Shimizu S. Statistical Causal Discovery: LiNGAM Approach

Tokyo: Springer, 2022. — 99 p.This is the first book to provide a comprehensive introduction to a new semiparametric causal discovery approach known as LiNGAM, with the fundamental background needed to understand it. It offers a general overview of the basics of the LiNGAM approach for causal discovery, estimation principles, and algorithms.
This semiparametric approach is one of the most exciting new topics in the field of causal discovery. The new framework assumes parametric assumptions on the functional forms of structural equations but makes no assumption on the distributions of exogenous variables other than non-Gaussianity. It provides data-analysis tools capable of estimating a much wider class of causal relations even in the presence of hidden common causes. This feature is in contrast to conventional nonparametric approaches based on conditional independence of variables.
This book is highly recommended to readers who seek an in-depth and up-to-date overview of this new causal discovery approach to advance the technique as well as to those who are interested in applying this approach to real-world problems. This LiNGAM approach should become a standard item in the toolbox of statisticians, machine learners, and practitioners who need to perform observational studies.AcronymsA Starting Point for Causal Inference
Framework of Causal Inference
Identification and Estimation of the Magnitude of Causation
Identification and Estimation of Causal Structures
Concluding RemarksBasics of LiNGAM Approach
Basic LiNGAM Model
Independent Component Analysis
LiNGAM Model
Identifiability of the LiNGAM model
Concluding RemarksEstimation of the Basic LiNGAM Model
ICA-Based LiNGAM Algorithm
DirectLiNGAM Algorithm
Multigroup Analysis
LiNGAM Model for Multiple Groups
DirectLiNGAM Algorithm for Multiple LiNGAMs
Concluding RemarksEvaluation of Statistical Reliability and Model Assumptions
Evaluation of Statistical Reliability
A Bootstrap Approach
Bootstrap Probability
Multiscale Bootstrap for LiNGAM
Evaluation of Model AssumptionsExtended Models
LiNGAM with Hidden Common Causes
Identification and Estimation of Causal Structures of Confounded Variables
LiNGAM Model with Hidden Common Causes
Identification Based on Independent Component Analysis
Estimation Based on Independent Component Analysis
Identification and Estimation of Causal Structures of Unconfounded Variables
Other Hidden Variable Models
LiNGAM Model for Latent Factors
LiNGAM Model in the Presence of Latent ClassesOther Extensions
Cyclic Models
Time-Series Models
Nonlinear Models
Discrete Variable Models