Bayesian statistics

Contents

Bayesian statistics is a statistical paradigm that updates degrees of belief using probability to represent uncertainty. It interprets probability as a measure of belief rather than frequency and combines prior information with data through a mathematical rule to produce posterior statements. The approach has influenced scientists across Royal Society, University of Edinburgh, Princeton University, Harvard University, and University of Cambridge traditions and is central in many modern research programs at institutions like Massachusetts Institute of Technology, Stanford University, Columbia University, University of California, Berkeley, and University of Oxford.

History

The roots trace to work by Thomas Bayes and posthumous publication in the Philosophical Transactions of the Royal Society, followed by extensions by Pierre-Simon Laplace in texts associated with the French Academy of Sciences and applications in the Napoleonic Wars era. Nineteenth- and early twentieth-century debates involved figures linked to University of Göttingen and University of Cambridge mathematics faculties, while twentieth-century developments engaged researchers associated with Bell Labs, RAND Corporation, Institute for Advanced Study, University of Chicago, and Columbia University. Twentieth-century controversies involved exchanges between proponents at University of California, Berkeley and critics connected to London School of Economics and Princeton University. The late twentieth and early twenty-first centuries saw computational advances at Los Alamos National Laboratory, Argonne National Laboratory, and Lawrence Berkeley National Laboratory that enabled widespread adoption in projects at National Institutes of Health, European Organization for Nuclear Research, NASA, and tech companies like IBM, Google, Microsoft Research, and Facebook.

Bayesian practice centers on updating belief via Bayes's theorem, a rule historically linked to Thomas Bayes and formalized by Pierre-Simon Laplace in correspondence with institutions such as the Académie des Sciences. Foundational debates engaged thinkers associated with Princeton University and Cambridge University Press publications, and later philosophers of science at University of Pittsburgh and University of Michigan who contrasted personalist and objective interpretations. Key principles incorporate prior elicitation techniques used in clinical trials overseen by Food and Drug Administration, regulatory frameworks at European Medicines Agency, and decision-theoretic concepts advanced by scholars affiliated with RAND Corporation and Columbia University. Coherence, exchangeability, and likelihood principles were debated in seminars at Harvard University and workshops at International Statistical Institute.

Bayesian methods include parametric models developed in textbooks originating from faculty at University of Chicago and nonparametric approaches tied to research at University College London and University of California, Berkeley. Hierarchical and multilevel models trace to applications in longitudinal studies at Johns Hopkins University and social science analyses at London School of Economics. Popular models such as conjugate families appear in curricula at Massachusetts Institute of Technology and Stanford University, while mixture models and Dirichlet processes were advanced by researchers at University of Washington and University of Toronto. Model selection and Bayesian model averaging have been applied in projects funded by National Science Foundation and evaluated in workshops at International Biometric Society.

Computational progress owes much to algorithms developed at Bell Labs, numerical libraries from Argonne National Laboratory, and software ecosystems produced by teams at Harvard University, Stanford University, Columbia University, Massachusetts Institute of Technology, and University of Cambridge. Markov chain Monte Carlo methods originated in work linked to Los Alamos National Laboratory and were expanded by scholars who later worked at University of California, Berkeley and Princeton University. Variational inference techniques have been promoted by research groups at Google, Facebook, and Stanford University, while sequential Monte Carlo methods saw applications in projects at NASA and European Space Agency. Probabilistic programming languages and tools developed by teams at University of Oxford, University of Cambridge, Massachusetts Institute of Technology, University College London, and Stanford University have broadened access across industries including Microsoft Research, Amazon, and biotechnology firms collaborating with National Institutes of Health.

Bayesian approaches are widely used in clinical trials overseen by Food and Drug Administration and European Medicines Agency, in genetics projects at National Institutes of Health and Wellcome Trust, and in particle physics experiments at European Organization for Nuclear Research. They inform climate modeling work at Intergovernmental Panel on Climate Change, remote sensing projects at National Aeronautics and Space Administration, and econometric studies linked to World Bank and International Monetary Fund. Machine learning applications appear in research labs at Google, Facebook, DeepMind, and OpenAI, while industry adopters include IBM, Microsoft Research, Amazon Web Services, and financial firms in Wall Street hubs. Conservation biology teams associated with Smithsonian Institution and World Wildlife Fund use hierarchical models, and epidemiological modeling during outbreaks has involved collaborations between Centers for Disease Control and Prevention, World Health Organization, and university public health schools.

Critiques arose in academic exchanges at London School of Economics and Princeton University around subjectivity of priors and frequentist comparisons promoted by statisticians from University of Chicago and Columbia University. Debates over reproducibility engaged editors at Nature and Science and regulatory discussions at Food and Drug Administration and European Medicines Agency. Computational approximations and scalability have been scrutinized in industry fora at Google and Amazon Web Services and at conferences organized by Association for Computing Machinery and NeurIPS Foundation. Ethical concerns about priors in policy settings have been raised in hearings before legislative bodies in United States Congress and committees at European Parliament.