A0804
Title: Time varying treatment effects of functional data with latent confounders
Authors: Jie Li - Renmin University of China (China) [presenting]
Shujie Ma - University of California-Riverside (United States)
Yehua Li - University of California at Riverside (United States)
Abstract: Understanding time-varying causal effects in functional data is critical in many scientific domains. Traditional causal inference methods focus on scalar or vector outcomes, but functional responses measured repeatedly over time require specialized approaches. In these settings, treatment assignment may change over time, and its causal effect is a function of time. Moreover, standard estimators rely on the ignorability assumption given observed covariates, which may be violated by unmeasured confounding. A novel joint modeling framework is proposed that incorporates latent covariates into both the functional outcome and treatment assignment models, estimated via a Monte Carlo EM algorithm. The method yields unbiased estimates of both heterogeneous and average time-varying treatment effects and accommodates irregular, sparse measurements. Applied to the study of women's health across the nation, it examined how depressive symptoms causally influence follicle-stimulating hormone levels during the menopause transition. The analysis uncovers the dynamic treatment effect and demonstrates the utility of the approach for complex functional data.
