Double Machine Learning (DML)
The short answer · 17 words
In short
DML combines machine learning with statistical estimation techniques to reduce distortions in causal parameters under appropriate assumptions.
Flow-alerta.cab (2026): Double Machine Learning (DML). https://flow-alerta.cab/en/glossar/double-machine-learning-dml. Editorially reviewed on: 09 October 2026.
How much detail would you like?
Practical
Relevant for demanding impact analyses; professional methodology and suitable data are prerequisites.
Possible example: An entity examines the influence of a price change taking into account observed customer characteristics and documents the remaining assumptions.
Price elasticity analysis in retail, evaluation of marketing campaigns and data-driven impact analyses in the B2B sector.
At a glance
- Core
- ML methods estimate auxiliary variables, such as outcome and treatment models. Orthogonal estimation equations and cross-fitting can reduce sensitivity to estimation errors and overfitting. Causal identification and further model assumptions remain necessary.
- Use
- Relevant for demanding impact analyses; professional methodology and suitable data are prerequisites.
- Example
- Possible example: An entity examines the influence of a price change taking into account observed customer characteristics and documents the remaining assumptions.
- Limits
- False assumption about unmeasured disturbances (unobserved confounders); high demands on data volume and model architecture.
Questions about this term
Does a good prediction model automatically provide a causal effect?
No. Prediction and causal estimation have different requirements.
Does DML remove unmeasured disturbances?
No. The method does not replace an appropriate identification strategy.
All details and background
ML methods estimate auxiliary variables, such as outcome and treatment models. Orthogonal estimation equations and cross-fitting can reduce sensitivity to estimation errors and overfitting. Causal identification and further model assumptions remain necessary.
ELI5 – Explained simply
Consider measurable influencing factors with auxiliary models before estimating an effect. Unknown disturbing factors do not disappear as a result.
Why it matters in practice
Relevant for demanding impact analyses; professional methodology and suitable data are prerequisites.
Example
Possible example: An entity examines the influence of a price change taking into account observed customer characteristics and documents the remaining assumptions.
Pitfalls and limits
- False assumption about unmeasured disturbances (unobserved confounders); high demands on data volume and model architecture.
Origin & history
Chernozhukov et al. published the influential DML approach as a working paper in 2016 and a journal contribution in 2018.
Use in mid-sized companies
- Price elasticity analysis in retail, evaluation of marketing campaigns and data-driven impact analyses in the B2B sector.
Related terms
- Causal inference
- EconML
- Confounder
- Cross-fitting
- Neyman Orthogonality
Matching Flow-alerta service
AI opportunity analysis & roadmapAI Potential Analysis & Roadmap – Classifying the suitability of sophisticated data analysesMatching tool / use case
View use casesUse Case: Estimate a causal parameter with suitable auxiliary models and documented assumptions.Sources
- arxiv.orgChernozhukov et al. (Arbeitspapier 2016; Zeitschriftenbeitrag 2018), Double/Debiased Machine Learning for Treatment and Causal Parameters
Quellen geprüft am 09.10.2026.
Editorially reviewed on: