Adjustment set

As explained in Causal AI

An adjustment set is a set of variables that satisfies the backdoor criterion for estimating a causal effect between specified variables.

Suppose you want to know whether increasing engagement in an online game causes more in-game purchases. Players who spend more time playing may both engage more and purchase more, so comparing engaged and less-engaged players does not necessarily answer the intervention question. An adjustment set identifies which observed variables to account for when estimating that causal effect. section 10.3.1 section 11.3

For a treatment X and outcome Y, an adjustment set Z satisfies the backdoor criterion: together, its variables block every backdoor path from X to Y, and none is a descendant of X. In a causal DAG, a backdoor path can carry an association through a common cause rather than through the effect of X on Y. Blocking those paths lets observed data identify the intervention query under the stated causal assumptions. section 10.3.1

Figure 10.8 E←G→I is a backdoor path where G is a “confounder” that is a common cause of E and I. We are interested in the statistical signal flowing along the causal path from E to I, but that signal is “confounded” by the noncausal noise from additional statistical information through G on the backdoor path E←G→I.

The path E←G→I shows how a common cause G can confound the apparent relationship between E and I. Adjusting for G blocks that backdoor path. section 10.3.1

Operationally, backdoor adjustment averages the outcome distribution within each value of Z, weighted by how common that value is: P(Y | do(X=x)) = Σz P(Y | X=x, Z=z)P(Z=z). For continuous adjustment variables, the corresponding operation is integration. The formula uses observational quantities to answer a causal query; it does not make an invalid adjustment set valid. section 10.3.1

Selecting variables for a purchase-effect estimate

For the effect of Side-quest Engagement on In-game Purchases, Guild Membership, Time Spent Playing, and Player Skill Level form a valid adjustment set. Prior Experience is unobserved, but the selected variables still block the backdoor path through it. In DoWhy, supply the data, treatment, outcome, and DAG to CausalModel, then call identify_effect() to inspect the backdoor estimand and its assumptions. section 11.4.1 section 11.3 section 11.3.1

After identification, keep variable selection distinct from estimation:

• A linear regression can include the adjustment variables as covariates; its treatment coefficient estimates the average treatment effect when that estimator is appropriate. section 11.4.1
• Double machine learning and meta learners offer other ways to estimate the backdoor effect using the adjustment variables. section 11.4.3
• Do not control for every available covariate: adjusting for a mediator can block the causal path of interest, while adjusting for a collider can introduce bias. section 11.4.1
• Check sensitivity to omitted common causes. If an important confounder leaves a backdoor path open, the estimated effect can change substantially. section 11.5.5

How Causal AI covers it

How deep it goes

Defines the graphical criterion, derives its adjustment formula, and carries the selected set into estimation and robustness checks.

Examples

Uses online gaming DAGs and a DoWhy purchase-effect workflow with guild membership, playing time, and player skill.

What it assumes

Builds on causal DAGs and observational data while explaining how backdoor paths, confounders, and d-separation determine valid controls.

What you can do after

Apply it via DoWhy to identify a backdoor estimand, choose controls, and test sensitivity to omitted confounders.

How it connects

Connects DAGs and d-separation to confounding, backdoor identification, regression, and machine-learning effect estimation.

liveBookmanning.com

Read Causal AI on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Causal Inference for Data ScienceAn adjustment set is a set of variables conditioned on to block noncausal paths while preserving the causal paths between a treatment and an outcome.

Compare how 2 books explain adjustment set →

Related concepts

• backdoor identificationused by · Backdoor identification is the identification of a causal effect by adjusting for variables that block noncausal backdoor paths between an exposure and an outcome.
• backdoor criterionuses · The backdoor criterion is a graphical condition for identifying a causal effect by requiring a variable set to block every backdoor path while excluding treatment descendants.
• average treatment effect estimationuses · Average treatment effect estimation quantifies the population-average difference in outcomes that would result from two alternative treatments or interventions.
• estimandaddresses · An estimand is the population quantity or probabilistic target that a statistical analysis seeks to determine.
• total effectuses · A total effect is the overall causal effect of a treatment or variable on an outcome, including all direct and mediated pathways.
• backdoor pathaddresses · A backdoor path is a path between a putative cause and an outcome that begins with an arrow into the cause and can create a noncausal association.

This explainer was generated by Manning's knowledge graph from the text of Causal AI. The section links point to the parts of the book it draws on.