Statistical Efficiency of Thinning Markov Chain Samplers

A paper by Art B. Owen from Stanford University analyzes the statistical efficiency of thinning Markov chain Monte Carlo (MCMC) output. It provides theoretical examples and bounds, showing thinning can improve efficiency when computation cost per sample is high and autocorrelations decay slowly. Supplementary materials are available online.

Use Cases

• Optimizing MCMC thinning frequency based on computational cost and autocorrelation structure described in the paper.
• Analyzing statistical efficiency trade-offs for AR(1) processes as discussed in the theoretical framework.
• Benchmarking MCMC sampling strategies using the provided efficiency bounds for bounded autocorrelations.

Strengths

• Authored by a researcher from Stanford University, suggesting academic rigor.
• Provides specific theoretical conditions for optimal thinning, such as when θ > (1 - ρ)²/(2ρ) for AR(1) processes.
• Includes supplementary materials online for further detail.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Row count and column-level documentation are absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

paperswithcode

Collection Method

Likely a theoretical paper with accompanying analysis or simulation results.