[Paper Review] DoubleML -- An Object-Oriented Implementation of Double Machine Learning in R

Philipp Bach, Victor Chernozhukov+3 more|arXiv (Cornell University)|Mar 17, 2021

Mental Health Research Topics67 references21 citations

TL;DR

This paper introduces DoubleML, an R package that implements the double/debiased machine learning framework using an object-oriented design. It enables valid causal inference in high-dimensional models—such as partially linear and interactive regression models—by leveraging Neyman orthogonality, sample splitting, and machine learning estimators from the mlr3 ecosystem, ensuring robust and efficient estimation of causal parameters even with complex nuisance functions.

ABSTRACT

The R package DoubleML implements the double/debiased machine learning framework of Chernozhukov et al. (2018). It provides functionalities to estimate parameters in causal models based on machine learning methods. The double machine learning framework consist of three key ingredients: Neyman orthogonality, high-quality machine learning estimation and sample splitting. Estimation of nuisance components can be performed by various state-of-the-art machine learning methods that are available in the mlr3 ecosystem. DoubleML makes it possible to perform inference in a variety of causal models, including partially linear and interactive regression models and their extensions to instrumental variable estimation. The object-oriented implementation of DoubleML enables a high flexibility for the model specification and makes it easily extendable. This paper serves as an introduction to the double machine learning framework and the R package DoubleML. In reproducible code examples with simulated and real data sets, we demonstrate how DoubleML users can perform valid inference based on machine learning methods.

Motivation & Objective

• To provide a flexible, extensible, and reproducible implementation of the double machine learning framework in R for causal inference.
• To enable valid inference in high-dimensional causal models by combining Neyman orthogonality, sample splitting, and machine learning for nuisance parameter estimation.
• To support a wide range of causal models, including partially linear, interactive, and instrumental variable models, via an extensible object-oriented architecture.
• To integrate seamlessly with the mlr3 ecosystem for flexible model specification, tuning, and resampling.
• To facilitate reproducible research through integrated replication code and support for both simulated and real-world data examples.

Proposed method

• The package uses an object-oriented design based on R6 classes to encapsulate model-specific components, including score functions and estimation logic.
• It implements Neyman-orthogonal estimating equations to ensure robustness to estimation errors in nuisance functions.
• Sample splitting is employed to decouple estimation of the target parameter from nuisance function estimation, improving inference validity.
• Nuisance functions (e.g., conditional mean or propensity scores) are estimated using state-of-the-art machine learning methods from the mlr3 ecosystem.
• The framework supports various resampling schemes and includes built-in procedures for standard error estimation, confidence intervals, and simultaneous inference via multiplier bootstrap.
• New model types can be added by defining new score functions that are linear in the target parameter and satisfy Neyman orthogonality.

Experimental results

Research questions

• RQ1 How can double machine learning be effectively implemented in R to enable valid inference in high-dimensional causal models?
• RQ2 To what extent does the object-oriented design of DoubleML enhance flexibility and extensibility for diverse causal models?
• RQ3 Can the integration with the mlr3 ecosystem support robust and scalable estimation across a wide range of machine learning methods?
• RQ4 How does DoubleML perform in finite samples across different data-generating processes, including PLR, PLIV, IRM, and IIVM models?
• RQ5 What is the empirical performance of DoubleML in terms of coverage probability, size, and power in simulation studies with high-dimensional confounders?

Key findings

• DoubleML successfully implements the double machine learning framework in R with full support for partially linear and interactive regression models, including instrumental variable extensions.
• The package achieves valid inference in high-dimensional settings by ensuring Neyman orthogonality and using sample splitting to reduce bias in nuisance estimation.
• In simulation studies with n=500 and n=1000 observations, DoubleML maintains appropriate coverage rates for confidence intervals across PLIV, IRM, and IIVM models.
• The method demonstrates robustness to model misspecification and estimation errors in high-dimensional nuisance functions, as evidenced by stable size and power in repeated simulations.
• The integration with mlr3 enables flexible and automated tuning of machine learning models for nuisance functions, improving estimation efficiency.
• The package supports advanced inference, including simultaneous confidence regions and adjusted p-values, via a multiplier bootstrap procedure for joint inference.