Scholarlink

<p>The Residualized LASSO (RLASSO) approach is widely employed for estimating heterogeneous treatment effects (HTE) due to its proficiency in high-dimensional settings and variable selection capabilities. However, its reliance on linear models for both outcome and treatment assignment imposes limitations. Specifically, the linearity assumption can restrict the flexibility required to capture complex, nonlinear relationships between covariates and treatment or outcomes. In observational studies, where such nonlinearities are prevalent, this may lead to biased treatment effect estimates. To overcome the linear constraints of traditional RLASSO, we propose the RLASSO-SVM model, which integrates Support Vector Machines (SVM) into the R-learner framework to enable flexible, nonlinear modeling of both treatments and outcomes. This hybrid approach leverages SVM for residual estimation while employing LASSO to ensure interpretable treatment effect estimation, thereby enhancing performance in high-dimensional, nonlinear contexts. Results from Monte Carlo simulations and real-world applications demonstrate that the proposed RLASSO-SVM approach achieves superior precision compared to other machine learning methods used for estimating HTE .</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Heterogeneous treatment effects estimation with residualized LASSO and support vector machines

  • 期刊论文

  • 发表日期：

    2025年11月25日

查看全文
  • Rafiullah,
  • Hong Wang

摘要

The Residualized LASSO (RLASSO) approach is widely employed for estimating heterogeneous treatment effects (HTE) due to its proficiency in high-dimensional settings and variable selection capabilities. However, its reliance on linear models for both outcome and treatment assignment imposes limitations. Specifically, the linearity assumption can restrict the flexibility required to capture complex, nonlinear relationships between covariates and treatment or outcomes. In observational studies, where such nonlinearities are prevalent, this may lead to biased treatment effect estimates. To overcome the linear constraints of traditional RLASSO, we propose the RLASSO-SVM model, which integrates Support Vector Machines (SVM) into the R-learner framework to enable flexible, nonlinear modeling of both treatments and outcomes. This hybrid approach leverages SVM for residual estimation while employing LASSO to ensure interpretable treatment effect estimation, thereby enhancing performance in high-dimensional, nonlinear contexts. Results from Monte Carlo simulations and real-world applications demonstrate that the proposed RLASSO-SVM approach achieves superior precision compared to other machine learning methods used for estimating HTE .