No. 155 of 500 No. 27 in Machine Learning

Statistical Rethinking

A Bayesian Course with Examples in R and Stan

By Richard McElreath

Published
March 16, 2020
Print length
612 pages
Reading level
Intermediate
ISBN-13
978-0367139919
Publisher
Chapman and Hall/CRC
Written for
Data scientists, ML engineers and researchers who want rigorous intuition for Bayesian modeling and causal inference.

Synopsis

Richard McElreath wrote Statistical Rethinking for scientists who were taught statistics as a list of tests and want to understand what their models actually do. He treats every statistical model as a golem, a powerful but mindless machine that does exactly what it is told, and builds Bayesian inference from first principles: counting possibilities, sampling from posterior distributions, and fitting linear, generalized linear and multilevel models step by step in R and Stan. Readers write the calculations themselves before the book hands any of them to software, so the logic stays visible.

The second edition, published in 2020, makes causal reasoning central. Directed acyclic graphs appear throughout to show when adding a variable helps an analysis and when it quietly creates confounding, and new material covers prior design, splines, cross-validation, instrumental variables and Hamiltonian Monte Carlo. For machine learning practitioners, the payoff is a working grasp of uncertainty, overfitting and causal identification that many prediction-focused courses skip. McElreath's free recorded lectures and community translations of the code into Python, Julia and brms make the book usable as a complete self-study course.

Scorecard

Editorial score

How it scored

Editorial score: 84.3 / 1005 weighted criteria

  • Influence30%

    25.5/30
  • Authority25%

    21.3/25
  • Practical value20%

    17.0/20
  • Currency15%

    12.0/15
  • Readability10%

    8.5/10
  • Editorial score

    84.3/100

How the rank was computed

Rank No. 155 of 500 comes from a composite of 59.3: observable signals 21.3 of 55 points plus editorial review 38.0 of 45 points.

All 11 inputs
InputPoints
Observable signals21.3/55
Reach (Amazon ratings volume)9.5/16.5
Consensus across best-of lists2.8/13.8
Awards and bestseller evidence0.0/8.3
Reader rating, sample-adjusted7.9/8.3
Amazon sales rank1.1/5.5
Recency0.0/2.8
Editorial review38.0/45
Author authority9.6/11.3
Influence11.5/13.5
Practical value7.7/9.0
Readability3.8/4.5
Currency5.4/6.8

Each criterion is scored from 1 to 10 and weighted, and the points add up to the editorial score. The rank combines observable signals (55 percent) with this score (45 percent). Read the methodology.

Verdict

A highly readable route into Bayesian modeling and causal inference that teaches data analysis as careful scientific reasoning rather than ritual.

Strengths

  • Computational, code-first teaching of Bayesian ideas
  • Causal inference integrated throughout the text
  • Free lectures and multi-language code ports

Keep in mind

Examples rely on McElreath's own rethinking R package, and readers seeking deep learning or large-scale predictive methods will find little coverage here.

Key takeaways

  1. Statistical models are golems: powerful, literal-minded tools that answer exactly the question encoded in them, whether or not it is the right one.
  2. Causal diagrams reveal when controlling for a variable removes bias and when it introduces confounding, so model choice must follow a stated causal theory.
  3. Multilevel models pool information across groups, producing better estimates for small samples and serving as a sensible default for clustered data.
  4. Overfitting is managed with regularizing priors and with information criteria or cross-validation that estimate out-of-sample predictive accuracy.

Notable facts

  • The International Society for Bayesian Analysis lists the second edition among its DeGroot Prize winners, under 2023, an award for a published book in statistical science. Source (opens in a new tab)
  • McElreath publishes free lecture videos for the course, plus links to code translations for brms, PyMC, NumPyro and Julia. Source (opens in a new tab)

About the author

Richard McElreath

Richard McElreath is an evolutionary anthropologist and director of the Department of Human Behavior, Ecology and Culture at the Max Planck Institute for Evolutionary Anthropology in Leipzig, where he has worked since 2015. He was previously a professor of anthropology at the University of California, Davis, holds a PhD in anthropology from UCLA, and studies the role of culture in human evolution.

Questions

Is Statistical Rethinking good for machine learning practitioners?
Yes, particularly for those who build models without a strong statistics background. It teaches uncertainty, regularization, overfitting and causal reasoning through hands-on Bayesian modeling. It does not cover neural networks or large-scale prediction pipelines, so it complements rather than replaces a machine learning textbook.
Which edition of Statistical Rethinking should I read?
Read the second edition, published by CRC Press in 2020. It adds directed acyclic graphs for causal inference throughout, plus material on prior design, cross-validation, instrumental variables and Hamiltonian Monte Carlo. McElreath's website notes that a third edition is in progress, with no publication date announced.
Do I need to know R to use Statistical Rethinking?
Basic familiarity helps, because the book's examples use R and Stan through McElreath's rethinking package. The code is introduced gradually, and community translations exist for Python libraries such as PyMC and NumPyro, for Julia and for the brms package, so readers can follow along in their preferred language.

Sources

As an Amazon Associate, Best AI Books earns from qualifying purchases. Best AI Books is sponsored by Iternal.ai, whose founder and CEO, John Byron Hanby IV, wrote The AI Strategy Blueprint, our #1 ranked book. How we rank books.