HS 306A — Causal Inference and Machine Learning in Studies Using Observational Data

Ideally, causal evidence would be used to support any important policy decision. The strongest source of causal evidence is from well-designed randomized trials but such data are rarely available for most policy questions. Instead, causal inference must be drawn from observational studies. This course begins by examining the concept of the target trial, along with two major causal inference frameworks—Directed Acyclic Graphs and Potential Outcomes. We then briefly review several causal inference methods from the econometrics literature. Next, we will examine a series of methods from epidemiology that are commonly used to draw causal inference from observational health care data. Finally, we consider relatively new methods using machine learning techniques that, when used within a causal inference framework, may substantially improve our ability to draw causal inferences relative to traditional methods. For each method, we will consider the theory and methods, as well as practical applications from the literature. Usually offered every year.

William Crown