Bayesian Statistics Overview

Principles

Definition

Bayesian statistics is a statistical inference framework that utilizes Bayes' theorem to update the probability for a hypothesis or parameter estimate as more evidence or data becomes available, providing a principled method for incorporating prior knowledge and quantifying uncertainty.

Ontological type

Core Concepts

Inference Methods

Modeling Frameworks

Key developments

Key Figures

Computational Foundations era

A. F. M. Smith [1], associated with Park University [3] and University College London [4], contributed during the Computational Foundations era. His key contribution in this era is reflected in the paper Sampling-Based Approaches to Calculating Marginal Densities [7], which introduced sampling-based methods to approximate marginal densities and made Bayesian posterior calculations tractable for complex models. Alan E. Gelfand [2], with appointments at Stanford University [5] and University of Minnesota [6], contributed to this era. Gelfand's contributions include the same Sampling-Based Approaches to Calculating Marginal Densities [7], which advanced Bayesian computation by enabling practical, scalable inference through sampling methods, shaping how complex models were analyzed in statistics and AI communities.

MCMC Revolution era

The MCMC revolution established Gibbs sampling and Metropolis-Hastings as the practical backbone of Bayesian computation, with Gelfand and Smith (1990) showing how sampling from full conditionals enables efficient posterior inference for complex models. Gelman, Carlin, Stern, Dunson, and Vehtari popularized a coherent Bayesian workflow, advancing hierarchical modeling, convergence diagnostics such as the Gelman-Rubin statistic, and careful post-processing to address identifiability in high dimensions. Radford Neal's work on scalable MCMC and Bayesian neural networks in the 1990s introduced slice sampling and advanced proposal schemes that extended MCMC to large, structured models. Haario, Saksman, and Tamminen developed adaptive Metropolis methods in the early 2000s, enabling automatic tuning of proposals and more efficient exploration of complex posterior landscapes.

Scalable Pragmatic Workflows era

Andrew Rambaut [1] is a prominent figure with affiliations at Harvard University [3] and the University of California, Los Angeles [4]. Posterior Summarization in Bayesian Phylogenetics Using Tracer 1.7 [7] advanced practical summarization of posterior distributions in phylogenetics, enabling scalable, reproducible workflows and decision-focused reporting. Paul-Christian Bürkner [2] is a leading figure linked with Heidelberg University [5] and Ghent University [6]. Paul-Christian Bürkner [2] further contributed through brms: An R Package for Bayesian Multilevel Models Using Stan [8], which democratizes Bayesian multilevel modeling and supports scalable, reproducible workflows.