Model-based replacement of rounded zeros in CODA: classical and robust approaches

Abstract
The log-ratio methodology represents a powerful set of methods and techniques to apply in the statistical analysis of compositional data (CODA). These techniques are used for the estimation of rounded zeros or values below the detection limit, in case that the underlying data are of compositional nature. An algorithm based on iterative log-ratio regressions is developed, combining a particular family of isometric log-ratio transformations with censored regression. We proof that classical and robust regression methods are equivalent in this context when they are based on additive or isometric log-ratio transformations. Based on Monte Carlo experiments, simulations are performed to assess the performance of classical and robust methods. To illustrate the introduced method, a real study involving geochemical data is carried out.
Year
2012
Category
Refereed journal
Output Tags
WP7.2 - Enhancing health benefits from food through production and processing