Structural Equation Models describe acyclic causal relationships between random variables. Linear Non-Gaussian Acyclic Models (LiNGAMs) apply a further restriction that allows the definition of the DirectLiNGAM algorithm to estimate the causal order and causal effects. We study the usage of Antitonic Score Matching (ASM) for the various linear regressions that are used in this algorithm. We also propose a modification to the existing DirectLiNGAM algorithm that theoretically minimizes causal violations given already occurred violations. We also discuss using methods for LiNGAMs that are based in Independent Component Analysis (ICA). A modification for a likelihood-based method for overcomplete ICA is also discussed.     «

Structural Equation Models describe acyclic causal relationships between random variables. Linear Non-Gaussian Acyclic Models (LiNGAMs) apply a further restriction that allows the definition of the DirectLiNGAM algorithm to estimate the causal order and causal effects. We study the usage of Antitonic Score Matching (ASM) for the various linear regressions that are used in this algorithm. We also propose a modification to the existing DirectLiNGAM algorithm that theoretically minimizes causal vio...     »