Using DAGs to Guide Analysis
What you'll learn: How to use DAGs to determine which variables to include (or exclude) in your regression models to estimate causal effects correctly.
From Theory to Practice
You've learned how to draw DAGs, identify confounders and colliders, and understand d-separation. Now comes the payoff: using your DAG to decide exactly which variables to control for in your statistical analysis.
The Adjustment Set Problem
When estimating a causal effect (say, treatment → outcome), you need to block all backdoor paths while keeping the frontdoor path open. Your DAG shows you how:
- Identify backdoor paths between treatment and outcome
- Find minimal adjustment sets — the smallest set of variables that, when conditioned on, blocks all backdoor paths
- Avoid conditioning on colliders — which would create spurious associations
- Avoid conditioning on mediators on the causal path — which would block part of the effect you want to measure
Practical Workflow
Step 1: Draw your DAG based on domain knowledge
Step 2: Apply the backdoor criterion to find valid adjustment sets
Step 3: Choose variables you actually have data for
Step 4: Run regression controlling for those variables — and only those variables
Think of the DAG as your analysis blueprint. Just as you wouldn't build a house without consulting the architectural plans, you shouldn't run causal regressions without consulting your DAG. It tells you precisely which confounders to adjust for and which colliders to avoid.
Common Mistakes DAGs Help You Avoid
- Including "everything available" as controls (might condition on colliders or mediators)
- Omitting necessary confounders (leaves backdoor paths open)
- Conditioning on post-treatment variables that create selection bias
Key Takeaway: A properly constructed DAG translates causal assumptions into concrete statistical decisions, telling you exactly which variables belong in your regression model and which must be excluded.