Regression Models

Routing Summary

This folder covers Bayesian regression from BDA3 Part IV, Statistical Rethinking Chapters 4-5, and PyMC tutorials. Contains 9 notes plus the Shrinkage Priors sub-topic.

Sub-topics

Sub-topicNotesCovers
Shrinkage Priors5Global-local shrinkage, the horseshoe and regularized (Finnish) horseshoe, effective number of nonzeros, choosing the global scale — Piironen & Vehtari (2017)

Concept Map

ConceptNoteTypeDepends OnKey Result
Priors as regularization, ridge/lasso/horseshoeBayesian Linear RegressionconceptProbability and Bayesian Inference, MCMC Basics, Bayesian Workflow - OverviewPriors encode regularization; horseshoe for sparsity
Varying intercepts/slopes, multilevel regressionHierarchical Linear ModelsconceptBayesian Linear Regression, Hierarchical Models, Generalized Linear Models, Efficient MCMCPartial pooling improves out-of-sample prediction
Logistic, Poisson, weakly informative priors for GLMsGeneralized Linear ModelsconceptBayesian Linear Regression, Hierarchical Linear Models, Nonparametric Models OverviewLink functions connect linear predictor to outcome
Multiple imputation, MCAR/MAR/MNAR, Rubin’s rulesMissing Data ModelsconceptData Collection Models, Hierarchical Models, MCMC BasicsMultiple imputation propagates missing-data uncertainty
Bayesian counterfactual prediction, do-operatorCounterfactual InferenceconceptBayesian Linear Regression, Generalized Linear Models, Model Checking, Spurious Association and ConfoundsPosterior predictive sampling implements the do-operator
Interaction effects, moderator analysis, spotlight graphsModeration AnalysisconceptSpurious Association and Confounds, Bayesian Linear Regression, Generalized Linear ModelsModeration = how context changes causal effect size
DAG-based missing data (MCAR/MAR/MNAR via causal graphs)Missing Data - Statistical RethinkingconceptMissing Data Models, Spurious Association and Confounds, Data Collection Models, Counterfactual InferenceDAGs clarify which missing-data mechanism applies
Gaussian models, MAP, prior predictive simulationLinear Models in Statistical RethinkingconceptProbability and Bayesian Inference, Bayesian Workflow - OverviewPrior predictive simulation validates model assumptions
Multivariate regression, confounds, post-treatment biasSpurious Association and ConfoundsconceptLinear Models in Statistical Rethinking, Bayesian Linear Regression, Probability and Bayesian InferenceDAGs identify which variables to include/exclude

Notes

Sources