Directed Acyclic Graph (DAG)

A directed acyclic graph (DAG) represents each variable as a node and each assumed direct causal effect as an arrow from cause to effect. "Acyclic" means you can never follow arrows back to where you started — nothing causes itself. A DAG makes a researcher's causal assumptions explicit and visual, turning vague reasoning about "controlling for" variables into a checkable structure.

DAGs matter because they provide rules for which variables to adjust for. By tracing paths between exposure and outcome, you can identify confounding paths (which adjustment closes), mediators (which adjustment should usually leave open), and colliders (which adjustment wrongly opens). A valid adjustment set "blocks" all non-causal paths while leaving the causal path intact — the backdoor criterion.

The common pitfall is omitting variables or arrows you don't want to think about; a DAG is only as good as the assumptions drawn into it. It also encodes structure, not effect sizes, so it cannot tell you how strong a bias is, only whether one can exist.

For example, a simple cancer-prevention DAG might show Income → Screening Access → Stage at Diagnosis and Income → Stage at Diagnosis, signaling that income confounds the screening–stage relationship and should be adjusted for.