A causal DAG puts variables at the nodes and direct causal effects on the arrows, with missing arrows asserting the absence of an effect, which makes it a statement of theory rather than a summary of data. Its payoff is mechanical: association flows along paths built from forks (confounders), chains (mediators), and colliders, and Pearl’s backdoor criterion turns “what should I control for?” into a graph-reading exercise.
The discipline it enforces is the point: a DAG forces the analyst to take a stand on how the world works, including unobserved variables, in a form colleagues can inspect and dispute.