The goal of Tanya’s dissertation research is to examine and develop new methodology in the area of Causal Inference.

The first component of the research introduces a class of one-step estimators of causal parameters in generalized linear counterfactual models. By careful study of influence curves and the projection of the inverse of probability treatment weighted estimator (Robins) onto the tangent space of the propensity score, the one-step estimator makes gains in efficiency and is asymptotically linear and consistent. In addition to theoretical validation, the results are verified with simulations and data analysis.

Data arising from studies with unmeasured confounders present a type of confounding that standard regression methods cannot adequately account for. In the second component of the work, Tanya uses instrumental variables to construct a specific class of generalized estimating equations that is appropriate to these data. The use of instrumental variables will allow researchers to expand the use of already existing data. This approach will be applied to a multi-hospital study comparing two treatments for ruptured cerebral aneurysms, and to a study examining the effect of caffeine consumption on the risk of spontaneous abortions in a cohort of California women.