Overview
George Casella (1951–2012) was a statistician at Cornell University (Biometrics Unit) and later the University of Florida. He is best known for accessible expositions of computational Bayesian methods, particularly the Gibbs sampler, as well as foundational work on improper posteriors and convergence failures in MCMC.
Key Contributions
- "Explaining the Gibbs Sampler" (1992) (with Edward I. George): Tutorial derivation of Gibbs sampler convergence via Markov chain transition matrices; Rao-Blackwell density estimation; failure mode for improper marginals.
- "Rao-Blackwellisation of Sampling Schemes" (1996) (with Christian P. Robert): Post-simulation improvement for Accept-Reject and Metropolis estimators by integrating out ancillary uniform variables; O(n²) recurrence; 15–60% MSE reduction; RB importance sampling beats RB Metropolis.
- Hobert and Casella (1996): Formal analysis of functional compatibility and Gibbs sampling with improper posteriors — establishes conditions under which chains with improper targets are recurrent vs. transient.
Related