双重机器学习处理效应估计的蒙特卡洛模拟

Monte Carlo Simulation of Treatment Effect Estimation of Double Machine Learning

基于机器学习进行因果推断是学术界研究的一个热点,双重机器学习是最新的因果效应估计方法之一,然而,相关理论并没有对不同机器学习方法之间的选择提供指导。鉴于此,文章运用蒙特卡洛模拟方法,研究不同情况下常见机器学习方法在双重机器学习处理效应估计中的表现,比较分析各种机器学习方法的估计结果,研究发现:基于不同机器学习方法的因果效应估计结果存在明显差异,双重机器学习方法的表现受到非线性函数形式以及样本容量的大小的影响;最后,在此基础上提出对机器学习方法选择的建议。

Causal inference based on machine learning is a hot topic in academic research, and double machine learning is one of the latest causal effect estimation methods. However, there is no relevant theory to guide the selection between different machine learning methods. Hence, this paper adopts Monte Carlo simulation to investigate the performance of commonly used machine learning methods in treatment effect estimation of double machine learning in different conditions, and then compares and analyzes the estimation results of different machine learning methods. The study finds that the results of causal effect estimation based on different machine learning methods are obviously different, and that the performance of double machine learning methods is affected by the form of nonlinear functions and sample size. Finally, the paper raises some suggestions on machine learning method selection.