Course contentsShow
← Data Science
Lesson 1472 of 2,14536. Causal Inference BasicsFree lesson

The Backdoor Criterion

Master the backdoor criterion for identifying which variables to control for to block non-causal paths and isolate causal effects.

The Backdoor Criterion

What you'll learn: How to systematically identify which variables to control for to isolate a true causal effect and block spurious associations.

Why We Need the Backdoor Criterion

You've built your DAG and identified the causal path from treatment to outcome you care about. But your DAG also shows confounders, mediators, and colliders—some variables help, others hurt. The backdoor criterion gives you a formal rule for choosing exactly which variables to condition on to block all non-causal "backdoor" paths while keeping the causal path open.

The Rule Explained

A set of variables Z satisfies the backdoor criterion relative to treatment X and outcome Y if:

  1. No variable in Z is a descendant of X (don't control for mediators or anything caused by treatment)
  2. Z blocks every path between X and Y that contains an arrow pointing INTO X (these are "backdoor paths"—non-causal routes through confounders)

Think of it like this: backdoor paths are side entrances that let spurious associations sneak in. The backdoor criterion tells you which guards (control variables) to post to lock those doors, without blocking the front entrance (your causal path).

Practical Application

If you have X → Y (causal path) and Z → X → Y plus Z → Y (a confounder creating a backdoor path X ← Z → Y), controlling for Z blocks the backdoor while preserving the causal arrow. But if W is a collider on a non-open path, don't control for it—you'd create spurious association.

Key Takeaway: The backdoor criterion is your decision rule: control for variables that block all backdoor paths into your treatment without being affected by the treatment itself, ensuring you measure only the causal effect.