Overview
Steven N. MacEachern is a statistician at Ohio State University. He is best known for proving, with L. Mark Berliner, that systematic subsampling (thinning) of a Gibbs sampler output never improves posterior mean estimates — a result that settled the common practitioner intuition to "thin to reduce autocorrelation."
Key Contributions
- Co-authored MacEachern-Berliner (1994), the canonical three-page proof that thinning a stationary Markov chain is always variance-inefficient compared to using the full chain.
- Extended Geyer's (1992) analogous result by requiring only stationarity rather than reversibility, thereby covering the standard fixed-scan Gibbs sampler.
Related