Scholarlink

<p>Bayesian causal forests (BCF) are a powerful method for estimating heterogeneous treatment effects (HTE) in the presence of confounding variables. However, the accuracy of BCF critically depends on the correct estimation of propensity scores (PS); misspecified PS can lead to biased estimates of the treatment effect. To address this limitation, we propose BCF with covariate balancing propensity scores (BCF-CBPS), which integrates covariate balancing directly into the PS estimation step. By replacing conventional PS with CBPS, our approach improves covariate balance, enabling BCF to more accurately estimate true treatment effect heterogeneity. Extensive Monte Carlo simulations and real-world applications demonstrate that BCF-CBPS achieves higher precision and robustness compared to standard BCF and other machine learning methods for HTE estimation. These results highlight the effectiveness of combining covariate balancing with the BCF model for reliable estimation of treatment effects in complex observational studies.</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

Bayesian causal forests with covariate balancing propensity score: a novel approach for heterogeneous treatment effects estimation

  • 期刊论文

  • 发表日期：

    2026年1月31日

查看全文
  • Rafiullah,
  • Hong Wang

摘要

Bayesian causal forests (BCF) are a powerful method for estimating heterogeneous treatment effects (HTE) in the presence of confounding variables. However, the accuracy of BCF critically depends on the correct estimation of propensity scores (PS); misspecified PS can lead to biased estimates of the treatment effect. To address this limitation, we propose BCF with covariate balancing propensity scores (BCF-CBPS), which integrates covariate balancing directly into the PS estimation step. By replacing conventional PS with CBPS, our approach improves covariate balance, enabling BCF to more accurately estimate true treatment effect heterogeneity. Extensive Monte Carlo simulations and real-world applications demonstrate that BCF-CBPS achieves higher precision and robustness compared to standard BCF and other machine learning methods for HTE estimation. These results highlight the effectiveness of combining covariate balancing with the BCF model for reliable estimation of treatment effects in complex observational studies.