Debiased/ Double Machine Learning under Non-Standard Settings

• Applicant: Professor Dr. Martin Spindler
• Subject Area: Statistics and Econometrics
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 555312726

Complex, high-dimensional data sets have become increasingly common in research and industry with the rise of digitisation. Analyzing them requires machine learning (ML) methods such as regularization based methods (e.g., Lasso, Ridge, Elastic Net), random forest, boosting (e.g. boosted trees) and neural networks / deep learning, to name only a few. Importantly, these methods are tailored for prediction problems. However, in many applications, the interest goes beyond just predicting an outcome and extends to the estimation and inference of a target parameter in a high-dimensional settings. The so-called debiased / double machine learning approach allows for valid inference in such high-dimensional settings. In the project the double machine learning approach will be extended to non standard settings, including spatial, temporal and spatial-temporal data. Most real world applications fall into these data types and therefore the results should help to perform valid inference in such complex settings.

• DFG Programme: Research Grants
• International Connection: USA
• Cooperation Partner: Professor Victor Chernozhukov, Ph.D.