Research

Statistical inference in nonparametric/semiparametric models

When analyzing data, it may be desirable to assume a nonparametric/semiparametric model rather than a parametric model to avoid relying on strong assumptions. In this setting, the target of inference is typically an overall summary of the underlying data-generating mechanism. Examples include average treatment effects under various identifying causal assumptions (e.g., no unmeasured confounding or access to an instrumental variable).

I am interested in developing valid procedures for statistical inference of such summaries in nonparametric/semiparametric models, typically utilizing asymptotic theory. In particular, I am interested in those procedures based on asymptotically normal plug-in estimators, which respect known bounds on the summary of interest. I am interested in applying the targeted minimum loss-based estimation (TMLE) framework to construction of such estimators and exploring novel approaches of using TMLE to solving outstanding methodological problems.

Numerical learning of statistical procedures

Traditionally, statistical procedures have been derived via analytic calculations whose validity often relies on asymptotic results. I am interested in using machine learning techniques to automatically learn statistical procedures that can perform well in small to moderate samples. In particular, I am interested in training strategies that can incorporate (possibly vague) prior knowledge into the learned statistical procedures in order to improve finite-sample performance in general models.

Causal inference

I develop methods aiming at statistical inference of a useful quantity that arise from causal inference. Such procedures may be useful in both observational studies and trials.