Identifying the Causal Effects of Cross-World Policies
- Noah Weber ,
- Levi Boyles ,
- Shuayb Zarar
Neural Information Processing Systems (NeurIPS) |
Wkshp. on Causal Discovery and Causality-Inspired Machine Learning
In this work, we introduce the notion of a counterfactual response to what we call a Cross World Policy. Cross World policies are defined as a type of dynamic treatment regime which assigns treatments based on a fixed function of the naturally observed value of causally prior covariates, including the treatment itself. Cross World policies share commonalities with treatment effects on the treated (albeit in a dynamic treatment regime setting) and generalize the idea of shift interventions on the treated (SITs) developed in Sani et al. (2020). We give examples of potential
queries of interest which may be described as Cross World policies and complete identification criteria for estimation from observed data.