Causal inference
Living DAGs as shared epistemic infrastructure in epidemiology.
Causal DAGs make assumptions explicit: which variables are confounders, mediators, colliders, or instruments, and which paths must be blocked to estimate a causal effect.
In practice, DAGs are often drawn once for a single paper and then discarded. Living DAGs treat the diagram itself as durable knowledge — versioned, annotated, and reusable across studies and teaching.
Living DAGs (Reynolds 2026)
Reynolds (2026, Am J Epidemiol) argues that epidemiology should move toward DAGs that are:
- Shared — visible to the community, not locked in supplementary PDFs
- Annotated — edges carry evidence levels and rationale, not just arrows
- Versioned — updates are tracked as understanding evolves
- Reusable — others can adapt a published structure to new settings
DAGpedia is an implementation of that direction: a catalog where each entry is both a machine-readable graph and a human-readable narrative.
Why this matters
| Problem | Living DAG approach |
|---|---|
| Hidden assumptions | Explicit graph + narrative edge rationale |
| Irreproducible diagrams | DAGitty source + validation in CI |
| One-off figures | Persistent URLs and git history per DAG |
You do not need to agree with every edge in the library — contributions must make assumptions explicit, not claim universal truth.