Abstract Directed acyclic graphs (DAGs) are now standard tools for selecting covariates and identifying estimands in causal inference. Yet in most applications, DAGs are treated as static and study-specific and then discarded rather than maintained as cumulative infrastructure. This Opinion piece argues that DAGs can serve a much broader role: as epistemic infrastructure that supports cumulative science. By treating DAGs as living, shared representations of causal systems—annotated with levels of evidence, revised over time, and tested empirically—we enable a mode of scientific practice that is transparent, collaborative, and intervention-oriented. Examples from spaceflight risk management and cerebral palsy research demonstrate how DAGs are already being used this way. I call on the field of epidemiology to adopt this approach more broadly: to share, refine, and re-use DAGs not just as tools of analysis, but as frameworks for designing better questions and building a more cumulative science.
Robert J. Reynolds (Fri,) studied this question.