Probabilities of exoplanet signals from posterior samplings

TL;DR

This paper introduces a simple, accurate method for estimating marginal likelihoods from posterior samples, improving model selection in exoplanet detection by reliably comparing models with different numbers of planets.

Contribution

The authors present a novel truncated posterior mixture estimate for marginal likelihoods that is more accurate and versatile than existing methods, applicable to various Bayesian models.

Findings

The method accurately estimates model probabilities in simulated scenarios.

It outperforms the deviance information criterion in convergence and accuracy.

Applied to real data, it reliably assesses the probability of an exoplanet orbiting HD 3651.

Abstract

Estimating the marginal likelihoods is an essential feature of model selection in the Bayesian context. It is especially crucial to have good estimates when assessing the number of planets orbiting stars when the models explain the noisy data with different numbers of Keplerian signals. We introduce a simple method for approximating the marginal likelihoods in practice when a statistically representative sample from the parameter posterior density is available. We use our truncated posterior mixture estimate to receive accurate model probabilities for models with differing number of Keplerian signals in radial velocity data. We test this estimate in simple scenarios to assess its accuracy and rate of convergence in practice when the corresponding estimates calculated using deviance information criterion can be applied to receive trustworthy results for reliable comparison. As a test case, we determine the posterior probability of a planet orbiting HD 3651 given Lick and Keck radial velocity data. The posterior mixture estimate appears to be a simple and an accurate way of calculating marginal integrals from posterior samples. We show, that it can be used to estimate the marginal integrals reliably in practice, given a suitable selection of parameter \lambda, that controls its accuracy and convergence rate. It is also more accurate than the one block Metropolis-Hastings estimate and can be used in any application because it is not based on assumptions on the nature of the posterior density nor the amount of data or parameters in the statistical model.