Causal Machine Learning

This section contains the Causal ML stream, an eight-notebook sequence on machine learning for causal inference. Notebooks 1 to 3 build a prediction toolkit (train/test splits, regularization, trees and forests), Notebook 4 makes the turn from prediction to treatment effects, and Notebooks 5 to 8 cover the causal tools: post-double-selection lasso, propensity scores, double machine learning, and causal forests. The notebooks are meant to be read in order.

Modules

Prediction, Inference, and Causality

The first notebook in the Causal ML stream. We meet the supervised-learning way of thinking: prediction versus inference, train/test splits, loss functions…

Regularization: Ridge and Lasso

The second notebook in the Causal ML stream. We encounter the ‘too many controls’ problem, where ordinary least squares falls apart once the number of covariates approaches…

Trees, Forests, and Boosting

The third notebook in the Causal ML stream. We show a different way to be flexible: decision trees, which split the data into boxes and predict the local average. A single…

From Prediction to Treatment Effects

The fourth notebook in the Causal ML stream. We move away from prediction to causality. We meet potential outcomes, the average treatment effect, selection bias, and the…

Post-Double-Selection Lasso

The fifth notebook in the Causal ML stream. We return to the growth data from Notebook 2 and the convergence question. A naive regression shows no convergence, and so does a…

Propensity Scores and Doubly-Robust Estimation

The sixth notebook in the Causal ML stream. We return to the 401(k) question and adjust for confounders a second way: by modelling who gets treated. We estimate the…

Double Machine Learning

The seventh notebook in the Causal ML stream. We combine everything: the partially linear model, the residual-on-residual idea from Notebook 4, and any learner we like for…

Heterogeneous Effects and Causal Forests

The eighth and final notebook in the Causal ML stream. We move from the average effect to who benefits. We meet the conditional average treatment effect, estimate it with…