Robust Causal Inference on Public Health Interventions Using Epidemic Models

Estimating the causal effects of public health interventions in epidemics is challenging due to time-varying confounding and dynamic feedback between interventions and outcomes. Parametric epidemic models are widely used but can suffer from bias and small effective sample size. In this talk, I will present two key contributions to address these challenges. First, I propose an estimating equation to correct collider bias in epidemic models, preventing spurious causal inferences. Second, I introduce a shrinkage method for estimating causal effects across regions without relying on hierarchical model assumptions, providing robust condence intervals. Using COVID-19 data, I demonstrate how these methods evaluate the effectiveness of public health interventions and offer practical tools for decision-making in epidemic management.