Русский
Notes on identification and estimation of causal effects using observational data
Received: 06 Jul 2025 Accepted: 19 Mar 2026
Published: 2026, vol. 30, issue 1, pp. 36–55
Abstract
The beginning of the 21st century in data science is characterized by the emergence of new interdisciplinary fields as well as an expansion of the general theoretical framework and development of statistical methods for solving novel problems. This article describes known approaches to addressing the problem of statistical identification of unidirectional (causal) relationships between variables using non-experimental data, and highlights distinctive features of statistical models employed for this purpose. It also considers a case study on generating data with nonlinear dependencies among variables through convolutional neural networks, where two different estimation techniques are investigated — Bayesian networks and double machine learning. The results show that both these approaches yield inaccurate estimates of individual effects in the considered scenario, and recommendations are provided regarding aggregated effect evaluation.
Keywords: causal identification, treatment effects, DAG-models, double machine learning, CATE
BibTeX
@article{IS-Chentsov-Toropov2026,
author = {Chentsov, Aleksandr Mikhailovich and Toropov, Nikita Igorevich},
title = {{Notes on identification and estimation of causal effects using observational data}},
journal = {Intelligent Systems. Theory and Applications},
year = {2026},
volume = {30},
number = {1},
pages = {36--55},
}
AMSBIB
\Bibitem{IS-Chentsov-Toropov2026}
\by A.\,M.~Chentsov, N.\,I.~Toropov
\paper Notes on identification and estimation of causal effects using observational data
\jour Intelligent Systems. Theory and Applications
\yr 2026
\vol 30
\issue 1
\pages 36--55
\lang In Russian