Marketing^AI

The Experimental Selection Correction Estimator: Using Experiments to Remove Biases in Observational Estimates∗


Listen Later

This academic paper proposes a novel method called the Experimental Selection Correction (ESC) estimator for combining observational and experimental data to estimate the causal effects of a treatment on a primary long-term outcome. Researchers often have access to large observational datasets with many outcomes but non-random treatment assignment, and smaller experimental datasets with random assignment but only secondary, intermediate outcomes. The ESC method utilizes the difference in the distribution of secondary outcomes between the experimental and observational data to adjust for selection bias in the observational data, allowing for a more reliable estimate of the treatment effect on the primary outcome under a new assumption called latent unconfoundedness. The authors apply this method to estimate the impact of third-grade class size on high school graduation rates, finding that their selection-corrected estimates align with experimental results and differ significantly from biased observational estimates.

...more
View all episodesView all episodes
Download on the App Store

Marketing^AIBy Enoch H. Kang