[R-sig-eco] Compositional Data Analysis: Simplex Scaling Function Selection

[Rich Shepard]

   The data sets to be analyzed are proportions based on counts. Each row is
closed; the proportions total 1.00. I understand the need to transform the
raw data to a CoDA simplex, and my reading strongly suggests that the
isometric log ratio (ilr) function is the most appropriate for these data.

   In Boogaart and Delgado's "Analyzing Compositional Data With R", page 30
describes the various available scaling functions. The advice needed is
which is most appropriate for my data sets:
   'aplus' -- Aichison (ratio) geometry in the real data space.
   'rcomp' -- Real (interval) compositional scale.
   'acomp' -- Aichison (ratio) compositional scale.
Or, one of the others.

   Reading several docs leaves me confused over which one should be used for
both descriptive statistics and further analyses such as CCA, clustering, or
time series.

   There is no statistics group on stackexchange.com and stackoverflow.com is
for programming questions, not statistical questions. If there is a better
place for this question, please point me to it as there will be other CoDA
questions as the analyses proceed.

Rich