Structural Causal Model (DAG)

A structural causal model represented as a directed acyclic graph (DAG) encodes assumed causal relationships among variables, showing which variables to adjust for and, with causal discovery algorithms, suggesting causal structure from data.

방향성 비순환 그래프(DAG)로 나타낸 구조적 인과모형은 변수 간 가정된 인과관계를 표현하여 보정할 변수를 알려주며, 인과탐색 알고리즘을 쓰면 자료로부터 인과구조를 제안한다.

How do I map causal relationships among variables and decide which ones to control for?

변수 간 인과관계를 그리고 어떤 변수를 통제해야 하는지 어떻게 정하나요?

A structural causal model drawn as a directed acyclic graph lays out assumed causal paths, so the backdoor criterion identifies which confounders to adjust for and which colliders or mediators to leave out. Structure-learning algorithms such as PC can suggest a graph from observational data, but the result should be checked against subject-matter knowledge.

When to use Structural Causal Model (DAG)

• Several variables whose causal ordering must be stated before estimation
• Choosing an adjustment set for confounding
• Exploring possible causal structure among observed variables

When to use something else

• The causal structure is known and only the effect size is needed; use regression, matching or weighting with the chosen adjustment set
• Unmeasured confounding is likely and an instrument or natural experiment exists; use instrumental variables or regression discontinuity

Assumptions and requirements

• Acyclic causal structure
• Causal Markov condition and faithfulness for structure learning
• No unmeasured common causes unless explicitly modeled
• Conditional independence tests suited to the data type

What the analysis reports

• Directed acyclic graph
• Minimal sufficient adjustment sets
• Implied conditional independencies and tests
• Estimated edges or edge confidence from structure learning

How to report it

Based on the directed acyclic graph, the minimal sufficient adjustment set for the effect of [exposure] on [outcome] comprised [variables].

Related methods

• Propensity Score Matching
• Instrumental Variables Regression
• Bayesian Network

Run Structural Causal Model (DAG) on Quorant