A Double Machine Learning Approach for Combining Experimental and Observational Studies.

Area of Science:

• Econometrics
• Machine Learning
• Causal Inference

Background:

• Experimental and observational studies are crucial for empirical research but often suffer from untestable assumptions, limiting their validity.
• Combining data from different study types (data fusion) is a promising approach, but existing methods struggle with assumption violations.

Purpose of the Study:

• To develop a robust framework for combining experimental and observational studies using double machine learning.
• To enable practitioners to rigorously test for assumption violations and consistently estimate treatment effects.
• To propose a falsification test for external validity and ignorability under relaxed assumptions.

Main Methods:

• A double machine learning approach is proposed to integrate experimental and observational data.
• The framework incorporates a falsification test to assess external validity and ignorability.
• Consistent treatment effect estimators are derived, even when key assumptions are violated.

Main Results:

• The proposed method allows for testing assumption violations, improving the reliability of causal inference.
• Consistent treatment effect estimation is achievable even if one assumption is violated.
• Comparative analyses demonstrate the superiority of this framework over existing data fusion techniques.

Conclusions:

• The developed double machine learning framework offers a powerful tool for combining diverse study designs.
• It enhances the validity and reliability of treatment effect estimation in empirical research.
• Real-world case studies confirm the practical utility and broad applicability of the approach.