DoubleML

As explained in Causal Inference for Data Science

DoubleML is an open-source software library for estimating causal effects with double machine learning and related methods in R and Python.

DoubleML is an open-source R and Python library for estimating causal effects with double machine learning and related methods. It is useful when flexible machine-learning models are needed for nonlinear relationships and when estimate precision and reliability matter, such as in healthcare or finance. section 8.0 section 8.6

Rather than treating predictive accuracy as causal evidence, use the library after identifying confounders to adjust for. Its methods assume that relevant confounders have been measured or selected carefully, for example with the back-door criterion. section 8.6 section 8.0

Key interfaces and methods include:

• DoubleMLPLR, the partially linear regression estimator, takes ml_m to predict treatment from confounders and ml_l to predict outcome from confounders. Despite its name, PLR supports regression and classification problems. section 8.1.3
• Confidence intervals can be calculated analytically after fitting a DoubleML estimator and accessed through its summary output. section 8.2.2
• Interactive regression models (IRM) provide the library’s implementation of augmented inverse probability weighting (AIPW) for binary treatments. section 8.3.1
• DoubleMLDID applies difference-in-differences with covariates and machine-learning models, including nonlinear relationships. section 11.4.5

Estimating a 401(k) effect

A practical PLR analysis estimates the causal effect of enrolling in a 401(k) plan on net financial assets. Different machine-learning models can be paired with the estimator to model treatment and outcome relationships with confounders. section 8.1.3

Difference-in-differences with an age covariate

For an ATET question over time, create DoubleMLData with outcome, treatment, age covariate, and time columns; then fit DoubleMLDID using a regression outcome learner and logistic treatment learner. section 11.4.5

Two-fold cross-fitting applied to the DML algorithm. Once the process is finished, you must switch roles between datasets 1 and 2. This process can be generalized to k-fold cross-fitting by breaking the data into k datasets instead of two.

Use the diagram to see two-fold cross-fitting: train and estimate on alternating dataset roles, then generalize the procedure to k folds. section 8.1.3

How Causal Inference for Data Science covers it

How deep it goes

Introduces estimator classes and inputs but defers detailed library instruction to external tutorials and documentation.

Examples

Uses 401(k) enrollment and net financial assets, right-heart catheterization, and age-adjusted difference-in-differences datasets.

What it assumes

Assumes carefully measured or selected confounders and familiarity with the back-door criterion before estimation.

What you can do after

Apply it via DoubleMLPLR, IRM, or DoubleMLDID to estimate causal effects and obtain uncertainty intervals.

How it connects

Connects double machine learning with the adjustment formula, confidence intervals, AIPW, partially linear regression, and difference-in-differences.

liveBookmanning.com

Read Causal Inference for Data Science on liveBook Get the book at manning.com

Related concepts

• double machine learningimplements · Double machine learning is a causal-effect estimation method that uses machine-learning nuisance models with orthogonalized estimation to reduce bias from nuisance-function errors.
• instrumental variableuses · An instrumental variable is a variable that affects a treatment or exposure while influencing an outcome only through that treatment or exposure, under causal identifying assumptions.
• difference-in-differencesuses · Difference-in-differences is a causal inference technique that estimates an intervention effect by comparing outcome changes over time between treated and comparison groups.
• scikit-learnuses · scikit-learn is an open-source Python library providing tools for classical machine learning, including algorithms, preprocessing, model selection, and evaluation.
• augmented inverse probability weightingused by · Augmented inverse probability weighting is a doubly robust causal estimation method that combines an outcome model with inverse-probability treatment weighting.
• confidence intervaluses · A confidence interval is a range estimated from data by a statistical procedure designed to provide a specified long-run coverage probability for an unknown quantity.

This explainer was generated by Manning's knowledge graph from the text of Causal Inference for Data Science. The section links point to the parts of the book it draws on.