Course contentsShow
← Data Science
Lesson 1470 of 2,14536. Causal Inference BasicsFree lesson

Confounders in DAGs

Understand how confounders appear in DAGs as common causes of treatment and outcome, and why they create spurious associations.

Confounders in DAGs

What you'll learn: How confounders show up in causal diagrams and create misleading associations between treatment and outcome.

What Is a Confounder in a DAG?

In a Directed Acyclic Graph (DAG), a confounder is a variable that points arrows to both your treatment and your outcome. It's a common cause that creates a "backdoor path" — an association between treatment and outcome that exists even without any causal effect.

Picture this: You're studying whether coffee consumption (treatment) affects productivity (outcome). But consider age — older workers might drink more coffee AND naturally have different productivity patterns due to experience. Age is a confounder because it influences both variables simultaneously.

In DAG notation:

Age → Coffee → Productivity
 ↓_____________↑

The arrow from Age to Coffee and Age to Productivity creates a spurious (false) correlation. Even if coffee has zero causal effect on productivity, you'd still see them correlated because of their shared cause: age.

Why Confounders Create Spurious Associations

When a confounder exists, the correlation you observe between treatment and outcome contains two components:

  1. The true causal effect (if any)
  2. The association created by the common cause

Without controlling for the confounder, you can't separate these. You might think coffee boosts productivity when you're really just observing that older workers drink more coffee and happen to be more productive for unrelated reasons.

The confounder "opens" a non-causal pathway between your variables, contaminating your estimate of the true causal effect.

Key Takeaway: Confounders appear in DAGs as common causes with arrows pointing to both treatment and outcome, creating backdoor paths that produce spurious associations you must block to identify true causal effects.