The package provides R and Python implementations of Isotonic Calibrated Debiased Machine learning (IC-DML) for average treatment effects from our paper on Automatic Doubly Robust Inference for Linear Functionals via Calibrated Debiased Machine Learning.
The CDML (Calibrated Debiased Machine Learning) package implements tools for automatic doubly robust inference for linear functionals in causal inference problems. Many causal estimands of interest, such as average causal effects of static, dynamic, and stochastic interventions, can be expressed as linear functionals of the outcome regression function. CDML leverages modern machine learning to provide debiased estimators that are doubly robust asymptotically linear, thus providing not only doubly robust consistency but also facilitating doubly robust inference (e.g., confidence intervals and hypothesis tests). The C-DML estimator integrates cross-fitting, isotonic calibration, and debiased machine learning estimation (AIPW). It maintains asymptotic linearity when either the outcome regression or the propensity score is estimated sufficiently well, allowing the other to be estimated at arbitrarily slow rates or even inconsistently. For inference, we use a bootstrap-assisted approach to construct doubly robust confidence intervals.
Currently, the supported estimands are the Average Treatment Effect (ATE), treatment-specific means, and the Average Treatment effect on the Treated (ATT). Internal functions support more general functionals of the outcome regression, given that the functional and Riesz representer are specified. The code can be modified to accommodate more general parameters and please reach out to me if you would like me to implement other estimands.
To install the CDML package in R, run the following command. For an example of using this package, see the Vignette.rmd file.
devtools::install_github("Larsvanderlaan/cdml")To run the Python code, download the code files in the Python folder and refer to the vignette.ipynb file for usage instructions.