Scott L. Schwartz
Research Associate - Department of Integrative BiologyUniversity of Texas at Austin
Bayesian Mixture Modeling Approaches for Intermediate Variables and Causal Inference
This thesis examines causal inference related topics involving intermediate variables,\par and uses Bayesian methodologies to advance analysis capabilities in these areas. First,\par joint modeling of outcome variables with intermediate variables is considered in the\par context of birthweight and censored gestational age analyses. The proposed method-\par ology provides improved inference capabilities for birthweight and gestational age,\par avoids post-treatment selection bias problems associated with conditional on gesta-\par tional age analyses, and appropriately assesses the uncertainty associated with cen-\par sored gestational age. Second, principal stratification methodology for settings where\par causal inference analysis requires appropriate adjustment of intermediate variables is\par extended to observational settings with binary treatments and binary intermediate\par variables. This is done by uncovering the structural pathways of unmeasured con-\par founding a ecting principal stratification analysis and directly incorporating them\par into a model based sensitivity analysis methodology. Demonstration focuses on a\par study of the e_cacy of inuenza vaccination in elderly populations. Third, exi-\par bility, interpretability, and capability of principal stratification analyses for contin-\par uous intermediate variables are improved by replacing the current fully parametric\par methodologies with semi-parametric Bayesian alternatives. This presentation is one\par of the rst uses of nonparametric techniques in causal inference analysis, and opens\par a connection between these two elds. Demonstration focuses on two studies, one\par involving a cholesterol reduction drug, and one examine the eect of physical activity\par \pard on cardiovascular disease as it relates to body mass index.