University of Wisconsin Professor of Sociology, Biostatistics, and Medical Informatics
Causal effects measure how actions change the world. Does the drug improve survival? Does portfolio diversification boost profit? Do social influencers actually influence anybody? These are causal questions, because they ask about the relative merit of one choice over another. Mere prediction cannot answer causal questions because it just extrapolates from the current state of affairs. By contrast, causality asks what would happen if business as usual was disrupted by intervention. Moving from prediction to causation requires a new way of thinking. Data alone provably don’t cut it—they must be combined with expert knowledge (“assumptions”) in a principled manner. This course introduces a powerful graphical approach for understanding what it takes to answer causal questions, and how to go about it in practice. We use directed acyclic graphs (DAGs) to bridge the gap between expert knowledge and raw data to estimate causal effects. This method was first developed in artificial intelligence by Turing-Award winner Judea Pearl and has since been adopted for use in health, business, and statistics. DAGs support rigorous reasoning, and yet they are intuitively accessible—ideal for practitioners in the trenches who have to communicate between experts and clients. This course empowers participants • to distinguish causal from non-causal questions; • to determine what data are needed to answer causal questions; • to link expert knowledge with data using DAGs, • and to understand promises and pitfalls of causal inference in practice. We will emphasize a concrete conceptual understanding of formal principles governing all causal inquries, and drill practical skills with many exercises and real examples. This course is platform-independent and will not deal with programming specifics. We think before we do.