Student seminar series
Sijie Chen
PhD Candidate
Room: M3 3127
Methods for Handling Unmeasured Confounding in Repeated Outcomes Data
Unmeasured confounding remains one of the most persistent challenges in causal inference. Although the assumption of no unmeasured confounding is fundamental to all causal models, it is inherently untestable in observational studies. Various methods have been developed to mitigate this issue, including the use of instrumental variables, proxy variables, negative control strategies, and the front-door criterion. While these approaches are well-studied in point-treatment settings, their extensions to time-varying treatments and outcomes have received comparatively little attention. Standard causal inference methods often yield biased estimates when covariates vary over time. To address this limitation, structural nested models (SNMs) provide a framework for unbiased estimation in longitudinal settings. In this proposal, we develop a structural nested mean model (SNMM) with instrumental variables (IV) to handle both time-varying treatment and outcome processes. For survival outcomes, we revisit the structural nested cumulative failure time model (SNCFTM) with IV proposed by Shi et al. (2021) and examine its performance under more complex causal structures.