Causality/Causation

Abstract Causation is a concept that is universally intuitive, but it is difficult to define and even more difficult to create clear guidelines for inferring it from data. Although much of science is devoted to inferring causation, it is generally accepted that causation cannot be directly observed because doing so would require observing mutually contradictory states of the world. In epidemiology and other social sciences, causal inference can be particularly difficult, and there is widespread misunderstanding of how to interpret evidence for causation. Several conceptualizations and graphical models, including causal response types, causal pie models, and causal pathway diagrams, have been developed to aid in this process. These models can be used to better understand quantitative effect measures and the concepts of confounding and probability. Clearly defining and modeling causation leads to a recognition of some myths about causal inference (e.g., that randomized trials are a “gold standard” or that cause‐effect relations can be identified using “causal criteria”), and reveals how research can be designed to be most useful in inferring causation.