Assessing Treatment Effects in Observational Data with Missing Confounders: A Comparative Study of Practical Doubly-robust and Traditional Missing-data Methods

This presentation examined methods for estimating exposure effects in observational studies with missing confounding variables. Traditional analytic methods for addressing missing data like multiple imputation (MI) and inverse probability weighting (IPW) rely on correct model specification. More robust methods, such as generalized raking and targeted maximum likelihood estimation, are discussed and compared to MI and IPW through numerical experiments. 

It was presented at the Eastern North American Region (ENAR) International Biometric Society Spring 2025 Meeting.