Interventions and the do-operator
What it means, mathematically, to set a variable. Graph surgery, P(y | do(x)) versus P(y | x), the backdoor adjustment formula derived from scratch, the front-door trick for hidden confounders, the do-calculus, and a first counterfactual.
50 min
After this lesson you can
- Define an intervention as a modification of a structural causal model, and draw the surgered graph
- Explain in words and symbols why P(y | do(x)) and P(y | x) differ
- Derive the backdoor adjustment formula and apply it to data
- State the front-door criterion and compute a front-door estimate
- Say what the do-calculus is for, and what identifiability means
- Compute a counterfactual in a simple structural model with abduction, action and prediction
In this lesson
- Setting a variable is editing the model
- Seeing it on data
- The backdoor adjustment formula
- When the confounder is hidden: the front door
- The do-calculus and identifiability
- The third rung: counterfactuals
This lesson is part of Descent Pro
Descent Pro unlocks every lesson on the route, from base camp to the summit. Every plan starts with a free trial.