A pragmatic Bayesian perspective on correlation analysis: The exoplanetary gravity - stellar activity case

TL;DR

This paper demonstrates how Bayesian methods can be applied to correlation analysis, specifically between exoplanetary gravity and stellar activity, providing more robust and informative results than traditional p-value approaches.

Contribution

It introduces a practical, Python-based Bayesian tool for correlation assessment, highlighting its advantages over classical methods and encouraging wider adoption in scientific research.

Findings

Bayesian analysis supports a correlation between planetary gravity and stellar activity.

Results are more robust and informative than traditional p-value methods.

The provided Python code is user-friendly and adaptable for various problems.

Abstract

We apply the Bayesian framework to assess the presence of a correlation between two quantities. To do so, we estimate the probability distribution of the parameter of interest, $\rho$, characterizing the strength of the correlation. We provide an implementation of these ideas and concepts using python programming language and the pyMC module in a very short ($\sim$130 lines of code, heavily commented) and user-friendly program. We used this tool to assess the presence and properties of the correlation between planetary surface gravity and stellar activity level as measured by the log($R'_{\mathrm{HK}}$) indicator. The results of the Bayesian analysis are qualitatively similar to those obtained via p-value analysis, and support the presence of a correlation in the data. The results are more robust in their derivation and more informative, revealing interesting features such as asymmetric posterior distributions or markedly different credible intervals, and allowing for a deeper exploration. We encourage the reader interested in this kind of problem to apply our code to his/her own scientific problems. The full understanding of what the Bayesian framework is can only be gained through the insight that comes by handling priors, assessing the convergence of Monte Carlo runs, and a multitude of other practical problems. We hope to contribute so that Bayesian analysis becomes a tool in the toolkit of researchers, and they understand by experience its advantages and limitations.