Intervention distribution

As explained in Causal Inference for Data Science

A probability distribution over variables generated under an intervention that fixes a treatment value rather than under natural treatment assignment.

An intervention distribution describes outcomes after fixing a treatment value, rather than observing natural treatment assignment. P|do(T = A) asks what recovery would look like if everyone received treatment A. section 2.3

This differs from a conditional probability: conditional probabilities describe what was observed, while intervention distributions estimate what would happen in a changed scenario. The construction mimics a hypothetical randomized controlled trial, where treatment is assigned independently of confounders. section 2.3 section 4.0

Kidney-stone treatment

Kidney-stone size can affect both treatment choice and recovery, creating confounding in observed data. To represent an RCT, remove the arrow from size to treatment; estimating P|do(T = A) then represents assigning treatment A to everyone and measuring the resulting recovery rate. section 2.3

Figure 2.7 Graph depicting an RCT. In an RCT, various factors may influence the outcome, but none influence the treatment, as the treatment is assigned randomly. This means there are no confounders.

Use the RCT graph to see how random treatment assignment removes confounding arrows while other factors may still influence the outcome. section 2.3

How Causal Inference for Data Science covers it

How deep it goes

Defines P|do(T = A) and connects it to RCT graphs, while deferring broader estimation methods.

Examples

Uses kidney-stone treatment, recovery rates, confounding, and a hypothetical randomized controlled trial.

What it assumes

Assumes familiarity with conditional probabilities, treatment, outcomes, confounding, and causal graphs.

What you can do after

Evaluate/understand outcomes under an imposed treatment and distinguish them from predictions based on observed conditional probabilities.

How it connects

Connects causal inference, randomized controlled trials, conditional probability, supervised learning, and the adjustment formula.

liveBookmanning.com

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

Also explained in

• 

  liveBookmanning.com
  Causal AIAn intervention distribution is the probability distribution over variables induced by externally setting or changing one or more variables in a causal model.

Compare how 2 books explain intervention distribution →

Related concepts

• conditional probabilitycontrasts with · Conditional probability is the probability of an event or variable value given that specified information, an event, or another variable’s value is known.
• experimentationequivalent to · Experimentation is the deliberate testing of an idea, intervention, or alternative by observing outcomes to gather evidence or compare effects.
• term frequency-inverse document frequencyused by · Term frequency-inverse document frequency is a text-weighting scheme that combines a term’s frequency in a document with its inverse frequency across a document collection.
• observational joint distributioncontrasts with · An observational joint distribution is the true joint probability distribution of variables as observed without externally imposed interventions.

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.