Causal DAG

A causal DAG (directed acyclic graph) is a diagram in which each variable is a node and each arrow points from a cause to its direct effect, with no loops. It is a picture of your assumptions about how the world generates the data.

DAGs turn fuzzy worries about confounding into precise, checkable rules: you can read off which variables are confounders, colliders, or mediators, and which you must or must not adjust for. They make assumptions transparent and arguable, but a DAG is only as good as the causal knowledge you put into it — the data cannot draw it for you.