Full talk · edited transcript

Pros and Cons of Double Machine Learning vs. Standard Linear Regression Approaches to Estimating Treatment Effects and Structural Parameters

Alexander Chentsov24 May 2025

Author: Alexander Chentsov (MIPT, HSE, Big Data Center)

The talk compares the standard approach to estimating causal effects with multiple regression (the London School of Economics, LSE, methodology), which is vulnerable to the problem of non-uniform convergence (uniform inference), with double machine learning (DML), which solves this problem at the price of greater complexity and higher variance of the estimates. The author works on DML theoretically and occasionally solves applied problems; the talk is rather a critique of the problems of multiple regression, with some mention of the problems of DML.

Continue reading — it's free

Google or a one-time email code. No password.