DataTerminology.comNEW
A graphical rule identifying covariate sets that block backdoor paths, enabling unbiased causal effect estimation from observational data.
DESCRIPTION
The backdoor criterion, introduced by Judea Pearl, provides conditions for selecting adjustment sets that block non-causal paths between treatment and outcome in a causal DAG. By controlling for appropriate covariates, analysts remove confounding and estimate causal effects from observational data. It relies on d-separation principles and avoids conditioning on colliders. The criterion underpins methods like propensity score adjustment and regression with selected controls. Complementary rules include the front-door criterion and instrumental variables for alternative identification strategies.
ALTERNATIVE DEFINITIONS
Wikipedia
In causal inference, the back-door criterion is a special case of d-separation used to identify a set of variables which blocks all back-door paths between a treatment and an outcome.
Emphasizes its relationship to d-separation and blocking back-door paths.
Hernán & Robins (Causal Inference: What If)
The backdoor criterion is a graphical condition that identifies sets of variables that, when conditioned upon, block all noncausal (backdoor) paths between treatment and outcome.
Frames the criterion as a graphical condition for valid adjustment.
RELATED TERMS
Directed Acyclic Graph (Causal) Do-Calculus Confounding d-separation Instrumental Variable Adjustment Set Causal Inference Propensity Score Directed Acyclic Graph
COMPETING TERMS
RELATIONS
- Broader: Causal Inference
KNOWLEDGE AREAS
TARGET AUDIENCE
Data Scientist AI Engineer Data Strategy Consultant Information Analyst
EXTERNAL SOURCES
- Wikipedia - Back-door criterion
- Judea Pearl (UCLA) - Pearl’s technical notes on backdoor
- Harvard T.H. Chan School - Open causal inference textbook
- BMJ - Causal inference overview article
TOP QUESTIONS
Why is the Backdoor Criterion important in causal inference?
It guides valid adjustment to remove confounding in observational studies. Without it, estimates may be biased due to uncontrolled backdoor paths between treatment and outcome.
Category: Relevancy
How do you apply the Backdoor Criterion to a DAG?
Identify backdoor paths from treatment to outcome, then find sets of covariates that block these paths without conditioning on colliders or their descendants.
Category: Requirements
What are common mistakes when using the Backdoor Criterion?
Conditioning on colliders introduces bias; over-adjusting for mediators distorts causal effects. Misreading the DAG structure is a frequent cause of errors.
Category: Mistakes
How does the Backdoor Criterion relate to propensity scores?
Propensity scores summarize the probability of treatment given covariates; selecting covariates using the backdoor criterion supports valid propensity score adjustment.
Category: Example
When should the Front-door Criterion be preferred over the Backdoor Criterion?
Use front-door when no valid backdoor adjustment set exists but a measurable mediator satisfies the front-door conditions for identification.
Category: Alternatives
What is a misunderstanding about blocking backdoor paths?
Blocking all paths isn’t always correct; you must avoid conditioning on colliders and not adjust away part of the causal effect through mediators.
Category: Misunderstanding
How do consistency and positivity assumptions interact with the Backdoor Criterion?
Even with a valid adjustment set, identification requires that treatments are well-defined (consistency) and each covariate pattern has treatment variation (positivity).
Category: Requirements
Is the Backdoor Criterion sufficient for unbiased estimation?
It is sufficient if assumptions hold and measurement is accurate; violations (measurement error, model mis-specification) can still bias results.
Category: Risks
How to learn more about the Backdoor Criterion?
Study Pearl’s do-calculus, causal DAGs, and practical guides like Hernán & Robins. Work through examples and counterexamples to internalize collider and mediator patterns.
Category: Overview