Double Machine Learning (DML)
Causal inference method that uses ML models to flexibly control for confounding while enabling valid statistical inference.
Double ML combines ML flexibility with causal validity: Two ML models control for confounding, residuals deliver the causal effect.
Explanation
DML uses two ML models: (1) Predict treatment from confounders, (2) Predict outcome from confounders. The residuals of both are then used for causal estimation. Cross-fitting prevents overfitting bias.
Marketing Relevance
Combines ML flexibility (non-linear confounder control) with classical econometrics validity – ideal for data-rich marketing settings.
Common Pitfalls
Needs good ML models for treatment and outcome. Overlap assumption must hold. Unobserved confounders remain a problem.
Origin & History
Chernozhukov et al. published DML in 2018. EconML (Microsoft) and DoubleML (Python) made it practical. Considered the bridge between ML and econometrics.
Comparisons & Differences
Double Machine Learning (DML) vs. Instrumental Variable
IV needs an exogenous instrument; DML controls confounding directly with ML models (more flexible, but needs observability).
Double Machine Learning (DML) vs. Propensity Score Matching
Propensity Score Matching uses one model; DML uses two (treatment and outcome) with cross-fitting for less bias.