Abstract
This paper investigates causal effect identification in latent variableLinear Non-Gaussian Acyclic Models (lvLiNGAM) using higher-order cumulants,addressing two prominent setups that are challenging in the presence of latentconfounding: (1) a single proxy variable that may causally influence thetreatment and (2) underspecified instrumental variable cases where fewerinstruments exist than treatments. We prove that causal effects areidentifiable with a single proxy or instrument and provide correspondingestimation methods. Experimental results demonstrate the accuracy androbustness of our approaches compared to existing methods, advancing thetheoretical and practical understanding of causal inference in linear systemswith latent confounders.