Node-splitting operation

As explained in Causal AI

A node-splitting operation replaces an intervention node in a causal graph with separate components representing its intervened and natural values.

A node-splitting operation transforms a causal DAG for an intervention by separating the target’s natural value from its imposed value. The original variable keeps its parents, while a new constant-valued node takes over its children. Downstream variables inherit intervention subscripts, making counterfactual quantities explicit within a single-world intervention graph (SWIG). section 10.7.3

Apply the transformation in three steps:

• Split each intervention target into a random variable retaining incoming edges and a constant intervention node retaining outgoing edges. section 10.7.3 section 11.6.1
• Append the intervention value as a subscript to every downstream variable. section 10.7.3
• Repeat the split for additional interventions, accumulating all upstream intervention subscripts. section 10.7.3

The resulting SWIG contains counterfactual variables and supports d-separation, enabling identification of suitable causal queries. Node-splitting can also represent level 3 queries whose factual conditions and hypothetical outcomes do not overlap, but this requires the single-world assumption: the natural value must remain knowable and unaffected by the intervention. section 10.7.3 section 10.7.5

Recommendation-policy intervention

For a recommendation policy T intervened from +t to −t, retain T with its profile parent C, create a constant intervention node −t for downstream variables, and represent outcomes such as W and A as W_T=−t and A_T=−t. Applying a second intervention to B adds B=−b to affected variables’ subscripts. section 10.7.3

Figure 11.13 The node-splitting transform splits the intervention target into a constant that keeps the children and a random variable that keeps the parents.

Shows the split between the constant intervention node that keeps children and the random variable that keeps parents. section 11.6.1

How Causal AI covers it

How deep it goes

Specifies edge reassignment, subscript inheritance, multiple interventions, and the single-world assumption, but not full identification algorithms.

Examples

Uses a Netflix recommendation policy intervention and downstream variables carrying single or multiple intervention subscripts.

What it assumes

Assumes causal DAGs, interventions, counterfactuals, and d-separation, while making the single-world assumption explicit.

What you can do after

Apply it via Pyro to implement SWIG-style interventions and infer identified causal quantities.

How it connects

Connects SWIGs with causal DAGs, graph surgery, d-separation, counterfactual graphs, and probabilistic inference.

liveBookmanning.com

Read Causal AI on liveBook Get the book at manning.com

Related concepts

• graph surgerykind of · Graph surgery is the modification of a causal graph to represent an intervention by changing or removing the intervened variable's incoming causal edges.
• single-world intervention graphused by · A single-world intervention graph is a causal graph formed by splitting intervention nodes to represent counterfactual variables under a single-world assumption.
• interventionimplements · An intervention is a deliberate, externally imposed change or control applied to a variable or system to alter or assess its behavior or outcomes.

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