Exploratory data analysis (EDA) involves graphical and spatial interpretation of the distributions of individual elements, an assessment of correlation between two or more elements, and unsupervised multivariate analysis of compositional data. It is an essential step in data analysis. Its purpose is to gain an understanding of the processes influencing geochemical data. An understanding of raw data distributions allows for the selection of appropriate data transformations as a precursor to parametric or robust statistical analysis of the data. Log transforms are used to de-skew mainly trace element distributions, although it should be recognised that many data sets contain multiple populations. Data may be levelled to correct for those processes identified through EDA. Compositional data is subject to geochemical closure which has the effect of creating false correlations, particularly where major elements vary due to fractionation, hydrothermal alteration or weathering. These effects are removed using the logs of element ratios. Correlation and regression analysis are used to infer geochemical processes from the data. The use of residuals following regression of a dependent variable against a controlling variable may be more informative than analysis of raw data. Statistical outliers, known as geochemical anomalies, can be identified in the data using well-established techniques.
Arne et al. (Fri,) studied this question.