Chatterjee Correlation Coefficient: A robust alternative for classic correlation methods in geochemical studies- (including ?TripleCpy? Python package)

Correlation coefficients (CC) are statistical tools that measure how strong a relationship is between two variables. In geochemical studies, these variables could be different elements' concentrations per sample, and the correlation between the concentrations provides a better perspective for the relevant elements in any type of mineralization and commodities. The most common classical CCs in geochemical studies are Pearson's and Spearman's (PCC and SCC). The advantage of such CCs is their simplicity, however, their efficiency could be limited while dealing with non-normal, noisy, closed, or open data (i.e., even after applying log-ratios to the data). Geochemists usually apply SCC to non-normal data, and PCC to normal and open data. What if we need a simple CC that is invariant under monotone transformations of the data, and robust enough to deal with both types of data, including normal/non-normal, noisy, and outliers? Chatterjee CC (CCC: Chatterjee, 2021), as a function of ranks correlation, is a new correlation method with a very simple and understandable formula, and quick computing, but significantly robust to deal with the aforementioned data types without having any assumptions for the variables' distributions. In this research, CCC is introduced to geoscientists further. To apply CCC to datasets, TripleCpy Python package was developed and introduced in this research, then applied to a data sample to demonstrate CCC's efficiency compared to PCC and SCC. This method is highly recommended for future geochemical studies.

Automated summary

This article introduces commonly used correlation coefficient (CC) methods in geochemical studies and discusses their advantages and limitations when dealing with different types of data. It then introduces a new correlation method, Chatterjee CC (CCC), which is simple, computationally efficient, and robust to various types of data. A Python package called TripleCpy was developed to apply CCC to datasets and its efficiency was demonstrated compared to traditional methods.