Richard McElreath is a director at the Max Planck Institute for Evolutionary Anthropology in Leipzig, Germany, whose textbook Statistical Rethinking: A Bayesian Course with Examples in R and Stan has redefined how Bayesian statistics is taught to scientists. Rather than presenting Bayesian inference as a collection of techniques, McElreath frames it as a way of thinking about scientific models, causal assumptions, and the relationship between theory and data. His approach integrates directed acyclic graphs (DAGs), simulation-based reasoning, and hierarchical models into a coherent workflow for scientific inquiry.
Life and Career
Born in the United States. Develops interests in anthropology, evolutionary biology, and quantitative reasoning.
Earns his Ph.D. in anthropology from UCLA, studying cultural evolution and the evolution of human cooperation using formal mathematical models.
Begins developing the course material that will become Statistical Rethinking, teaching Bayesian methods to anthropologists and evolutionary biologists.
Publishes the first edition of Statistical Rethinking, which rapidly gains a devoted following for its clarity, depth, and emphasis on causal reasoning.
Publishes the second edition of Statistical Rethinking, substantially expanding coverage of causal inference with DAGs, measurement error models, and multilevel models. Accompanying lecture videos become widely viewed online.
Statistical Rethinking
McElreath's textbook is distinctive for several reasons. First, it begins not with probability distributions or Bayes' theorem but with the question of what a statistical model is trying to accomplish scientifically. Before fitting any model, students learn to draw causal diagrams (DAGs) that represent their assumptions about the data-generating process, identify which variables to include (and exclude) from a regression, and reason about confounding, colliders, and mediation.
Second, the book builds Bayesian inference from the ground up using grid approximation and quadratic approximation before introducing MCMC, giving students genuine understanding of what the algorithms are computing. Third, it integrates model checking and posterior predictive simulation throughout, rather than treating them as afterthoughts.
McElreath argues that many statistical errors stem not from choosing the wrong test or estimator but from failing to think clearly about the scientific question. A regression coefficient is not a causal effect unless the model's causal assumptions justify that interpretation. By teaching students to reason with DAGs before they write any code, McElreath helps them avoid the most consequential errors in applied statistics, errors that no amount of computational sophistication can fix.
Evolutionary Anthropology and Cultural Evolution
McElreath's scientific research focuses on cultural evolution, social learning, and the evolution of human cooperation. He uses formal mathematical models (often evolutionary game theory) combined with empirical data from small-scale societies to understand how cultural practices spread, persist, and adapt. This work naturally demands sophisticated statistical methods, as the data are often hierarchically structured (individuals within communities within regions), sample sizes are small, and the scientific questions involve causal mechanisms rather than mere associations.
His research provides a compelling case study for why Bayesian methods are valuable in the sciences: they handle small samples gracefully through regularization, they accommodate complex hierarchical data structures, and they force researchers to state their assumptions explicitly in the form of prior distributions and model structure.
Causal Inference Integration
A hallmark of McElreath's approach is the integration of causal inference tools, particularly DAGs, into routine Bayesian workflow. He shows how to use d-separation rules to identify conditional independencies implied by a causal model, how to choose conditioning sets that avoid confounding bias, and how to recognize situations where conditioning on a variable introduces bias rather than removing it (the collider problem). This integration of Pearl's causal framework with Bayesian modeling is one of the most practically useful aspects of his teaching.
"Statistical models are not tests of hypotheses. They are machines for generating predictions from assumptions. The value of a model lies not in whether it is 'significant' but in whether its assumptions are reasonable and its predictions useful." — Richard McElreath, Statistical Rethinking
Legacy
McElreath's influence extends far beyond anthropology. Statistical Rethinking is used across biology, psychology, political science, and ecology. His freely available lecture videos have reached hundreds of thousands of viewers, and community translations of his code examples into Python, Julia, and other languages have expanded the book's reach further. By demonstrating that rigorous Bayesian thinking is accessible to working scientists, McElreath has done as much as anyone to advance the practical adoption of Bayesian methods in the empirical sciences.