Semiparametric Inference for Missing-Not-at-Random Data

This lecture by the instructor covers the challenges of systematically missing data in statistical analysis, the failure of standard approaches like complete-case analysis, and the need for semiparametric inference for non-monotone missing-not-at-random data. The lecture delves into the identification of missingness mechanisms, assumptions, and the No Self-Censoring Model. It explores odds ratio parameterization, components of OR factorization, and the graphical perspective of the model. The lecture also discusses the efficient influence function, a proposed doubly-robust estimator, and its application to HIV data. The conclusion highlights the general identified MNAR model, the semiparametric efficiency bound, and the performance of the proposed estimator through simulations and real-world applications.