B0864
Title: Bayesian nonparametrics for principal stratification
Authors: Antonio Canale - University of Padua (Italy) [presenting]
Dafne Zorzetto - University of Padova (Italy)
Fabrizia Mealli - University of Florence (Italy)
Falco Joannes Bargagli Stoffi - Harvard University (United States)
Francesca Dominici - Harvard University (United States)
Abstract: Principal stratification is a widely employed causal inference framework utilized in health and environmental sciences to address confounding factors after treatment. However, the application of principal stratification with continuous post-treatment variables presents several inferential challenges. Notably, the definition of latent principal strata becomes intricate due to the diverse responses of the intermediate variable to the treatment. To tackle this issue, leveraging dependent nonparametric mixture models is proposed to characterize the distribution of the post-treatment variable, enabling a model-based approach for defining principal strata. This novel method is demonstrated through simulations and an application focused on estimating the impact of air quality regulations on pollution levels and health outcomes.
