1 Quick Overview
The posterior distribution is obtained from the prior distribution and sampling model via Bayes’ rule:
\[p(\theta \mid y)=\frac{p(y \mid \theta) p(\theta)}{\int_{\Theta} p(y \mid \theta') p(\theta') d \theta'}.\]
1.1 Why Bayesian?
- Intuitive probability interpretation: Directly quantifies uncertainty about parameters as probability distributions
- Incorporates prior knowledge: Systematically combines domain expertise with data through the prior distribution
- Principled inference: Bayes’ rule provides a coherent framework for updating beliefs based on evidence
- Natural handling of uncertainty: Posterior distributions capture full uncertainty, not just point estimates
- Sequential analysis: Easily updates beliefs as new data arrives (posterior becomes new prior)
- Small sample inference: Performs well with limited data by leveraging prior information
- Prediction with uncertainty: Generates predictive distributions that quantify uncertainty in future observations
- Decision-making: Naturally incorporates loss functions for optimal decision rules
- Model comparison: Bayes factors provide a principled approach to comparing competing models
1.2 Some Bayesian Topics and their Computational Focus
| Topics | Key Concepts / Readings | Computing Focus |
|---|---|---|
| Introduction to Bayesian Thinking | Bayesian vs. Frequentist paradigms; Prior, likelihood, posterior | Review of R basics and reproducible workflows |
| Bayesian Inference for Simple Models | Conjugate priors, Beta-Binomial, Normal-Normal, Poisson-Gamma | Simulating posteriors, visualization |
| Prior Elicitation and Sensitivity | Informative vs. noninformative priors, Jeffreys prior | Prior sensitivity plots |
| Monte Carlo Integration | Law of large numbers, sampling-based inference | Random sampling and Monte Carlo approximation |
| Markov Chain Monte Carlo (MCMC) | Metropolis-Hastings, Gibbs sampler | Implementing MCMC in R |
| Convergence Diagnostics | Trace plots, autocorrelation, Gelman–Rubin statistic | coda, rstan, and bayesplot packages |
| Hierarchical Bayesian Models | Partial pooling, shrinkage, multilevel structures | rstanarm / brms |
| Midterm Project: Bayesian Linear Regression | Posterior inference for regression, model selection | brms, rstanarm, custom Gibbs samplers |
| Bayesian Model Comparison | Bayes factors, BIC, DIC, WAIC, LOO | Practical comparison via cross-validation |
| Model Checking and Diagnostics | Posterior predictive checks, residual analysis | pp_check in brms |
| Advanced Computation | Hamiltonian Monte Carlo (HMC), Variational Inference | Using Stan and CmdStanR |
| Bayesian Decision Theory | Utility functions, decision rules, loss minimization | Simple decision problems in R |
| Modern Bayesian Methods | Approximate Bayesian computation (ABC), Bayesian neural networks | Examples via rstan or tensorflow-probability |
| Student Project Presentations | Applications and case studies | Full workflow demonstration in R |
1.3 Interesting Article:
- Goligher, E.C., Harhay, M.O. (2023). What Is the Point of Bayesian Analysis?, American Journal of Respiratory and Critical Care Medicine, 209, 485–487.