A central question facing scientists in almost every field is: How do we know if one thing causes another, particularly if we cannot run a randomized experiment? My methodological work focuses on feasible causal inference: developing research methods that enable researchers across disciplines to more feasibly make credible causal inferences from observational data. This work falls into three major categories:
Developing new identification strategies that allow us to make causal claims based on assumptions that are more realistic--or at least more understandable and susceptible to debate--than those of existing approaches;
Sensitivity analyses that explore how violating important assumptions (such as unconfoundedness or the exclusion restriction) changes our inferences; and
Estimation and modeling: statistical estimation problems related to causal inference, such as weighting estimators, employing kernels and other tools from machine learning to relax functional form concerns, and time-series analysis.
Much of my substantive work has focused on civil war, indiscriminate violence, and mass atrocity. However In recent years I had also been increasingly involved in medical sciences and other fields where my methodological tools can add value. I also still work on the occasional neuroscience work, an earlier focus of mine.
Right now, my group is working feverishly to apply tools for feasible causal inference to improve the response to COVID-19. This includes using new identification strategies that allow us to learn the effects of unproven COVID-19 treatments like hydroxychloroquine and Remdesivir from those who are taking these treatments outside of randomized trials.
I teach courses on statistics, research design, causal inference, and machine learning. Occasionally I also teach about civil conflict and mass atrocity.