liveBook · Manning

Covariate adjustment

As explained in Causal Inference for Data Science

Covariate adjustment is the process of accounting for differences in measured covariates when estimating a treatment effect.

Covariate adjustment estimates a treatment effect by accounting for measured differences between treated and untreated cases. With propensity scores, conditioning on the score can replace conditioning on the full set of confounders, reducing the data to outcome, treatment, and score. section 5.3.4

A practical propensity-score adjustment workflow is:

  • Identify confounders before modeling, ideally using substantive expert knowledge. section 5.4
  • Estimate each case’s propensity score with a machine-learning model. section 5.4
  • Use the scores in the adjustment formula or as the covariate for a T-learner, applying cross-fitting to reduce overfitting. section 5.3.4

Right Heart Catheterization

Estimate the average treatment effect of an invasive right heart catheterization on survival using the RHC dataset, with treatment recorded by swang1 and outcome by death. section 5.4

Table 5.3 Necessary data with propensity scores calculated to apply the adjustment formula
O
T
S
o 1
t 1
s 1
…
…
…
O n
t n
s n

A compact table shows the outcome, treatment, and propensity-score columns required to apply the adjustment formula. section 5.3.4

How Causal Inference for Data Science covers it

How deep it goes
Derives propensity-score adjustment mathematically and reduces implementation to outcome, treatment, and score, while deferring advanced matching variations.
Examples
Uses the Right Heart Catheterization dataset, propensity-score data tables, and an ATE exercise.
What it assumes
Assumes identified confounders and introduces their conversion into propensity scores for adjustment.
What you can do after
Apply propensity scores to compute an average treatment effect while using cross-fitting to limit machine-learning overfitting.
How it connects
Connects propensity scores with confounders, the adjustment formula, T-learners, cross-fitting, and matching.

Also explained in

Compare how 2 books explain covariate adjustment →

Related concepts

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.