Spurious correlation and the closure property of compositional data in earth sciences

Abstract

 In the field of earth sciences, measurements typically yield compositional data that has a property known as closure. The application of common statistical methods to compositional data results in the exclusion of spurious correlations, which in turn yields findings that are not representative of the underlying data. This article presents a set of transformations for the opening of closed systems of compositional data. These transformations include the additive logarithmic ratio (alr), the centered logarithmic ratio (clr), and the isometric logarithmic ratio (ilr). All of the aforementioned transformations are defined in terms of logarithms of ratios. The clr transformation was then applied to a soil chemical data set. The results of applying cluster analysis on the clr-transformed data were also analyzed using Spearman's correlation coefficient matrix as distance. Furthermore, the impact of the clr transformation on spurious correlations, skewness, and outliers in the data was evaluated using R statistical software.

Key words: Closure property, Compositional data, Log-ratio transformations, Robust statistical method, Spurious correlation.