ID* identifies counterfactual queries from observational and experimental data by operating on a causal graph. It recursively removes irrelevant variables, decomposes graphs into c-components, applies do-calculus, and factorizes queries into simpler subqueries. When the graph cannot support further simplification, it returns a non-identifiable result. section 10.7.2
ID* fits into a family of graphical identification algorithms:
- ID handles interventional queries using observational data; ID* handles counterfactual queries using observational and experimental data. section 10.7.2
- IDC* extends ID* to queries conditioned on evidence, such as the effect of treatment on the treated. section 10.7.2
- Counterfactual graphs represent variables across intervention worlds, providing the structure over which identification proceeds. section 10.7.2
Identifying ETT with y0
Use y0's idc_star function with a DAG, an outcome event such as A under an alternative treatment, and observed treatment evidence. The algorithm returns an estimand when the query is identifiable.
