This episode explores the foundational concepts of linear regression as a tool for predictive inference and association analysis. It details the Best Linear Prediction (BLP) problem and its finite-sample counterpart, Ordinary Least Squares (OLS), emphasizing their statistical properties, including analysis of variance and the challenges of overfitting when the number of parameters is not small relative to the sample size. The text further introduces sample splitting as a method for robustly evaluating predictive models and clarifies how partialling-out helps in understanding the predictive effects of specific regressors, such as in analyzing wage gaps. Finally, it discusses adaptive statistical inference and the behavior of OLS in high-dimensional settings where traditional assumptions may not hold.

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

This episode explores a powerful method for identifying causal effects in non-experimental settings. The authors, affiliated with various universities, explain the basic RDD framework, where treatment assignment is determined by a running variable crossing a cutoff value. The text highlights how modern machine learning (ML) methods can enhance RDD analysis, particularly when dealing with numerous covariates, improving efficiency and allowing for the study of heterogeneous treatment effects. An empirical example demonstrates the application of RDD and ML techniques to analyze the impact of an antipoverty program in Mexico.

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

This episode introduces and explains the Difference-in-Differences (DiD) framework, a widely used method in social sciences for estimating causal effects in situations with treatment and control groups over multiple time periods. It elaborates on the core assumption of "parallel trends" and discusses how Debiased Machine Learning (DML) methods can be used to incorporate high-dimensional control variables, enhancing the robustness of DiD analysis. The text illustrates these concepts with a practical example applying DML to study the impact of minimum wage changes on teen employment, analyzing different machine learning models and assessing their performance. The authors also briefly touch on more advanced DiD settings, such as those involving repeated cross-sections, and provide exercises for further study.

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