CausalML Book Ch11: DAGs: Good and Bad Controls for Causal Inference

This episode focuses on causal inference and the selection of control variables within the framework of Directed Acyclic Graphs (DAGs). It explains various strategies for constructing valid adjustment sets to identify average causal effects, such as conditioning on parents or common causes of treatment and outcome variables. The text differentiates between "good" and "bad" controls, emphasizing how conditioning on certain pre-treatment or post-treatment variables can introduce or amplify bias. Through examples like M-bias and collider bias, the authors illustrate scenarios where adjusting for seemingly innocuous variables can lead to incorrect causal conclusions. Ultimately, the excerpt provides guidance on robust methods for causal identification while cautioning against common pitfalls in empirical research.

Disclosure

• The CausalML Book: Chernozhukov, V. & Hansen, C. & Kallus, N. & Spindler, M., & Syrgkanis, V. (2024): Applied Causal Inference Powered by ML and AI. CausalML-book.org; arXiv:2403.02467.
• Audio summary is generated by Google NotebookLM https://notebooklm.google/
• The episode art is generated by OpenAI ChatGPT