Title: The log-ratio methodology: Major concepts, robustness, and practical use
Authors: Peter Filzmoser - Vienna University of Technology (Austria) [presenting]
Abstract: In compositional data analysis, the interest is in analyzing relative information rather than the measured or observed values directly. This can be done by considering the log-ratios between all pairs of variables, and - to avoid over-parametrization - by constructing an orthonormal basis describing this information. Since (log-)ratios are taken, one could multiply the values of one observation by a positive constant without changing this relative information. This implies that compositional data analysis is not only limited to data sets with a constant sum of the observations - a frequent misunderstanding. We will introduce some of the major concepts of the log-ratio methodology for compositional data, and present their use in methods for multivariate data analysis. Special attention is given to robust statistical methods.