A0851
Title: Multivariate treatment effect estimation through Bayesian factor regression model
Authors: Dafne Zorzetto - Brown University (United States) [presenting]
Abstract: In the context of causal inference, investigating causal effects for multivariate potential outcomes has not yet been extensively explored. This is due to the missing data problem inherent in the potential outcomes framework, which leads to the challenges of capturing the causal effect of a treatment on correlated outcomes and understanding the heterogeneity of the causal effect between outcomes. The ability of Bayesian factor analysis is exploited to identify the latent treatment-specific factors that capture and characterize the causal effects within correlated multivariate outcomes. The innovative use of the dependent Dirichlet process as the distribution for the factor scores allows the overcoming of the problem of missing data through an aware and fair imputation. Aligned to real data questions in environmental epidemiology, the causal link between air pollution regulation and the concentration of various pollutants in the United States is investigated.
