Statistical Rethinking (2026 Edition)

Statistical Rethinking (2026 Edition) by Richard McElreath (MPI-EVA). A practical course on Bayesian data analysis, causal inference with DAGs, generative modeling, multilevel models, and computationa

• Instructor: Richard McElreath (Director, Department of Human Behavior, Ecology and Culture, Max Planck Institute for Evolutionary Anthropology, Leipzig)

• Repository: rmcelreath/stat_rethinking_2026

• Course Dates: 6 January – 13 March 2026

• Book: Statistical Rethinking: A Bayesian Course with Examples in R and Stan (2nd Edition, CRC Press / Chapman & Hall; with draft chapters for the 3rd Edition) — Book Site

• Software: R package rethinking via GitHub, interfacing cmdstanr / Stan; conversions available in Python (PyMC), Julia (Turing.jl), etc.

---

Overview & Philosophy

"The unfortunate truth about data is that nothing much can be done with it, until we theorize about what caused it. Therefore the meaning of any statistical estimate depends upon assumptions outside the data and statistical model. So we will prioritize these outside assumptions: causal models, how to analyze them, and how to use them to construct scientifically meaningful statistical procedures. We will use Bayesian data analysis to connect scientific models to data. And we will learn powerful computational tools for coping with high-dimension, imperfect data of the kind that biologists and social scientists face." — Richard McElreath

Statistical Rethinking shifts statistical practice away from the standard "Ptolemaic" menu of arbitrary null-hypothesis significance tests (t-tests, ANOVA, p-values) towards principled scientific modeling:

1. Causal Inference First: Using Directed Acyclic Graphs (DAGs) to identify confounders, colliders, pipes, and forks before formulating regression equations.

2. Generative Modeling: Designing a synthetic data generation process to verify that an estimator can recover known ground truth before applying it to empirical observations.

3. Bayesian Logic: Representing uncertainty through joint posterior probability distributions rather than point estimates, null distributions, and p-values.

4. Computational Workflow: Employing Markov Chain Monte Carlo (MCMC) and Hamiltonian Monte Carlo (HMC) via Stan (cmdstanr) to fit realistic, high-dimensional models.

---

2026 Edition Structure: Two Tracks

The 2026 edition splits instruction into two concurrent 10-week tracks running at MPI-EVA, pacing material at half-speed compared to previous intensive runs:

• Section A (Beginner): Geared towards researchers new to causal inference and Bayesian regression modeling. Covers Chapters 1–12 of Statistical Rethinking (2nd edition).

  • Playlist: Section A (Beginner) Playlist

• Section B (Experienced): Geared towards researchers who already understand the foundations and want advanced topics. Starts immediately with multilevel/hierarchical models, social networks, Gaussian processes, and generalized linear models.

  • Playlist: Section B (Experienced) Playlist

• Full Combined 2026 Playlist: 2026 Chronological Playlist

• Reference Archive: Statistical Rethinking 2023 Playlist

---

Calendar & Topical Outline

Week	Section	Topic	Lecture Link	Reading
01	Beginner	Introduction	Lecture	Chapters 1 & 2
	Experienced	Multilevel Models	Lecture	Chapter 12
02	Beginner	Garden of Forking Data	Lecture	Chapters 2 & 3
	Experienced	Multilevel Model Expansion	Lecture	Chapter 13
03	Beginner	Geocentric Models	Lecture	Chapter 4
	Experienced	Correlated Features	Lecture	Chapter 13
04	Beginner	Categories & Causes	Lecture	Chapter 4
	Experienced	Group-level Confounds / Social Networks I	Lecture	Chapter 14
05	Beginner	Estimands and Estiplans	Lecture	Chapters 4 & 5
	Experienced	Social Networks II	Lecture	Chapter 14
06	Beginner	Elemental Confounds I	Lecture	Chapter 6
	Experienced	Gaussian Processes	Lecture	Chapter 15
07	Beginner	Good and Bad Controls	Lecture	Chapter 6
	Experienced	Measurement Models	Lecture	Chapter 15
08	Beginner	MCMC and Item Response Models	Lecture	Chapters 9 & 10
	Experienced	Missing and Censored Data	Lecture	Chapter 15
09	Beginner	Modeling Events	Lecture	Chapters 10 & 11
	Experienced	Generalized Linear Madness	Lecture	Chapter 16
10	Beginner	Confounds & Sensitivity Analysis	Lecture	Chapter 12
	Experienced	Hidden Markov Models	Lecture	Chapter 16

---

Methodological Themes for Biomedical & Pathology Research

1. DAGs and Non-Causal Associations: In clinical pathology studies, multivariable regression models frequently suffer from the "Table 2 Fallacy" — reporting coefficients for all covariates as though they have identical causal validity. McElreath illustrates why adjusting for mediators or colliders introduces bias, whereas conditioning on genuine backdoor paths correctly isolates the causal estimand.

2. Multilevel Shrinkage & Partial Pooling: Particularly critical for pathology and clinical data where observations are nested within patients, biopsy cores, staining batches, laboratory sites, or scanner devices. Varying intercepts and varying slopes share information across groups, adaptively shrinking noisy estimates towards the grand mean without underestimating uncertainty.

3. Measurement Error & Observer Variability: Histopathological grading (e.g. tumour cell percentage, Gleason score, Nottingham histological grade) involves subjective error and inter-observer disagreement. Rather than treating subjective scores as immutable fixed covariates, Bayesian measurement error models incorporate uncertainty directly into the parameter likelihoods.

4. Continuous Spatial Modeling with Gaussian Processes: Provides principled covariance functions over physical distances, applicable to spatial transcriptomics, spatial proteomics, and tissue microarchitecture analysis.