{
  "id": "back-door-criterion",
  "title": "Back-door criterion",
  "lang": "en",
  "source_lang": "en",
  "summary": "The back-door criterion is a graphical condition that identifies covariate sets sufficient to estimate a causal effect by adjustment for confounding.",
  "category": "mathematics",
  "subcategory": "probability-statistics",
  "type": "concept",
  "aliases": [
    "Backdoor criterion",
    "Pearl's back-door criterion"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/back-door-criterion/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/back-door-criterion/",
      "json": "https://aiwiki.page/zh/back-door-criterion.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "The back-door criterion is a graphical rule in causal inference for determining whether a set of observed variables can be used to adjust for confounding when estimating a causal effect. Formulated by Judea Pearl, it applies to a causal directed acyclic graph (DAG): an adjustment set must contain no descendants of the treatment and must block every path between treatment and outcome that begins with an arrow pointing into the treatment. When these conditions hold, the causal effect is identifiable from an appropriate observational distribution by covariate adjustment, subject to support and causal-model assumptions."
    },
    {
      "heading": "Causal graphs and back-door paths",
      "text": "A causal DAG represents variables as nodes and direct causal relationships as arrows. Its interpretation is stronger than that of an ordinary Bayesian network: arrows specify how variables are generated and how the system responds to interventions, not merely how an observational distribution factorizes.\n\nLet \\(X\\) denote a treatment or exposure and \\(Y\\) an outcome. A directed path such as\n\n\\[X\\to M\\to Y\\]\n\nrepresents a possible route through which changing \\(X\\) can affect \\(Y\\). By contrast, the path\n\n\\[X\\leftarrow C\\to Y\\]\n\ncan create an observational association because \\(C\\) is a common cause of \\(X\\) and \\(Y\\). It is a back-door path because its first edge has an arrowhead at \\(X\\).\n\nA path is a sequence of adjacent nodes; it need not follow the directions of its arrows. Consequently, back-door paths can be more complicated than a single common-cause fork. Their defining property is the direction of the first edge at the treatment, not their length or the orientation of subsequent edges.\n\nThe criterion separates paths that may transmit the causal effect of interest from paths that may create noncausal associations."
    },
    {
      "heading": "Formal definition",
      "text": "Let \\(G\\) be a causal DAG, and let \\(X\\) and \\(Y\\) be distinct nodes. A set \\(Z\\), disjoint from \\(\\{X,Y\\}\\), satisfies the back-door criterion relative to \\((X,Y)\\) if:\n\n- No member of \\(Z\\) is a descendant of \\(X\\). A descendant is a node reachable from \\(X\\) by a nonempty directed path.\n- \\(Z\\) blocks every path from \\(X\\) to \\(Y\\) that begins with an arrow into \\(X\\).\n\nBlocking is defined through d-separation. On a particular path, an internal node is a collider if both neighboring edges point into it. A path is open, or active, conditional on \\(Z\\) precisely when:\n\n- every noncollider on the path is outside \\(Z\\); and\n- every collider on the path is either in \\(Z\\) or has a descendant in \\(Z\\).\n\nA path is blocked if either requirement fails. Thus, conditioning on a noncollider blocks a path, whereas conditioning on a collider or one of its descendants can open a path that was previously blocked. D-separation provides graphical implications of conditional independence under the graph's Markov assumptions.\n\nFor a set \\(Z\\) containing no descendants of \\(X\\), the second condition can equivalently be checked by deleting every arrow leaving \\(X\\) and testing whether \\(Z\\) d-separates \\(X\\) and \\(Y\\) in the resulting graph. Deleting outgoing arrows removes paths that begin by following the treatment's causal influence.\n\nThe empty set may satisfy the criterion. In that case, no covariate adjustment is needed to block back-door paths."
    },
    {
      "heading": "The adjustment formula",
      "text": "If \\(Z\\) satisfies the back-door criterion, then the intervention distribution is given, for discrete variables, by\n\n\\[P(Y=y\\mid \\operatorname{do}(X=x)) = \\sum_z P(Y=y\\mid X=x,Z=z)\\,P(Z=z).\\]\n\nHere, the do-operator denotes an intervention that sets \\(X\\) to \\(x\\), replacing its usual causal assignment mechanism. It differs from ordinary conditioning on \\(X=x\\), which selects observations in which the treatment naturally takes that value.\n\nFor continuous adjustment variables, the corresponding expression uses integration:\n\n\\[p(y\\mid \\operatorname{do}(X=x)) = \\int p(y\\mid x,z)\\,p(z)\\,dz.\\]\n\nThese formulas first compare outcomes within covariate strata and then average over the covariate distribution in the target population. Importantly, the weighting distribution is \\(P(Z=z)\\), not \\(P(Z=z\\mid X=x)\\).\n\nFor a numerical outcome, the intervention mean is\n\n\\[\\mathbb{E}[Y\\mid \\operatorname{do}(X=x)] = \\mathbb{E}_{Z}\\!\\left[ \\mathbb{E}[Y\\mid X=x,Z] \\right],\\]\n\nwhere the inner quantity is a conditional expectation. With binary treatment, the average treatment effect is therefore\n\n\\[\\operatorname{ATE} = \\mathbb{E}_{Z}\\!\\left[ \\mathbb{E}[Y\\mid X=1,Z] - \\mathbb{E}[Y\\mid X=0,Z] \\right].\\]\n\nRelation to potential outcomes\n\nIn the potential-outcomes framework, \\(Y(x)\\) denotes the outcome under treatment level \\(x\\). A valid back-door set supplies a graphical sufficient condition for conditional exchangeability,\n\n\\[Y(x)\\perp X\\mid Z,\\]\n\nunder the corresponding causal model.\n\nTogether with consistency, \\(Y=Y(x)\\) when \\(X=x\\), this gives\n\n\\[\\begin{aligned} \\mathbb{E}[Y(x)] &=\\mathbb{E}_{Z}\\!\\left[\\mathbb{E}[Y(x)\\mid Z]\\right]\\\\ &=\\mathbb{E}_{Z}\\!\\left[\\mathbb{E}[Y(x)\\mid X=x,Z]\\right]\\\\ &=\\mathbb{E}_{Z}\\!\\left[\\mathbb{E}[Y\\mid X=x,Z]\\right]. \\end{aligned}\\]\n\nIdentification from observed data also requires positivity: treatment level \\(x\\) must be supported within the covariate strata being averaged. For discrete variables, this means\n\n\\[P(X=x\\mid Z=z)>0\\]\n\nwhenever \\(P(Z=z)>0\\). A graphical adjustment set does not itself guarantee positivity."
    },
    {
      "heading": "Examples",
      "text": "A common cause\n\nConsider the graph with arrows\n\n\\[C\\to X,\\qquad C\\to Y,\\qquad X\\to Y.\\]\n\nThe path \\(X\\leftarrow C\\to Y\\) is open without adjustment. Since \\(C\\) is not a descendant of \\(X\\), conditioning on \\(C\\) blocks this path and satisfies the criterion:\n\n\\[P(y\\mid \\operatorname{do}(x)) = \\sum_c P(y\\mid x,c)P(c).\\]\n\nThe unadjusted association \\(P(y\\mid x)\\) instead averages using \\(P(c\\mid x)\\), which may differ across treatment groups.\n\nMore than one valid adjustment set\n\nSuppose the graph consists of\n\n\\[A\\to X,\\qquad A\\to B,\\qquad B\\to Y,\\qquad X\\to Y.\\]\n\nIts back-door path is\n\n\\[X\\leftarrow A\\to B\\to Y.\\]\n\nEither \\(\\{A\\}\\) or \\(\\{B\\}\\) blocks the path, and neither variable is a descendant of \\(X\\). Both are valid adjustment sets; so is \\(\\{A,B\\}\\).\n\nA valid set is minimal if none of its proper subsets is valid. Minimal does not mean uniquely determined, nor does it necessarily mean best for statistical precision.\n\nConditioning on a collider\n\nConsider the arrows\n\n\\[A\\to X,\\qquad A\\to C,\\qquad B\\to C,\\qquad B\\to Y,\\qquad X\\to Y.\\]\n\nThe back-door path\n\n\\[X\\leftarrow A\\to C\\leftarrow B\\to Y\\]\n\nis initially blocked at the collider \\(C\\). The empty set therefore satisfies the criterion in this graph. Conditioning only on \\(C\\) opens the path and can introduce collider bias, rather than remove confounding.\n\nThis illustrates why adjusting for every available variable is not a generally valid strategy. Conditioning or selection can create associations that were absent before adjustment, a mechanism also relevant to selection bias.\n\nA mediator\n\nIn the graph\n\n\\[X\\to M\\to Y,\\]\n\n\\(M\\) lies on a causal pathway and is a descendant of \\(X\\). It is therefore excluded by the back-door criterion. Adjusting for \\(M\\) generally blocks part of the total causal effect.\n\nEstimating direct and indirect effects is a separate problem addressed by causal mediation analysis. Conditioning on a mediator does not, by itself, identify a direct causal effect."
    },
    {
      "heading": "Identification and estimation",
      "text": "The criterion addresses identification: whether a causal quantity can be expressed in terms of an observed probability distribution under stated assumptions. It does not prescribe a unique estimator.\n\nOnce an adjustment set is justified, the adjustment formula can be estimated through methods such as:\n\n- outcome regression followed by standardization over the covariate distribution;\n- stratification or matching;\n- inverse probability weighting;\n- methods combining outcome and treatment-assignment models.\n\nThese approaches can target the same identified effect but differ in modeling assumptions, finite-sample behavior, and sensitivity to poor treatment overlap.\n\nIncluding \\(Z\\) in a linear regression is not automatically equivalent to evaluating the adjustment formula correctly. The functional form, interactions, and definition of the target effect matter. In nonlinear models, a treatment coefficient need not equal a marginal causal effect."
    },
    {
      "heading": "Scope and limitations",
      "text": "Dependence on the causal graph\n\nThe criterion is valid relative to an assumed causal structure. Observational associations alone generally do not determine all arrow directions or establish that every relevant common cause has been measured.\n\nIf an unobserved variable \\(U\\) creates the path\n\n\\[X\\leftarrow U\\to Y,\\]\n\nand no observed variables can block all relevant back-door paths, an observed back-door adjustment set may not exist. The rule does not justify treating unmeasured confounding as absent.\n\nA sufficient, not necessary, condition\n\nThe classical back-door criterion is sufficient for covariate adjustment but is not a necessary characterization of every valid adjustment set. Its blanket exclusion of treatment descendants is conservative: more general adjustment criteria permit some descendants when they cannot disrupt identification.\n\nFailure to find a back-door set also does not imply that the causal effect is unidentifiable by every method. In suitable graphs, the front-door criterion or more general do-calculus arguments identify effects without a back-door adjustment set. Instrumental-variable methods offer another approach under their own assumptions, though the identifiable effect depends on the model.\n\nGraphical validity does not determine precision\n\nDifferent valid sets may yield estimators with different variances. Additional covariates can improve outcome prediction, but they can also increase model complexity or worsen treatment overlap. A valid set need not be the most statistically efficient set, and a larger set need not be safer.\n\nThe estimand must be specified\n\nThe classical criterion concerns adjustment for the total effect of a specified treatment on an outcome. Joint interventions, longitudinal treatment regimes, direct effects, and effects in selected populations require additional care. In particular, a variable may confound a later treatment–outcome relationship while itself being affected by an earlier treatment, making ordinary single-stage adjustment inappropriate."
    },
    {
      "heading": "Historical role",
      "text": "Pearl's formulation made graphical blocking rules an explicit basis for covariate selection in causal inference. It connected substantive causal assumptions, represented by arrows, to an observational adjustment formula without requiring a particular linear or parametric model. The distinction between common causes, mediators, and colliders explains why adjustment is a causal modeling decision rather than merely a question of predictive association."
    }
  ],
  "links": [
    "causal-inference",
    "confounding",
    "directed-acyclic-graph",
    "bayesian-network",
    "d-separation",
    "conditional-independence",
    "conditional-probability",
    "conditional-expectation",
    "selection-bias",
    "statistical-identification",
    "estimator",
    "linear-regression",
    "instrumental-variables",
    "variance"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}