effort needed to learn the language. We hope to demonstrate
this.
Keywords: : compositional_data, R, S-Plus
|
