Causal Inference
The problem of interpretational errors in epidemiology Aaron* Aaron Sarvet
The “causal revolution” promises to aid in the translation of data into public good. In this talk I illustrate that merely doing a “causal” analysis is not enough if it provides “the right answer to the wrong question”. I present practical case studies in organ transplantation, and intensive care triage. These examples illustrate interpretational errors commonly arising in settings with scarce resources. As a constructive solution, I present work designed to prevent interpretational errors by allowing an investigator to explicitly target policy-relevant questions in these settings. More broadly, I contrast the principles of this work with contemporary trends in statistical methods development. I argue that methods can undermine public health when they prioritize statistical properties over faithfulness to real-world public health problems.