In the context of causal inference, the study of causal effects for multivariate potential outcomes remains relatively underexplored. This gap largely stems from the inherent difficulty of quantifying the overall causal impact of a treatment on correlated outcomes and understanding how these effects vary across different outcomes. Nevertheless, this research question is critically important in fields such as environmental epidemiology, where one may seek to assess the causal relationship between air pollution regulations and the concentrations of multiple pollutants, or between air pollution exposure and hospitalizations for various diseases.
To address this challenge, we leverage a Bayesian factor regression model to identify latent, treatment-specific factors that capture causal effects among correlated multivariate outcomes. We propose a methodology that (i) introduces novel causal estimands within a general framework for multivariate outcomes, and (ii) develops a multi-treatment Bayesian factor regression model that enables the identification and characterization of causal latent effects. The innovative use of the dependent Dirichlet process as the distribution for the factor scores further allows us to handle missing data through a principled, awareness-driven, and fair imputation mechanism.
The performance of the proposed method is demonstrated through both simulation studies and real-world applications in environmental epidemiology.
31 Ottobre 2025, ore 12
Roberta De Vito
Sapienza University of Rome
In person: Room 34 (4th floor) building CU002 Scienze Statistiche
Webinar: https://uniroma1.zoom.us/j/83625004899?pwd=bXCtz0mp759PUh2lkqT0BUoVa0Uegg.1
ID riunione: 836 2500 4899
Passcode: 123456