Teaching Causal Inference with Observational Data

Most students are more interested in causal inference than in population inference. Most students are more interested in observational causation than in randomized-experimental causation. Observational causation has been modeled from four different perspectives: Galton's correlation-regression method, Rubin's propensity-imputation method, Pearl's DAG-SEM method, and a Statistical Literacy combination of Wainer's graphical method of controlling for confounding and Cornfield's necessary conditions for nullification. Since Galton's regression method is well known and taught in the multivariate course, this session focuses on teaching observational causation using the last three methods. If we want our students to have a lasting appreciation of statistic, we need to provide a separate introductory course on causal inference that stands along side the traditional course on population inference where together they form a two-semester sequence.