Robust Compositional Data Analysis

Abstract

Many practical data sets contain outliers or other forms of data inhomogeneities. Robust statistics offers concepts how to deal with these situations where the data do not follow strict model assumptions. These concepts are designed for the usual Euclidean space, and they can be easily applied to compositional data if they are represented in this space as well. It turns out that the isometric logratio (ilr) transformation is best suitable in the context of robust estimation. Depending on the method applied, an interpretation of result is usually done in a back-transformed space