This repository provides R code for conducting Doubly Robust Estimation of the Average Treatment Effect (ATE) using modern machine learning techniques. Developed for a working paper, the package includes two core components:
DR_AIPW.R: Implements Augmented Inverse Probability Weighting (AIPW)DR_TMLE.R: Implements Targeted Maximum Likelihood Estimation (TMLE)
The code enables users to estimate causal effects with high-dimensional confounders by combining rigorous statistical methods with flexible machine learning models. This approach provides consistency and efficiency under less restrictive modeling assumptions, making it especially suitable for observational studies.
- 🧠 Implements two doubly robust estimators: AIPW and TMLE
- 🔍 Supports five machine learning models for nuisance estimation:
glmnet: Regularized regression (LASSO / Elastic Net)ranger: Random Forestxgboost: Gradient Boosted Treese1071: Support Vector Machines (SVM)nnet: Neural Networks
- 📊 Automatically compares ATE estimates across ML methods
- 📎 Based on theory from an accompanying working paper
- Causal inference in observational data
- Policy evaluation
- High-dimensional data analysis
DR_AIPW.R: AIPW estimation with multiple ML learnersDR_TMLE.R: TMLE estimation with multiple ML learners
To use this package, simply run either file after ensuring the necessary R packages are installed:
source("DR_AIPW.R") # for AIPW estimation
source("DR_TMLE.R") # for TMLE estimation