Tutorials provide hands-on learning experiences to help attendees build practical skills in Stan, probabilistic programming languages, and probabilistic modelling. Instructors guide participants through structured examples, code, and workflows. Tutorials are designed to be accessible while still valuable for users at varying levels.

Presentation of confirmed tutorials

An introduction to applied Bayesian statistics with Stan

This full-day tutorial is an introduction to Bayesian data analysis using Stan. We begin with the fundamentals of Bayesian inference and the Bayesian workflow. We then survey the Stan ecosystem: the Stan probabilistic programming language, the inference algorithms that fit Stan models, the interfaces that call them from R and Python, and the downstream tools for diagnostics, posterior analysis, and model comparison.

Next, we introduce BRMS, a high-level R interface to Stan, and show how its design supports the Bayesian workflow. The tutorial closes with a worked example of a multilevel model implemented directly in Stan. By day’s end, participants will be able to run Stan and BRMS on their own data in their own compute environments. They will also be prepared to engage with the models and techniques presented in the rest of StanCon, and to apply workflow principles when evaluating that work or their own.

Organizers/instructors

Mitzi Morris, independent consultant, Columbia University.

Bayesian model diagnostics: Workflows and software tools

Course description
This tutorial provides an overview and hands-on experience of diagnostics for Bayesian models. We will present software tools for using concepts from recent Bayesian workflow research in practice. Along with gaining an understanding of the underlying concepts, participants will learn to use software such as the R packages posterior, bayesplot, and priorsense, and the Python package ArviZ. Exercises will be provided in both R and Python to accommodate each participant’s preferred language.

Topics
The material is structured around a modern Bayesian workflow, emphasizing the iterative nature of model building, fitting, evaluation, and critique. Rather than treating diagnostics as isolated checks, we frame them as integral tools for guiding model revision and improving the reliability of our models.

We will work primarily with posterior draws that have already been generated. This allows us to focus on diagnostics and workflow rather than on the details of fitting individual models. Using these draws, participants will explore key components of this workflow, including basic manipulation of posterior draws, checking convergence with R-hat (including rank-normalized and nested variants), effective sample size (ESS), and Pareto-k diagnostics, and model evaluation and critique with prior sensitivity checks and visual predictive checks, including PIT ECDFs, calibration plots, and interval-based summaries.

The focus will be on providing the latest recommendations based on recent research, and showing how to translate these into practice.

Organizers/instructors

Osvaldo Martin – Research fellow at Aalto University. Core developer of ArviZ and other packages for Bayesian statistics and probabilistic programming.

Teemu Säilynoja – Data Scientist at PyMC Labs, and a doctoral candidate focusing on model calibration assessment. Contributor to bayesplot, posterior, SBC, as well as PyMC-Marketing.

Noa Kallioinen – Doctoral researcher in Bayesian workflow at Aalto University with focus on prior specification and sensitivity checking. Lead developer of priorsense, and contributor to posterior.