Michael KnausMar 13, 2023
Interested in learning #causalML for policy evaluation/learning?
I consolidated my teaching material from different courses for Master and PhD Economics students. Result:
- 10 slide decks
- 22 R notebooks
See https://github.com/MCKnaus/causalML-teaching and thread below
GitHub - MCKnaus/causalML-teaching: This repository consolidates my teaching material for "Causal Machine Learning".

This repository consolidates my teaching material for "Causal Machine Learning". - GitHub - MCKnaus/causalML-teaching: This repository consolidates my teaching material for "Causal M...

GitHub
Show threadMichael KnausMar 13, 2023
0. Welcome slides
- Pep talk
- Managing expectations
https://nbviewer.org/github/MCKnaus
Jupyter Notebook Viewer

Show threadMichael Knaus
1. Stats/’metrics recap:
- Conditional Expectation Functions and how to model them
- Convergence rates
Both play a crucial role for #DoubleML and students tend to have no good intuition what the latter means
https://nbviewer.org/github/MCKnaus
Jupyter Notebook Viewer

Mar 13, 2023 at 3:16pmWeb
Show threadMichael KnausMar 13, 2023
2. Supervised ML intro/recap through the causal ML lense:
- Lasso and especially Post-Lasso as basis of Double Selection later
- Regression trees and random forest as basis of Causal Tree and Forest later
https://nbviewer.org/github/MCKnaus
Jupyter Notebook Viewer

Show threadMichael KnausMar 13, 2023
3. Causal inference intro/recap:
- Potential outcomes
- Experiments
- Pitch of structural causal models and directed acyclic graphs
- Pitch of single-world intervention graphs
https://nbviewer.org/github/MCKnaus
Jupyter Notebook Viewer

Show threadMichael KnausMar 13, 2023
4. Constant effect estimation
- Double Selection
- Double ML for partially linear regression
- Introduction of Neyman orthogonality
https://nbviewer.org/github/MCKnaus/causalML-teaching/blob/main/Slides/CML4_DS_PLR.pdf
Show threadMichael KnausMar 13, 2023
5. Estimation of average PO and ATE allowing for heterogeneous effects
- Double ML for augmented inverse probability weighting (AIPW)
- Definition of AIPW ATE pseudo-outcome that has a prominent role in following lectures
https://nbviewer.org/github/MCKnaus
Jupyter Notebook Viewer

Show threadMichael KnausMar 13, 2023
6. The generic Double ML recipe
- Double ML with linear Neyman orthogonal scores
- Average treatment effect on the treated
- IV estimation with constant effects and for LATE
- Influence functions and their usage
https://nbviewer.org/github/MCKnaus
Jupyter Notebook Viewer

Show threadMichael KnausMar 13, 2023
7. Predicting treatment effects aka CATE estimation
- Why this is a nontrivial problem and T-learner not the end of the story
- Causal tree and forest of Imbens
& @susanathey
et al.
- R-learner (Nie & Wager)
- DR-learner of Kennedy

https://nbviewer.org/github/MCKnaus
Jupyter Notebook Viewer

Show threadMichael KnausMar 13, 2023
8. Heterogeneous effects with inference
- Generic ML of with BLP/GATES/CLAN (low-dimensional summary statistics of predicted CATEs)
- DR-learner with kernel or series regression

https://nbviewer.org/github/MCKnaus
Jupyter Notebook Viewer

Show threadMichael KnausMar 13, 2023
9. Offline/online policy learning
- Toy example to motivate that policy learning is different from CATE prediction
- Policy learning (binary and multiple treatments)
- Pitch of multi-armed bandits
https://nbviewer.org/github/MCKnaus
Jupyter Notebook Viewer