Causal Identification with Directed Acyclic Graphs (do-calculus)
Also known as: do-calculus, backdoor adjustment, Pearl causal identification, DAG ile Nedensel Tanımlama (do-calculus)
DAG causal identification is a framework, developed by Judea Pearl (2009), that encodes causal assumptions as a directed acyclic graph and uses the do-calculus rules to determine whether and how a causal effect can be identified from observational data. It systematically handles confounders, instrumental variables, and backdoor paths.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use DAG identification when you want to estimate a causal effect from observational (non-experimental) data and you can articulate the causal structure from theory. It needs a reasonable sample (at least about 50 observations relative to the number of covariates), a DAG specified in advance, and either the backdoor or frontdoor criterion satisfied. Because unobserved confounding cannot be ruled out by data alone, pair it with sensitivity analysis. It is not appropriate when the causal structure is unknown or the sample is too small for the number of covariates.
Strengths & limitations
- Makes causal assumptions explicit and visual, so confounders, instruments, and backdoor paths can be reasoned about transparently.
- Provides formal do-calculus rules that determine whether an effect is identifiable and which variables to adjust for.
- Works with continuous, binary, and categorical variables across many study structures.
- Conclusions are only as good as the assumed graph; a wrong DAG yields wrong causal claims.
- Unobserved confounders invalidate the causal interpretation unless the backdoor (or frontdoor) criterion is met.
- Requires enough observations relative to the number of covariates; with n < 50 the effect cannot be reliably estimated even when it is identified.
Frequently asked
What is the backdoor criterion?
A set of variables Z satisfies the backdoor criterion for the effect of X on Y if it blocks every spurious path from X to Y that enters X through an incoming arrow, without including any descendant of X. Adjusting for such a Z identifies the causal effect from observational data.
Does a DAG let me prove causality from data alone?
No. The graph encodes assumptions you bring from theory; the data cannot confirm the graph itself. Identification tells you that, if the assumptions hold, the effect is computable. Always accompany it with a sensitivity analysis for unobserved confounding.
What if a key confounder is unobserved?
Backdoor adjustment then fails. The frontdoor criterion can sometimes identify the effect through an observed mediator, or you can switch to a sensitivity analysis or a design-based method such as propensity score matching.
How is this different from just adding control variables to a regression?
A DAG tells you which variables to adjust for and which to avoid. Blindly controlling for everything can open backdoor paths (for example by conditioning on a collider) and bias the estimate, so the graph guides the adjustment set rather than the model mechanics.
Sources
- Pearl, J. (2009). Causality: Models, Reasoning, and Inference (2nd ed.). Cambridge University Press. ISBN: 978-0521895606
- Pearl, J., Glymour, M., & Jewell, N. P. (2016). Causal Inference in Statistics: A Primer. Wiley. ISBN: 978-1119186847
How to cite this page
ScholarGate. (2026, June 1). Causal Identification with Directed Acyclic Graphs (do-calculus). ScholarGate. https://scholargate.app/en/causal-inference/dag-identification
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Instrumental Variables in Health ResearchHealth Economics↔ compare
- Inverse Probability WeightingCausal inference↔ compare
- Mediation AnalysisStatistics↔ compare
- Propensity Score MatchingResearch Statistics↔ compare
- Sensitivity Analysis for Unmeasured ConfoundingCausal inference↔ compare