On thinning of chains in MCMC

By W. Link, M. Eaton. 2012

366 citations recorded by the source corpus.

Subfields: Environmental Chemistry, Environmental Engineering, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/f1d47f13d8ee38ce9b4b234b9ce6ccf826760dff

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Environmental Chemistry

In Environmental Engineering

In Statistics and Probability

Citations in the corpus

Improving Transparency and Replication in Bayesian Statistics: The WAMBS-Checklist (2017, neutral): “Although it is true that univariate priors used for variances (or standard deviations/ precisions) have some similar complexities (see, e.g., van de Schoot et al., 2015), we highlight some of the issues specific to the multivariate treatment of these priors. In our experience, the multivariate prior”
Partition MCMC for Inference on Acyclic Digraphs (2015, methodology): “Since processing the DAGs after they have been sampled can be quite computationally expensive, thinning is often appropriate (Link and Eaton 2012). By modifying the original algorithm of Madigan and York (1995) to include the possibility of reversing an edge, the convergence speed can be greatly imp”
Bayesian estimation supersedes the t test. (2013, background): “It turns out that in most typical applications, clumpiness can be adequately smoothed simply by running a long chain without thinning, and the long chain produces reliable estimates of the posterior distribution (e.g., Link & Eaton, 2012; Jackman, 2009, p. 263).”
Support points (2016, background): “This is performed for several reasons (see [37]): it reduces high autocorrelations in the MCMC chain, saves computer storage space, and reduces processing time for computing derived posterior quantities.”
Evaluating population viability and efficacy of conservation management using integrated population models (2018, methodology): “We ran three independent chains for 400,000 iterations after a burn- in of 200,000 iterations and thinned chains by 10 to reduce autocorrelation (Link & Eaton, 2012).”