Lesson 2 of 6 · 12 min
DAGs and confounding
Use a directed acyclic graph to find back-door paths and a valid adjustment set.
A DAG encodes your assumptions about which variables cause which. Every arrow is a claim. Paths that enter the treatment through its causes, called back-door paths, create bias in naive comparisons.
To block a back-door path, condition on a pre-treatment variable along it. Never condition on mediators, which removes part of the effect, or on colliders, which opens new spurious paths.
Key points
- Back-door paths bias naive comparisons
- Adjust for pre-treatment confounders
- Do not adjust for mediators or colliders
Knowledge check
Sign in to mark this lesson complete and resume where you left off.