Removing Hidden Confounding by Experimental Grounding
This addresses the challenge of making reliable causal predictions from observational data, which is crucial for fields like education and healthcare, though it is incremental in improving existing correction methods.
The paper tackles the problem of hidden confounding in causal inference from observational data by introducing a method that uses limited experimental data to correct for confounding, even when the observational and experimental data do not fully overlap. The method is demonstrated on real-world educational data, showing efficacy under weaker assumptions than existing approaches.
Observational data is increasingly used as a means for making individual-level causal predictions and intervention recommendations. The foremost challenge of causal inference from observational data is hidden confounding, whose presence cannot be tested in data and can invalidate any causal conclusion. Experimental data does not suffer from confounding but is usually limited in both scope and scale. We introduce a novel method of using limited experimental data to correct the hidden confounding in causal effect models trained on larger observational data, even if the observational data does not fully overlap with the experimental data. Our method makes strictly weaker assumptions than existing approaches, and we prove conditions under which it yields a consistent estimator. We demonstrate our method's efficacy using real-world data from a large educational experiment.