Causal Density Functions
Researchers have introduced a new statistical concept called causal density functions, which are defined as Radon-Nikodym derivatives. These functions compare interventional laws to observational laws, serving as local density ratios for causal effects. The proposed method allows for the estimation, calibration, and scoring of directed influence by providing a pointwise change-of-measure object, which can be directly tested using observational expectations reweighted by the density ratio. AI
IMPACT Introduces a novel statistical framework for analyzing causal relationships, potentially improving AI model interpretability and causal inference.