Topics in Bayesian Causal Inference
Seminar · TU Dortmund · Winter semester 2026/27
Prof. Dr. Alexander Marx and Dr. Javier Aguilar
Opening session: 16 October 2026, 10:00
Welcome to the course website for the seminar Topics in Bayesian Causal Inference, taught at TU Dortmund in the winter semester 2026.
We present general information, news, the paper list, and more here.
Summary
Causal questions often concern the effect of an intervention. In many empirical settings, the goal is to estimate how an outcome would change under a treatment policy, exposure, or decision. This can become challenging in the presence of observational data since treated and untreated units may differ systematically, relevant confounders may be high-dimensional, and the causal quantities of interest depend on potential outcomes that are not jointly observed.
Bayesian inference approaches these challenges by placing probability distributions on unknown quantities and updating them in light of the observed data. In causal inference, this provides a coherent way to model missing potential outcomes, flexible outcome regressions, treatment-assignment mechanisms, and heterogeneous treatment effects while propagating the associated uncertainty. Bayesian inference does not, however, identify causal effects by itself. Causal identification still depends on assumptions linking the observed data to the counterfactual quantities of interest, including consistency, conditional exchangeability, and overlap.
In this seminar, we focus on Bayesian methods for estimating causal effects, with particular emphasis on observational studies, and extend Bayesian causal inference to mediation, noncompliance, and other complex treatment settings. Our central theme is the interaction between causal identification and Bayesian modelling, that is, how flexible models, prior information, and posterior uncertainty can support treatment-effect estimation while keeping the underlying causal assumptions explicit.
Our interests
The idea of the seminar is to read selected research papers on Bayesian causal inference and ask how Bayesian methods help us:
- define causal estimands using potential outcomes and counterfactuals
- deal with non-random treatment assignment and confounding
- model heterogeneous treatment effects
- use priors and regularization in complex causal models
- quantify uncertainty about missing counterfactual outcomes and causal effects.
The introductory lectures will cover the necessary fundamentals of Bayesian inference.
Our opening session on 16 October 2026 explains the seminar format, introduces the paper list, and describes how topics will be assigned.
Prerequisite: a previous course in Causality or Causal Inference.
Contact and help
Feel free to send us an email with questions about the seminar and topic assignment. Please put [BCI seminar] in the subject line.