Research: Evaluating Doubly-Robust and Double Machine Learning Estimators for Causal In...

Research question: How do doubly-robust and double-ML estimators perform in estimating causal effects from observational datasets compared to traditional methods?

Causal inference is central to understanding the impact of interventions and policies from observational data, especially when randomized controlled trials are infeasible. Doubly-robust and double machine learning (DML) estimators have emerged as promising tools, combining robustness to model misspecification with the flexibility of modern machine learning.

Despite their theoretical appeal, there is limited empirical evidence comparing the performance of doubly-robust and double-ML estimators against classical approaches across varied real-world datasets. Many studies focus on synthetic benchmarks or narrow scenarios, leaving open questions about their practical advantages and limitations.

This project will systematically review literature, formulate hypotheses, and empirically evaluate these methods using large public observational datasets. The student will implement and compare the estimators, assess their robustness, and analyze conditions under which each method excels or fails. The expected contribution is a reproducible benchmarking study and practical recommendations for applied researchers.

Such rigorous evaluation matters for advancing reproducible analytics and guiding practitioners in selecting methodological tools for causal inference, impacting fields from epidemiology to economics.