Causal adjustment

As explained in Causal Inference for Data Science

The adjustment formula is a causal inference formula that identifies an effect by averaging outcome distributions conditional on treatment across the target population distribution of a confounding variable.

Causal adjustment estimates what would happen if everyone received a chosen treatment, rather than simply comparing the outcomes of people who happened to receive different treatments. It removes confounding bias by combining outcome estimates within confounder groups according to the target population’s distribution. section 2.2.5 section 2.2

A confounder affects both treatment assignment and the outcome. For example, age can influence which medication a patient receives and how quickly they recover. A raw comparison then mixes the medication effect with age differences; adjustment instead compares treatments within relevant groups and reweights those results. section 1.4.2 section 2.0

Figure 1.6 Apart from how treatment may affect recovery, there’s a third element affecting both simultaneously. These elements are called confounders. For instance, in this example, age is a confounder.

Use the graph to distinguish the treatment-to-outcome relationship from the shared cause that must be accounted for. section 1.4.2

Apply causal adjustment as a structured causal analysis:

• Draw a DAG and state the intervention question, such as the outcome if everyone received treatment A versus B. section 2.2 section 3.1
• Decide whether adjustment is valid for that graph and select variables to adjust for; do not adjust automatically. section 2.2.3 section 2.6
• Check positivity: each relevant confounder combination must have received the treatment at some point. section 2.5.5 section 2.9
• Estimate conditional outcomes by treatment and confounders, then average them over the confounder distribution. section 2.2.5 section 4.2.2

Kidney-stone treatments

Treatment A can look worse overall when it is used more often for difficult large-stone cases, even though it has higher recovery rates for both small and large stones. Adjusting each treatment’s size-specific recovery rates to the same patient mix reverses the misleading aggregate comparison and selects A. section 2.1.3 section 2.2

Adjustment is not a universal correction. It can estimate an intervention effect for a confounder structure, but may be biased when applied to a mediator. Missing relevant confounders also remains a central risk. With many confounders, estimating the needed conditional probabilities becomes difficult and may use machine learning with cross-fitting. section 2.5.10 section 2.9 section 4.6

How Causal Inference for Data Science covers it

How deep it goes

Derives discrete and expected-outcome forms, then defers validity criteria, variable selection, and advanced estimators to later treatments.

Examples

Uses medication by age, kidney-stone treatments by stone size, mobile-app engagement, and recommender-system purchases.

What it assumes

Assumes domain knowledge to draw causal graphs while teaching confounders and intervention-based treatment questions from concrete examples.

What you can do after

Apply it to estimate and compare treatment outcomes under a specified intervention after assessing confounding and positivity.

How it connects

Connects confounders, DAGs, interventions, positivity, mediators, average treatment effect, machine learning, and potential outcomes.

liveBookmanning.com

Read Causal Inference for Data Science on liveBook Get the book at manning.com

Related concepts

• propensity scoreuses · A propensity score is the estimated conditional probability that a unit receives a particular treatment or action given observed covariates or context.
• language modeluses · A language model is a computational model that represents patterns in linguistic tokens to estimate sequence probabilities and predict or generate language from context.
• confounderuses · A confounder is a variable associated with both an explanatory or treatment variable and an outcome, whose influence can distort their apparent relationship or estimated causal effect.
• supervised learningused by · Supervised learning is a machine-learning approach that learns to map inputs to target outputs from examples paired with known labels or values.
• treatment effectproduces · A treatment effect is the difference in an outcome attributable to receiving an intervention or exposure rather than a comparison condition.
• s-learnerused by · An S-learner is a causal estimation algorithm that fits one outcome model containing the treatment and confounders and predicts outcomes under alternative treatment values.

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