Inference Fundamentals

Routing Summary

This folder covers the foundations of Bayesian inference from BDA3 Part I and Statistical Rethinking Chapters 1-3. Contains 8 notes plus the Empirical Bayes sub-topic.

Sub-topics

Sub-topicNotesCovers
Empirical Bayes5Robbins’ formula, the James-Stein estimator, Stein’s paradox / risk dominance, parametric EB and the empirical-Bayes view of shrinkage — Efron, Large-Scale Inference Ch. 1

Concept Map

ConceptNoteTypeDepends OnKey Result
Bayes’ theorem, three steps of Bayesian analysisProbability and Bayesian Inferenceconceptraw/BDA3.pdfPosterior = Prior x Likelihood / Evidence
Conjugate priors, beta-binomial, noninformative priorsSingle-Parameter ModelsconceptProbability and Bayesian InferencePosterior is a compromise between prior and data
Nuisance parameters, marginal posteriors, bioassayMultiparameter ModelsconceptProbability and Bayesian Inference, Single-Parameter ModelsMarginalization integrates out nuisance parameters
Normal approximation, Bernstein-von MisesAsymptotics and Frequentist ConnectionsconceptMultiparameter Models, Probability and Bayesian InferencePosteriors converge to normal with enough data
Exchangeability, partial pooling, eight schoolsHierarchical ModelsconceptSingle-Parameter Models, Probability and Bayesian Inference, Multiparameter ModelsPartial pooling shrinks toward group mean
Z-score shrinkage, variance ratio, simulation evidencePartial Pooling as Multiple Comparisons CorrectiontheoremHierarchical Models, Multiple Comparisons - Bayesian Perspective, Type S and Type M ErrorsShrinkage factor = always < 1
Philosophy of statistical modelingStatistical Rethinking - The Golem of Pragueoverviewraw/StatRethink-Bayes.pdfHypotheses != models; all models are golems
Bayesian updating, grid approximationGarden of Forking DataconceptStatistical Rethinking - The Golem of Prague, Probability and Bayesian InferenceSmall world vs large world distinction
Working with posterior samples, HPDI, posterior predictivePosterior Sampling and SummarizationconceptGarden of Forking Data, Probability and Bayesian InferenceSummarize posteriors with intervals and predictions

Notes

Sources

  • BDA3.pdf — Bayesian Data Analysis, 3rd Edition (Gelman et al.), Part I (pp. 1-137)
  • StatRethink-Bayes.pdf — Statistical Rethinking (McElreath, 2015), Chapters 1-3
  • multiple2f.pdf — “Why we (usually) don’t have to worry about multiple comparisons” (Gelman, Hill & Yajima, 2009)