Reading a Dag

The one idea: A DAG is a one-page map of what you think causes what in your business. Its real power is that it tells you which things to hold steady when you compare groups — and holding the wrong thing steady doesn't just waste time, it can hand you a confident, wrong answer.

⚠️ The trap: "The more variables I control for, the more careful and rigorous my analysis is." That feels right and it's wrong. The diagram — not the data — decides which handful of variables to hold steady, and holding the wrong one (a collider: a variable your treatment and outcome both cause) manufactures a confident, wrong answer instead of removing bias.

A DAG — directed acyclic graph — is the one-page version of your theory about how the business works. Variables are dots, arrows are causal claims, and the whole thing fits on a whiteboard. It looks like a doodle. It is actually the most consequential artifact in a causal analysis, because the DAG decides which variables you control for. As the next two lessons show, controlling for the wrong variable doesn't just waste effort — it manufactures wrong answers wrapped in confident confidence intervals.

Here's the part almost everyone misses on first contact. The arrows you draw are the cheap claims. The arrows you leave out are the expensive ones. An arrow says "X might affect Y" — a soft, defensible statement. A missing arrow says "X has zero direct effect on Y, for everyone, always." Scott Cunningham makes the point bluntly in Causal Inference: The Mixtape: a missing arrow is one of the strongest beliefs you can…