Treatment effect

As explained in Causal Inference for Data Science

The change in an outcome caused by changing an exposure or treatment while holding the relevant causal context fixed.

A treatment effect answers a change question: how would an outcome differ if the treatment changed while the relevant causal context stayed fixed? It is not the same as an observed association. People who receive a treatment may differ from those who do not for reasons that also affect the outcome, so correlation alone can confuse coincidence with cause. section 1.0 section 1.4.2

The treatment, also called a decision variable, is the variable whose effect is sought; the outcome is the result of interest. The average treatment effect (ATE) summarizes the causal effect of one variable on another. A causal model makes the assumed relationships explicit, often with a directed acyclic graph (DAG), so potential confounders and mediators can be considered rather than ignored. section 1.4.2 section 1.8 section 1.1.3

Medication and recovery

Suppose a doctor wants to know whether a new medication improves recovery. Age affects both which medication a patient receives and how quickly the patient recovers. Comparing raw recovery rates can therefore mix the medication’s effect with age-related differences. Age is a confounder: it creates a misleading path between treatment and outcome unless the analysis accounts for it. section 1.4.2

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 diagram to separate the treatment-to-recovery relationship from age, which influences both treatment choice and recovery. section 1.4.2

When treatment is binary, one causal contrast asks for the difference between the expected outcome after setting treatment to 1 and the expected outcome after setting it to 0: E[Y|do(D = 1)] − E[Y|do(D = 0)]. For a continuous treatment, a linear-model coefficient can represent the outcome change from a one-unit treatment increase while other included variables remain unchanged. That interpretation requires the causal assumptions behind the model, not merely a fitted regression. section 9.2.1 section 6.0

Estimating a treatment effect depends on what the data and causal model support:

• Adjust for observed confounders with the adjustment formula; in linear models, this corresponds to controlling for confounders in regression. section 1.1.5 section 9.2.1
• Use an instrumental variable when treatment–outcome confounders are unobserved but an instrument supplies independent treatment variation under additional assumptions. section 9.2 section 9.2.1
• Use time-related approaches such as differences in differences, regression discontinuity, or synthetic controls when an experiment is impractical and their assumptions fit the setting. section 1.1.5 section 11.0

Release effect as a counterfactual gap

To estimate whether releasing a new movie increased interest in an older one, train a model before release using an unaffected, correlated comparison series. Its post-release predictions represent the estimated without-release outcome. The gap between actual interest and those predictions estimates the release’s treatment effect, provided the comparison series remains unaffected and the method’s assumptions are credible. section 11.3 section 11.0

Randomized experiments, including A/B tests, are preferred when feasible because random assignment eliminates confounding. When they are not feasible, state the causal goal and assumptions, check whether important confounders are missing, and consider new data or surrogate variables. A useful estimate is one whose model complexity and assumptions are appropriate for the decision at stake. section 1.8 section 1.1.4 section 1.1.5

How Causal Inference for Data Science covers it

How deep it goes

Defines treatment effects through DAGs, intervention notation, regression coefficients, and multiple estimation strategies; defers detailed adjustment computation.

Examples

Uses medication recovery, airline pricing and demand, instrumental-variable synthetic data, and a movie-release time series.

What it assumes

Introduces causal effects intuitively from correlation and confounding, then uses linear regression and intervention notation for formal estimation.

What you can do after

Evaluate/understand whether an observed treatment–outcome difference supports a causal effect and choose an appropriate estimation approach.

How it connects

Connects treatment effects to confounders, causal DAGs, the adjustment formula, instrumental variables, A/B tests, and synthetic controls.

liveBookmanning.com

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

Related concepts

• regression discontinuity designmeasured by · Regression discontinuity design is a causal inference method that estimates treatment effects by comparing observations near a cutoff where treatment assignment changes.
• average treatment effect estimationmeasured by · Average treatment effect estimation quantifies the population-average difference in outcomes that would result from two alternative treatments or interventions.
• causal adjustmentproduced by · Causal adjustment estimates a causal effect by conditioning on variables that block noncausal pathways and averaging over their target-population distribution.
• total effectincludes · A total effect is the overall causal effect of a treatment or variable on an outcome, including all direct and mediated pathways.
• synthetic controlmeasured by · Synthetic control is a causal inference method that estimates a treated unit’s counterfactual outcome using a weighted combination of comparison units.
• instrumental variablemeasured by · An instrumental variable is a variable that affects a treatment or exposure while influencing an outcome only through that treatment or exposure, under causal identifying assumptions.

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.