Statistics Seminar

Abstract: Causal inference is a necessary goal in research for addressing many of the most pressing questions around policy and practice. However we know that it is difficult to create situations where we can be confident in our causal claims. In the past decade, causal methodologists have increasingly capitalized on and touted the benefits of more complicated machine learning algorithms to estimate causal effects. These methods can take some of the guesswork out of analyses, decrease the opportunity for “p-hacking,” and may be better suited for more fine-tuned tasks such as identifying varying treatment effects and generalizing results from one population to another. But they also fail to resolve some of the most pressing questions (should we ever believe an ignorability assumption) and raise additional questions. Should these more advanced methods change our fundamental views about how difficult it is to infer causality? Do sufficient guardrails exist to ensure appropriate use and interpretation? How can we provide tools that make it easy for non-technical researchers to use these new methods in a responsible way? I will discuss these issues, describe a new tool aimed at addressing some of these questions. I'll also present results from ongoing research that sheds light on ways we can support applied researchers in both doing better research and being transparent about their assumptions and methods.