Ranking Methods for Compositional Data

Abstract

In the analysis of compositional data, zero components or a non-normal distribution of the log-ratios
\n severely limits the analysis that can be done. Ranking across both cases and components, before
\n scaling, is suggested as a possible new approach. The idea is illustrated with a data set which
\n compares the results with other methods including replacement of Os by small positive quantities and
\n other ranking methods.